From b7304af1483270712dd41cec6835815d7acadd85 Mon Sep 17 00:00:00 2001
From: Cursor Agent <cursoragent@cursor.com>
Date: Tue, 8 Sep 2026 18:38:33 +0000
Subject: [PATCH 01/13] docs(challenge): experimental caveat, honesty review,
 lives
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Clarify that CodeDNA is still experimental per agent×language (issues/PRs welcome), require truthful metrics with optional Meet verification, and note public lives on submitted tests.

AI-Agent: composer-2.5
AI-Provider: cursor
AI-Session: s_20260908_001
AI-Visited: docs/challenge.md,docs/challenge.it.md,docs/challenge.html,challenge/SUBMISSION_TEMPLATE.md,challenge/SUBMISSION_TEMPLATE.it.md,README.md,README-it.md
AI-Message: Honesty + Meet review + lives added; experimental agent/language note mirrored to HTML

Co-authored-by: Fabrizio Corpora <Larens94@users.noreply.github.com>
---
 .codedna                            | 16 ++++++++--------
 README-it.md                        |  2 +-
 README.md                           |  2 +-
 challenge/SUBMISSION_TEMPLATE.it.md |  2 ++
 challenge/SUBMISSION_TEMPLATE.md    |  2 ++
 docs/challenge.html                 | 19 +++++++++++++++++--
 docs/challenge.it.md                | 26 ++++++++++++++++++++++++--
 docs/challenge.md                   | 28 +++++++++++++++++++++++++---
 8 files changed, 80 insertions(+), 17 deletions(-)

diff --git a/.codedna b/.codedna
index 7daf86b..ee8416a 100644
--- a/.codedna
+++ b/.codedna
@@ -30,14 +30,6 @@ packages:
 cross_cutting_patterns: {}
 
 agent_sessions:
-  - agent: "gpt-5"
-    provider: "openai"
-    date: "2026-08-21"
-    session_id: "s_20260821_gitbook_docs"
-    task: "Redesign documentation like GitBook"
-    changed: ["docs/install.html","docs/site.css",".codedna"]
-    visited: ["docs/install.html","docs/site.css",".codedna"]
-    message: "Reworked the documentation shell with a GitBook-inspired three-column layout, searchable hierarchical sidebar, keyboard shortcut, breadcrumb, page metadata, sticky outline, pagination, feedback, and responsive spacing. HTML/CSS checks, Ruff, and 284 tests pass."
   - agent: "gpt-5"
     provider: "openai"
     date: "2026-08-21"
@@ -70,3 +62,11 @@ agent_sessions:
     changed: ["codedna_tool/languages/php.py","codedna_tool/languages/base.py","codedna_tool/languages/vbnet.py","codedna_tool/languages/__init__.py","codedna_tool/cli.py","integrations/.roorules","integrations/install.sh","integrations/README.md","docs/languages.md","README.md","README-it.md","CHANGELOG.md","tests/test_issues_4_6_7.py","tests/test_docs.py"]
     visited: ["codedna_tool/languages/php.py","codedna_tool/languages/base.py","codedna_tool/languages/csharp.py","codedna_tool/languages/__init__.py","codedna_tool/cli.py","integrations/.clinerules"]
     message: "Validated #4/#6/#7 still open. Fixed PHP mixed-HTML injection + full-file has_codedna_header; added VbNetAdapter; added Roo --tools roo. 306 tests pass."
+  - agent: "composer-2.5"
+    provider: "cursor"
+    date: "2026-09-08"
+    session_id: "s_20260908_001"
+    task: "Challenge honesty + experimental notes"
+    changed: ["docs/challenge.md","docs/challenge.it.md","docs/challenge.html","challenge/SUBMISSION_TEMPLATE.md","challenge/SUBMISSION_TEMPLATE.it.md","README.md","README-it.md"]
+    visited: ["docs/challenge.md","docs/challenge.it.md","docs/challenge.html","challenge/SUBMISSION_TEMPLATE.md","challenge/SUBMISSION_TEMPLATE.it.md","README.md","README-it.md"]
+    message: "Documented experimental agent×language caveat + honesty/Meet review + public lives in challenge rules."
diff --git a/README-it.md b/README-it.md
index f7bc644..5c36030 100644
--- a/README-it.md
+++ b/README-it.md
@@ -48,7 +48,7 @@ Nessuna infrastruttura. Nessuna pipeline di retrieval. Nessuna memoria esterna.
 
 ## CodeDNA Challenge — €200
 
-Valutazione pubblica sul **tuo** progetto (non SWE-bench). Almeno 10 task con/senza CodeDNA, stack L0/L1/L2 a pari livello, PR con metriche — anche se vanno contro CodeDNA.
+Valutazione pubblica sul **tuo** progetto (non SWE-bench). Almeno 10 task con/senza CodeDNA, stack L0/L1/L2 a pari livello, PR con metriche — anche se vanno contro CodeDNA. Affermazioni oneste: il team può chiedere una review Meet; live sui test quando possibile.
 
 - Bacheca: [larens94.github.io/codedna/challenge.html](https://larens94.github.io/codedna/challenge.html)
 - Regolamento: [`docs/challenge.it.md`](docs/challenge.it.md) (IT) · [`docs/challenge.md`](docs/challenge.md) (EN)
diff --git a/README.md b/README.md
index 1642ae5..7312378 100644
--- a/README.md
+++ b/README.md
@@ -48,7 +48,7 @@ No infrastructure. No retrieval pipeline. No external memory. The code carries i
 
 ## CodeDNA Challenge — €200
 
-Public **own-project** evaluation (not SWE-bench). ≥10 tasks with/without CodeDNA, fair L0/L1/L2 stack parity, metrics PR — even if results go against CodeDNA.
+Public **own-project** evaluation (not SWE-bench). ≥10 tasks with/without CodeDNA, fair L0/L1/L2 stack parity, metrics PR — even if results go against CodeDNA. Claims must be honest: the team may request a Meet review; live sessions on submitted tests when possible.
 
 - Board: [larens94.github.io/codedna/challenge.html](https://larens94.github.io/codedna/challenge.html)
 - Rules: [`docs/challenge.md`](docs/challenge.md) (EN) · [`docs/challenge.it.md`](docs/challenge.it.md) (IT)
diff --git a/challenge/SUBMISSION_TEMPLATE.it.md b/challenge/SUBMISSION_TEMPLATE.it.md
index 2e8c30b..29b7ad0 100644
--- a/challenge/SUBMISSION_TEMPLATE.it.md
+++ b/challenge/SUBMISSION_TEMPLATE.it.md
@@ -53,3 +53,5 @@
 - [ ] Nessun secret / sorgente proprietaria in questa PR
 - [ ] Solo log redatti
 - [ ] Modalità + parity di stack dichiarate nel JSON (`mode`, `stack`)
+- [ ] Metriche e narrativa sono veritiere (non inventate); posso ripercorrerle in una call di review (Meet / simile) se richiesto
+- [ ] Capisco che claim inventati = squalifica
diff --git a/challenge/SUBMISSION_TEMPLATE.md b/challenge/SUBMISSION_TEMPLATE.md
index 71b2333..4474782 100644
--- a/challenge/SUBMISSION_TEMPLATE.md
+++ b/challenge/SUBMISSION_TEMPLATE.md
@@ -53,3 +53,5 @@
 - [ ] No secrets / proprietary source in this PR
 - [ ] Redacted logs only
 - [ ] Mode + stack parity declared in JSON (`mode`, `stack`)
+- [ ] Metrics and narrative are truthful (not invented); I can walk through them on a review call (Meet / similar) if asked
+- [ ] I understand fabricated claims = disqualification
diff --git a/docs/challenge.html b/docs/challenge.html
index c9e57a9..60eba1e 100644
--- a/docs/challenge.html
+++ b/docs/challenge.html
@@ -212,6 +212,13 @@ <h1 data-i18n-html="h1">CodeDNA Challenge<br><em>€200</em></h1>
       <a class="btn-secondary" href="challenge.md" data-i18n="ctaRules">Full rules (EN)</a>
       <a class="btn-secondary" href="challenge.it.md" data-i18n="ctaRulesIt">Regolamento (IT)</a>
     </div>
+    <p class="note" style="margin-top:1.75rem;max-width:640px" data-i18n-html="experimentalNote">
+      <strong>Experimental.</strong> Results can vary by AI coding agent and language.
+      If annotations or the CLI misbehave on your stack, open a
+      <a href="https://github.com/Larens94/codedna/issues/new">GitHub issue</a> or a fix
+      <a href="https://github.com/Larens94/codedna/compare">pull request</a> so we can harden support.
+      That feedback is part of the challenge.
+    </p>
   </div>
 </header>
 
@@ -276,6 +283,10 @@ <h3 data-i18n="s3t">Open the PR</h3>
       <p data-i18n="l2">Graph / structural memory — parity required.</p>
     </div>
   </div>
+  <p class="note" style="margin-top:1.5rem" data-i18n-html="honestyNote">
+    <strong>Honesty.</strong> Do not invent results. Fabrizio or the review team may request a live review (e.g. Google Meet) to verify your runs.
+    When possible we will also host public lives discussing submitted tests.
+  </p>
 </section>
 
 <section class="section" id="board">
@@ -333,6 +344,7 @@ <h2 data-i18n="boardTitle">Public board</h2>
     h1:'CodeDNA Challenge<br><em>€200</em>',
     lead:'Test CodeDNA on your own project. Open a metrics PR to enroll — no separate signup. We update this board when submissions are merged.',
     ctaExample:'Copy metrics.json →', ctaRules:'Full rules (EN)', ctaRulesIt:'Regolamento (IT)',
+    experimentalNote:'<strong>Experimental.</strong> Results can vary by AI coding agent and language. If annotations or the CLI misbehave on your stack, open a <a href="https://github.com/Larens94/codedna/issues/new">GitHub issue</a> or a fix <a href="https://github.com/Larens94/codedna/compare">pull request</a> so we can harden support. That feedback is part of the challenge.',
     statusTitle:'Challenge status',
     statusLead:'Maintainer-updated board. Source file: docs/challenge-board.json.',
     metaPrize:'Prize', metaPrizeSub:'paid after review',
@@ -346,11 +358,12 @@ <h2 data-i18n="boardTitle">Public board</h2>
     l0:'In-source headers — the layer under test.',
     l1:'LLM wiki, skill packs, instruction files — parity required.',
     l2:'Graph / structural memory — parity required.',
+    honestyNote:'<strong>Honesty.</strong> Do not invent results. Fabrizio or the review team may request a live review (e.g. Google Meet) to verify your runs. When possible we will also host public lives discussing submitted tests.',
     boardTitle:'Public board',
     boardLead:'Language + framework, project size and file count are required so comparisons stay honest.',
     emptyTitle:'No submissions on the board yet',
     emptyBody:'When a metrics PR is reviewed and merged, we add a row here by editing challenge-board.json.',
-    boardNote:'This board is static and curated. Opening a PR enrolls you automatically; appearing here means maintainers accepted the submission format.',
+    boardNote:'This board is static and curated. Opening a PR enrolls you automatically; appearing here means maintainers accepted the submission format. Agent × language gaps are expected — file issues or fix PRs.',
     thHandle:'Entrant', thStack:'Language / framework', thSize:'Size', thFiles:'Files',
     thMode:'Mode', thScore:'Pass C / DNA', thFavors:'Favors', thPr:'PR',
     unlock:'min {n} to unlock', windowFallback:'dates TBD'
@@ -361,6 +374,7 @@ <h2 data-i18n="boardTitle">Public board</h2>
     h1:'CodeDNA Challenge<br><em>€200</em>',
     lead:'Prova CodeDNA sul tuo progetto. Apri una PR con le metriche per iscriverti — niente signup separato. Aggiorniamo questa bacheca quando le submission vengono mergiate.',
     ctaExample:'Copia metrics.json →', ctaRules:'Rules (EN)', ctaRulesIt:'Regolamento (IT)',
+    experimentalNote:'<strong>Sperimentale.</strong> I risultati possono variare per agente AI di coding e linguaggio. Se annotazioni o CLI si comportano male sul tuo stack, apri una <a href="https://github.com/Larens94/codedna/issues/new">issue</a> o una <a href="https://github.com/Larens94/codedna/compare">pull request</a> di fix così possiamo rafforzare il supporto. Quel feedback fa parte della challenge.',
     statusTitle:'Stato challenge',
     statusLead:'Bacheca aggiornata dai maintainer. File sorgente: docs/challenge-board.json.',
     metaPrize:'Premio', metaPrizeSub:'pagato dopo review',
@@ -374,11 +388,12 @@ <h2 data-i18n="boardTitle">Public board</h2>
     l0:'Header in-source — layer sotto test.',
     l1:'Wiki LLM, skill pack, instruction file — parity obbligatoria.',
     l2:'Memoria graph / strutturale — parity obbligatoria.',
+    honestyNote:'<strong>Onestà.</strong> Non inventare i risultati. Fabrizio o il team di review possono chiedere una review live (es. Google Meet) per verificare le run. Quando possibile faremo anche live pubbliche sui test inviati.',
     boardTitle:'Bacheca pubblica',
     boardLead:'Linguaggio + framework, dimensione progetto e numero file sono obbligatori per confronti onesti.',
     emptyTitle:'Ancora nessuna submission in bacheca',
     emptyBody:'Quando una PR metrics viene accettata e mergiata, aggiungiamo una riga modificando challenge-board.json.',
-    boardNote:'Bacheca statica e curata. Aprire una PR ti iscrive automaticamente; comparire qui significa che i maintainer hanno accettato il formato.',
+    boardNote:'Bacheca statica e curata. Aprire una PR ti iscrive automaticamente; comparire qui significa che i maintainer hanno accettato il formato. Gap agente × linguaggio sono attesi — apri issue o PR di fix.',
     thHandle:'Partecipante', thStack:'Linguaggio / framework', thSize:'Size', thFiles:'File',
     thMode:'Mode', thScore:'Pass C / DNA', thFavors:'Favors', thPr:'PR',
     unlock:'min {n} per sbloccare', windowFallback:'date TBD'
diff --git a/docs/challenge.it.md b/docs/challenge.it.md
index 49789a7..4dd57e4 100644
--- a/docs/challenge.it.md
+++ b/docs/challenge.it.md
@@ -12,7 +12,7 @@ Questa challenge pone una sola domanda:
 
 > CodeDNA aiuta (o no) quando fai sviluppo reale assistito da AI — e possiamo misurarlo in modo onesto?
 
-I bug trovati durante la challenge sono benvenuti: apri una GitHub issue e li correggiamo upstream.
+**Sperimentale.** La qualità di CodeDNA dipende ancora dall’**agente AI di coding** e dal **linguaggio / framework** che usi. Alcune combinazioni funzionano meglio di altre. Se install, annotation, hook o refresh non si comportano bene sul tuo stack, **apri una GitHub issue o una PR di fix** — fa parte della challenge, e sistemiamo upstream a partire dai tuoi report.
 
 ---
 
@@ -36,15 +36,25 @@ I bug trovati durante la challenge sono benvenuti: apri una GitHub issue e li co
 | Submission valide minime | **5** per sbloccare il premio (obiettivo stretch: 10) |
 | Se ci sono meno di 5 PR valide | I risultati si pubblicano comunque; il premio **non** viene assegnato (o passa all’edizione successiva) |
 | Vincitore | Scelto da Fabrizio Corpora + team di review (non voto community) |
+| Onestà | Le affermazioni devono essere vere e supportate da evidenze — metriche o narrative inventate = squalifica |
+| Verifica | Fabrizio o il team di review possono chiedere una review live (Google Meet / simile) per ripercorrere run, log e setup del repo |
+| Live | Quando possibile ospiteremo live pubbliche sui test inviati (con consenso del partecipante dove serve) |
 
 ### Cosa valutiamo (in ordine)
 
-1. **Onestà del protocollo** — stack equo, modalità dichiarata, note riproducibili  
+1. **Onestà del protocollo** — stack equo, modalità dichiarata, note riproducibili; **niente risultati inventati**  
 2. **Qualità dei task** — mix reale facili/medi/difficili su una codebase vera  
 3. **Evidenze** — metriche + breve narrativa; a favore **o** contro CodeDNA va bene  
 4. **Bug report** — issue actionable aperte upstream contano positivamente  
 5. **Chiarezza** — un altro engineer può ripetere il confronto  
 
+### Onestà, call di review e live
+
+- **Non puoi inventare** esiti pass/fail, tempi, dettagli di stack o claim narrativi. Se non riesci a riprodurre un numero, marca il task come inconclusive e dillo.
+- Fabrizio Corpora e/o il team di review **possono chiedere una video call** (es. Google Meet) per verificare la submission: screen-share del progetto, riesecuzione di un task campione, walkthrough di `metrics.json` / note.
+- Rifiutare senza motivo valido una richiesta ragionevole di verifica può annullare l’eleggibilità al premio per quella entry.
+- Quando possibile faremo anche **live pubbliche** sui test della challenge (metodologia, sorprese, gap agente × linguaggio). Partecipare a una live è opzionale, salvo finalisti chiamati in verifica.
+
 ---
 
 ## Timeline (da compilare prima della pubblicazione)
@@ -194,6 +204,16 @@ codedna init . --no-llm   # oppure con LLM per le rules:
 
 Solo domande (opzionale): GitHub Discussions / Discord — il vecchio template “entry issue” non è obbligatorio.
 
+### Se qualcosa si rompe sul tuo agente o linguaggio
+
+CodeDNA è ancora sperimentale tra agenti e linguaggi. Percorso consigliato:
+
+1. Riproduci una volta (agente + linguaggio/framework + comando).
+2. Apri una **issue** (bug) o una **PR** con fix minimo / test di regressione.
+3. Continua la challenge se puoi; annota l’incidente in `metrics.json` → `bugs_reported` e in `notes.md`.
+
+Tooling rotto su un certo stack **non** ti squalifica — segnalarlo è utile.
+
 ---
 
 ## Cosa non è questa challenge
@@ -209,6 +229,8 @@ Solo domande (opzionale): GitHub Discussions / Discord — il vecchio template 
 - Issue / bug: GitHub Issues  
 - Q&A challenge: GitHub Discussions (Announcements / Q&A)  
 - Community: Discord (vedi badge in README)  
+- Call di verifica: Google Meet (o simile) su richiesta dei maintainer  
+- Live pubbliche: annunciate su Discussions / Discord quando programmate  
 - Maintainer: Fabrizio Corpora  
 
 ---
diff --git a/docs/challenge.md b/docs/challenge.md
index 36dd9c3..65a5c9d 100644
--- a/docs/challenge.md
+++ b/docs/challenge.md
@@ -12,7 +12,7 @@ This challenge asks one question:
 
 > Does CodeDNA help (or not) when you do real AI-assisted development work — and can we measure it honestly?
 
-Bugs found during the challenge are welcome: open a GitHub issue and we will fix them upstream.
+**Experimental.** CodeDNA quality still depends on the **AI coding agent** and the **language / framework** you use. Some combinations work better than others. If install, annotation, hooks, or refresh misbehave on your stack, **open a GitHub issue or a fix PR** — that is part of the challenge, and we will harden upstream from your reports.
 
 ---
 
@@ -36,14 +36,24 @@ Bugs found during the challenge are welcome: open a GitHub issue and we will fix
 | Minimum valid submissions | **5** to unlock the prize (stretch goal: 10) |
 | If fewer than 5 valid PRs | Challenge still publishes results; prize is **not** awarded (or rolled to the next edition) |
 | Winner | Selected by Fabrizio Corpora + review team (not community vote) |
+| Honesty | Claims must be true and evidence-backed — fabricated metrics or narratives = disqualification |
+| Verification | Fabrizio or the review team may request a live review (Google Meet / similar) to walk through runs, logs, and repo setup |
+| Lives | When possible we will host public live sessions discussing submitted tests (with entrant consent where needed) |
 
 ### What we score (in order)
 
-1. **Protocol honesty** — fair stack, declared mode, reproducible notes  
+1. **Protocol honesty** — fair stack, declared mode, reproducible notes; **no invented results**  
 2. **Task quality** — real easy/medium/hard mix on a real codebase  
 3. **Evidence** — metrics + short narrative; pro **or** against CodeDNA is fine  
 4. **Bug reports** — actionable issues filed upstream count positively  
-5. **Clarity** — another engineer can re-run your comparison
+5. **Clarity** — another engineer can re-run your comparison  
+
+### Honesty, review calls, and lives
+
+- You **must not invent** pass/fail outcomes, timings, stack details, or narrative claims. If you cannot reproduce a number, mark the task inconclusive and say so.
+- Fabrizio Corpora and/or the review team **may ask for a video call** (e.g. Google Meet) to verify your submission: screen-share the project, rerun a sample task, and walk through `metrics.json` / notes.
+- Refusing a reasonable verification request without a good reason can void prize eligibility for that entry.
+- When feasible we will also run **public live sessions** covering interesting challenge runs (methodology, surprises, agent × language gaps). Participation in a live is optional unless you are a finalist asked for verification.
 
 ---
 
@@ -194,6 +204,16 @@ codedna init . --no-llm   # or with an LLM for rules:
 
 Questions only (optional): GitHub Discussions / Discord — the old “entry issue” template is not required.
 
+### If something breaks on your agent or language
+
+CodeDNA is still experimental across agents and languages. Prefer this path:
+
+1. Reproduce once (agent + language/framework + command).
+2. Open an **issue** (bug) or a **PR** with a minimal fix / regression test.
+3. Continue the challenge if you can; note the incident in `metrics.json` → `bugs_reported` and in `notes.md`.
+
+Broken tooling on a given stack does **not** disqualify you — reporting it is valuable.
+
 ---
 
 ## What this challenge is not
@@ -209,6 +229,8 @@ Questions only (optional): GitHub Discussions / Discord — the old “entry iss
 - Issues / bugs: GitHub Issues  
 - Challenge Q&A: GitHub Discussions (Announcements / Q&A)  
 - Community: Discord (see README badge)  
+- Verification calls: Google Meet (or similar) when requested by maintainers  
+- Public lives: announced on Discussions / Discord when scheduled  
 - Maintainer: Fabrizio Corpora  
 
 ---

From 60924d8a59fa81a180b695d4c3a0e2eca9be5f05 Mon Sep 17 00:00:00 2001
From: Cursor Agent <cursoragent@cursor.com>
Date: Tue, 8 Sep 2026 20:34:37 +0000
Subject: [PATCH 02/13] docs(challenge): real projects, install path, Meet
 review, live board
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Require declared tech stack and CodeDNA install steps, reject toy/throwaway
sites (≥25 source files), mandate same-task control vs CodeDNA runs, require
bug reports when experimental agent/language gaps appear, and clarify that
the public ranking updates as valid PRs arrive with optional Meet presentation.

AI-Agent: composer
AI-Provider: cursor
AI-Session: s_20260908_challenge_rules
AI-Visited: docs/challenge.md,docs/challenge.it.md,docs/challenge.html,challenge/metrics.schema.json,challenge/metrics.example.json,challenge/SUBMISSION_TEMPLATE.md
AI-Message: Tightened eligibility; dates still TBD; draft #18 content included on this branch

Co-authored-by: Fabrizio Corpora <Larens94@users.noreply.github.com>
---
 CHANGELOG.md                        |   2 +-
 README-it.md                        |   2 +-
 README.md                           |   2 +-
 challenge/README.md                 |   8 +-
 challenge/SUBMISSION_TEMPLATE.it.md |  41 +++++++----
 challenge/SUBMISSION_TEMPLATE.md    |  41 +++++++----
 challenge/metrics.example.json      |  12 ++-
 challenge/metrics.schema.json       |  54 +++++++++++++-
 docs/challenge-board.README.md      |  10 ++-
 docs/challenge.html                 |  70 +++++++++---------
 docs/challenge.it.md                | 109 +++++++++++++++++++---------
 docs/challenge.md                   | 109 +++++++++++++++++++---------
 12 files changed, 315 insertions(+), 145 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 9aa3478..22eec31 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -6,7 +6,7 @@ All notable changes to CodeDNA will be documented in this file.
 
 ### Added
 
-- **CodeDNA Challenge (€200)** — bilingual rules, required `metrics.json` (languages + frameworks + size/files), docs site page + static board (`docs/challenge.html`, `docs/challenge-board.json`). PR opens enrollment (no separate signup). Separate from historical SWE-bench benchmarks.
+- **CodeDNA Challenge (€200)** — bilingual rules, required `metrics.json` (languages + frameworks + size/files ≥25 + `install` + `bugs_reported`), real-project bar (no toy sites), same-task control vs CodeDNA, Meet presentation verification, docs site page + live public board updated as valid PRs arrive. PR opens enrollment (no separate signup). Separate from historical SWE-bench benchmarks.
 
 ### Added
 
diff --git a/README-it.md b/README-it.md
index 5c36030..826c4c4 100644
--- a/README-it.md
+++ b/README-it.md
@@ -48,7 +48,7 @@ Nessuna infrastruttura. Nessuna pipeline di retrieval. Nessuna memoria esterna.
 
 ## CodeDNA Challenge — €200
 
-Valutazione pubblica sul **tuo** progetto (non SWE-bench). Almeno 10 task con/senza CodeDNA, stack L0/L1/L2 a pari livello, PR con metriche — anche se vanno contro CodeDNA. Affermazioni oneste: il team può chiedere una review Meet; live sui test quando possibile.
+Valutazione pubblica sul **tuo** progetto (non SWE-bench). Progetto reale funzionante (≥25 file), stack tecnologico + install CodeDNA dichiarati, ≥10 **stessi** task con/senza CodeDNA, parity L0/L1/L2, PR con metriche — anche se vanno contro CodeDNA. I bug vanno segnalati (sperimentale tra agenti). Possibile presentazione Meet; la bacheca pubblica si aggiorna quando arrivano PR valide.
 
 - Bacheca: [larens94.github.io/codedna/challenge.html](https://larens94.github.io/codedna/challenge.html)
 - Regolamento: [`docs/challenge.it.md`](docs/challenge.it.md) (IT) · [`docs/challenge.md`](docs/challenge.md) (EN)
diff --git a/README.md b/README.md
index 7312378..801989e 100644
--- a/README.md
+++ b/README.md
@@ -48,7 +48,7 @@ No infrastructure. No retrieval pipeline. No external memory. The code carries i
 
 ## CodeDNA Challenge — €200
 
-Public **own-project** evaluation (not SWE-bench). ≥10 tasks with/without CodeDNA, fair L0/L1/L2 stack parity, metrics PR — even if results go against CodeDNA. Claims must be honest: the team may request a Meet review; live sessions on submitted tests when possible.
+Public **own-project** evaluation (not SWE-bench). Real working project (≥25 files), declared tech stack + CodeDNA install, ≥10 **same** tasks with/without CodeDNA, fair L0/L1/L2 parity, metrics PR — even if results go against CodeDNA. Bugs must be reported (experimental across agents). Meet presentation may be required; public board updates as valid PRs arrive.
 
 - Board: [larens94.github.io/codedna/challenge.html](https://larens94.github.io/codedna/challenge.html)
 - Rules: [`docs/challenge.md`](docs/challenge.md) (EN) · [`docs/challenge.it.md`](docs/challenge.it.md) (IT)
diff --git a/challenge/README.md b/challenge/README.md
index 09a75e3..3101fb8 100644
--- a/challenge/README.md
+++ b/challenge/README.md
@@ -4,19 +4,21 @@
 
 ```text
 challenge/<github-handle>/
-  metrics.json    # REQUIRED
+  metrics.json    # REQUIRED (stack + install + same-task control/codedna results)
   README.md       # optional
   notes.md        # optional
 ```
 
-No official signup. Public board (maintainer-updated): [`docs/challenge.html`](../docs/challenge.html) · data [`docs/challenge-board.json`](../docs/challenge-board.json)
+No official signup. Public board / ranking (updated as valid PRs arrive): [`docs/challenge.html`](../docs/challenge.html) · data [`docs/challenge-board.json`](../docs/challenge-board.json)
 
 **Metrics format**
 
 - Example: [`metrics.example.json`](./metrics.example.json)
 - Schema: [`metrics.schema.json`](./metrics.schema.json)
 
-Required project fields: `languages`, `frameworks`, `approx_source_files`, `size_band` (`S`/`M`/`L`/`XL`).
+Required: `languages`, `frameworks`, `approx_source_files` (≥25), `size_band` (`S`/`M`/`L`/`XL`), `install` (agent + steps), `bugs_reported` (array; empty OK).
+
+Real working projects only — toy sites / hello-world demos are rejected. Same tasks must be run with and without CodeDNA. Meet presentation may be required.
 
 **Rules**
 
diff --git a/challenge/SUBMISSION_TEMPLATE.it.md b/challenge/SUBMISSION_TEMPLATE.it.md
index 29b7ad0..82adfbf 100644
--- a/challenge/SUBMISSION_TEMPLATE.it.md
+++ b/challenge/SUBMISSION_TEMPLATE.it.md
@@ -7,12 +7,29 @@
 ### Partecipante
 
 - Handle:
-- Issue di iscrizione collegata:
 - Modalità challenge:
   - [ ] Parity (stessi L1/L2 su Control e CodeDNA)
   - [ ] Dichiarata `codedna-only` vs stack superiore
 
-### Stack sotto test
+### Stack tecnologico (obbligatorio)
+
+- Linguaggio/i:
+- Framework (obbligatorio — non solo il linguaggio):
+- Agente / modello:
+- File sorgente approx. (≥ **25**):
+- Size band (`S`/`M`/`L`/`XL`):
+- Note stack (DB, monorepo, infra — opzionale):
+- URL pubblico (opzionale) / “privato — solo metriche”:
+- [ ] È un **progetto reale e funzionante** (non sito giocattolo / hello-world / demo throwaway)
+
+### Install CodeDNA (obbligatorio)
+
+- Agente usato con CodeDNA:
+- Passi esatti di install / init (incolla i comandi):
+- Valore di `codedna install --tools` (se usato):
+- Install + annotation ok? sì / no — note:
+
+### Stack sotto test (L0/L1/L2)
 
 | Livello | Tool / file | Presente in Control? | Presente in CodeDNA? |
 |---|---|---|---|
@@ -20,22 +37,16 @@
 | L1 | | | |
 | L2 | | | |
 
-### Progetto
-
-- Linguaggio/i:
-- Framework (obbligatorio — non solo il linguaggio):
-- File sorgente approx.:
-- Size band (`S`/`M`/`L`/`XL`):
-- URL pubblico (opzionale) / “privato — solo metriche”:
-
 ### Metrics JSON (obbligatorio)
 
 - [ ] Presente `challenge/<handle>/metrics.json`
 - [ ] Partito da [`metrics.example.json`](./metrics.example.json) / conforme a [`metrics.schema.json`](./metrics.schema.json)
 - [ ] `schema_version` = `"1.0"`
-- [ ] Compilati `project.languages` + `project.frameworks` + `approx_source_files` + `size_band`
+- [ ] Compilati `project.languages` + `project.frameworks` + `approx_source_files` (≥25) + `size_band`
+- [ ] Compilati `install.agent` + `install.steps`
+- [ ] Presente `bugs_reported` (array vuoto OK se nessuno)
 - [ ] ≥10 task con mix `difficulty` (easy ≥3, medium ≥3, hard ≥2)
-- [ ] Ogni task ha risultati `control` e `codedna`
+- [ ] Gli **stessi task** hanno risultati `control` e `codedna`
 - [ ] `summary.favors` impostato (`codedna` | `control` | `tie` | `inconclusive`)
 - [ ] I risultati possono favorire **oppure** sfavorire CodeDNA (onestà OK)
 
@@ -47,11 +58,13 @@
 ### Bug trovati in CodeDNA
 
 - Link (oppure “nessuno”) — elencarli anche in `metrics.json` → `bugs_reported`:
+- [ ] Se qualcosa si è rotto su agente/linguaggio, ho aperto issue o PR di fix
 
 ### Checklist
 
 - [ ] Nessun secret / sorgente proprietaria in questa PR
 - [ ] Solo log redatti
 - [ ] Modalità + parity di stack dichiarate nel JSON (`mode`, `stack`)
-- [ ] Metriche e narrativa sono veritiere (non inventate); posso ripercorrerle in una call di review (Meet / simile) se richiesto
-- [ ] Capisco che claim inventati = squalifica
+- [ ] Metriche e narrativa sono veritiere (non inventate)
+- [ ] Posso presentare progetto e processo di test in una call di review (Meet / simile) se richiesto
+- [ ] Capisco che claim inventati o progetti finti = squalifica
diff --git a/challenge/SUBMISSION_TEMPLATE.md b/challenge/SUBMISSION_TEMPLATE.md
index 4474782..6d1e8b9 100644
--- a/challenge/SUBMISSION_TEMPLATE.md
+++ b/challenge/SUBMISSION_TEMPLATE.md
@@ -7,12 +7,29 @@
 ### Entrant
 
 - Handle:
-- Linked entry issue:
 - Challenge mode:
   - [ ] Parity (same L1/L2 on Control and CodeDNA)
   - [ ] Declared `codedna-only` vs higher stack
 
-### Stack under test
+### Tech stack (required)
+
+- Language(s):
+- Framework(s) (required — not language alone):
+- Agent / model:
+- Approx. source files (≥ **25**):
+- Size band (`S`/`M`/`L`/`XL`):
+- Tech stack notes (DB, monorepo, infra — optional):
+- Public URL (optional) / “private — metrics only”:
+- [ ] This is a **real working project** (not a toy site / hello-world / throwaway demo)
+
+### CodeDNA install (required)
+
+- Agent used with CodeDNA:
+- Exact install / init steps (paste commands):
+- `codedna install --tools` value (if any):
+- Did install + annotation work? yes / no — notes:
+
+### Stack under test (L0/L1/L2)
 
 | Level | Tools / files | Present in Control? | Present in CodeDNA? |
 |---|---|---|---|
@@ -20,22 +37,16 @@
 | L1 | | | |
 | L2 | | | |
 
-### Project
-
-- Language(s):
-- Framework(s) (required — not language alone):
-- Approx. source files:
-- Size band (`S`/`M`/`L`/`XL`):
-- Public URL (optional) / “private — metrics only”:
-
 ### Metrics JSON (required)
 
 - [ ] `challenge/<handle>/metrics.json` present
 - [ ] Copied from [`metrics.example.json`](./metrics.example.json) / matches [`metrics.schema.json`](./metrics.schema.json)
 - [ ] `schema_version` = `"1.0"`
-- [ ] `project.languages` + `project.frameworks` + `approx_source_files` + `size_band` filled
+- [ ] `project.languages` + `project.frameworks` + `approx_source_files` (≥25) + `size_band` filled
+- [ ] `install.agent` + `install.steps` filled
+- [ ] `bugs_reported` present (empty array OK if none)
 - [ ] ≥10 tasks with `difficulty` mix (easy ≥3, medium ≥3, hard ≥2)
-- [ ] Each task has both `control` and `codedna` results
+- [ ] **Same tasks** each have both `control` and `codedna` results
 - [ ] `summary.favors` set (`codedna` | `control` | `tie` | `inconclusive`)
 - [ ] Results may favor **or** disfavor CodeDNA (honesty OK)
 
@@ -47,11 +58,13 @@
 ### Bugs found in CodeDNA
 
 - Links (or “none”) — also list them in `metrics.json` → `bugs_reported`:
+- [ ] If something broke on my agent/language, I filed an issue or fix PR
 
 ### Checklist
 
 - [ ] No secrets / proprietary source in this PR
 - [ ] Redacted logs only
 - [ ] Mode + stack parity declared in JSON (`mode`, `stack`)
-- [ ] Metrics and narrative are truthful (not invented); I can walk through them on a review call (Meet / similar) if asked
-- [ ] I understand fabricated claims = disqualification
+- [ ] Metrics and narrative are truthful (not invented)
+- [ ] I can present the project and test process on a review call (Meet / similar) if asked
+- [ ] I understand fabricated claims or fake projects = disqualification
diff --git a/challenge/metrics.example.json b/challenge/metrics.example.json
index fb036ff..38a1e81 100644
--- a/challenge/metrics.example.json
+++ b/challenge/metrics.example.json
@@ -44,7 +44,8 @@
       "NestJS",
       "FastAPI"
     ],
-    "size_band": "M"
+    "size_band": "M",
+    "tech_stack_notes": "NestJS API + FastAPI workers, Postgres, monorepo apps/api + apps/worker"
   },
   "success_definition": "Task passes if the agent produced a correct change without the human rewriting the core logic.",
   "tasks": [
@@ -313,5 +314,12 @@
     "favors": "codedna",
     "one_line_verdict": "CodeDNA reduced wrong-file mistakes on hard tasks; still honest about one medium task needing a steer."
   },
-  "bugs_reported": []
+  "bugs_reported": [],
+  "install": {
+    "agent": "cursor",
+    "tools_flag": "cursor",
+    "steps": "pipx install git+https://github.com/Larens94/codedna.git\ncodedna install --path . --tools cursor\ncodedna init . --no-llm",
+    "worked": true,
+    "notes": "Install OK on Cursor; no adapter issues on this NestJS/FastAPI stack."
+  }
 }
diff --git a/challenge/metrics.schema.json b/challenge/metrics.schema.json
index 61e5692..a1eaf0e 100644
--- a/challenge/metrics.schema.json
+++ b/challenge/metrics.schema.json
@@ -10,8 +10,10 @@
     "mode",
     "stack",
     "project",
+    "install",
     "tasks",
-    "summary"
+    "summary",
+    "bugs_reported"
   ],
   "properties": {
     "schema_version": {
@@ -154,8 +156,8 @@
         },
         "approx_source_files": {
           "type": "integer",
-          "minimum": 1,
-          "description": "Approximate number of source files in the project under test."
+          "minimum": 25,
+          "description": "Approximate number of source files. Minimum 25. Toy landing pages / hello-world demos are rejected; must be a real working project (small is OK)."
         },
         "visibility": {
           "type": "string",
@@ -193,6 +195,13 @@
             "XL"
           ],
           "description": "S=<50 files, M=50-199, L=200-999, XL=1000+"
+        },
+        "tech_stack_notes": {
+          "type": [
+            "string",
+            "null"
+          ],
+          "description": "Optional free-text stack notes (DB, infra, monorepo layout, etc.)."
         }
       }
     },
@@ -334,7 +343,44 @@
           }
         }
       },
-      "default": []
+      "default": [],
+      "description": "Bugs found in CodeDNA during the challenge. Empty array if none. If you hit agent/language issues you MUST file issues or fix PRs and list them here."
+    },
+    "install": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": [
+        "agent",
+        "steps"
+      ],
+      "description": "How CodeDNA was installed for this run (required \u2014 publish your install path).",
+      "properties": {
+        "agent": {
+          "type": "string",
+          "minLength": 1,
+          "description": "AI coding agent used (e.g. cursor, claude-code, opencode, codex)."
+        },
+        "tools_flag": {
+          "type": [
+            "string",
+            "null"
+          ],
+          "description": "Value passed to `codedna install --tools`, if any."
+        },
+        "steps": {
+          "type": "string",
+          "minLength": 1,
+          "description": "Exact install/init commands and steps you ran."
+        },
+        "worked": {
+          "type": "boolean",
+          "description": "Whether install + annotation completed successfully on this agent/stack."
+        },
+        "notes": {
+          "type": "string",
+          "description": "Problems, workarounds, agent-specific gaps."
+        }
+      }
     }
   },
   "$defs": {
diff --git a/docs/challenge-board.README.md b/docs/challenge-board.README.md
index a3478b5..87a6d1e 100644
--- a/docs/challenge-board.README.md
+++ b/docs/challenge-board.README.md
@@ -1,11 +1,12 @@
 # Updating the public challenge board
 
-The live board is [`docs/challenge.html`](../docs/challenge.html).  
-It reads [`docs/challenge-board.json`](../docs/challenge-board.json) (static — no backend).
+The live board / ranking is [`docs/challenge.html`](../docs/challenge.html).  
+It reads [`docs/challenge-board.json`](../docs/challenge-board.json) (static — no backend).  
+**Update it as soon as valid PRs arrive** so the ranking stays publicly visible.
 
 ## When a metrics PR is accepted
 
-1. Merge the PR (`challenge/<handle>/metrics.json`).
+1. Merge the PR (`challenge/<handle>/metrics.json`) after checking: real project (≥25 files), stack declared, `install` present, same-task control/codedna pairs, bugs filed when relevant.
 2. Append a row to `docs/challenge-board.json` → `submissions` using fields from that JSON:
 
 ```json
@@ -29,4 +30,5 @@ It reads [`docs/challenge-board.json`](../docs/challenge-board.json) (static —
 3. Bump `updated_at` (ISO date).
 4. Commit + push `main` so GitHub Pages refreshes.
 
-Enrollment does **not** require an issue — the metrics PR is enough.
+Enrollment does **not** require an issue — the metrics PR is enough.  
+Optional: invite the entrant to a Meet presentation before or after the board row if verification is needed.
diff --git a/docs/challenge.html b/docs/challenge.html
index 60eba1e..d943874 100644
--- a/docs/challenge.html
+++ b/docs/challenge.html
@@ -206,17 +206,17 @@
   <div class="page-hero-inner">
     <div class="eyebrow"><span class="eyebrow-dot"></span><span data-i18n="eyebrow">Public evaluation · €200</span></div>
     <h1 data-i18n-html="h1">CodeDNA Challenge<br><em>€200</em></h1>
-    <p class="lead" data-i18n="lead">Test CodeDNA on your own project. Open a metrics PR to enroll — no separate signup. We update this board when submissions are merged.</p>
+    <p class="lead" data-i18n="lead">Test CodeDNA on your own real project. Declare stack + install. Same tasks with vs without CodeDNA. Open a metrics PR to enroll — the public board updates as valid PRs arrive.</p>
     <div class="hero-actions">
       <a class="btn-primary" href="https://github.com/Larens94/codedna/blob/main/challenge/metrics.example.json" target="_blank" data-i18n="ctaExample">Copy metrics.json →</a>
       <a class="btn-secondary" href="challenge.md" data-i18n="ctaRules">Full rules (EN)</a>
       <a class="btn-secondary" href="challenge.it.md" data-i18n="ctaRulesIt">Regolamento (IT)</a>
     </div>
     <p class="note" style="margin-top:1.75rem;max-width:640px" data-i18n-html="experimentalNote">
-      <strong>Experimental.</strong> Results can vary by AI coding agent and language.
-      If annotations or the CLI misbehave on your stack, open a
+      <strong>Experimental.</strong> Results can vary by AI coding agent and language (Claude Code, OpenCode, Cursor, Codex, …).
+      Publish how you installed CodeDNA. If annotations or the CLI misbehave, open a
       <a href="https://github.com/Larens94/codedna/issues/new">GitHub issue</a> or a fix
-      <a href="https://github.com/Larens94/codedna/compare">pull request</a> so we can harden support.
+      <a href="https://github.com/Larens94/codedna/compare">pull request</a>.
       That feedback is part of the challenge.
     </p>
   </div>
@@ -251,22 +251,22 @@ <h2 data-i18n="statusTitle">Challenge status</h2>
 
 <section class="section" id="how">
   <h2 data-i18n="howTitle">How it works</h2>
-  <p class="section-lead" data-i18n="howLead">Own project · ≥10 tasks · with vs without CodeDNA · fair L0/L1/L2 stack · metrics.json PR.</p>
+  <p class="section-lead" data-i18n="howLead">Real working project (≥25 files) · declare languages/frameworks/agent · ≥10 same tasks with vs without CodeDNA · publish install · report bugs · metrics.json PR.</p>
   <div class="steps">
     <div class="step">
       <div class="step-num">01</div>
-      <h3 data-i18n="s1t">Run on your repo</h3>
-      <p data-i18n="s1p">Declare languages, frameworks, file count and size band. Easy / medium / hard mix.</p>
+      <h3 data-i18n="s1t">Real project + stack</h3>
+      <p data-i18n="s1p">No toy sites. Declare languages, frameworks, file count, size band, and how you installed CodeDNA.</p>
     </div>
     <div class="step">
       <div class="step-num">02</div>
-      <h3 data-i18n="s2t">Keep the stack fair</h3>
-      <p data-i18n="s2p">If you use wiki, skills or Graphify, both conditions get them — unless you declare codedna-only mode.</p>
+      <h3 data-i18n="s2t">Same tasks, fair stack</h3>
+      <p data-i18n="s2p">Run the same ≥10 tasks with and without CodeDNA. Keep L1/L2 parity unless you declare codedna-only mode.</p>
     </div>
     <div class="step">
       <div class="step-num">03</div>
-      <h3 data-i18n="s3t">Open the PR</h3>
-      <p data-i18n="s3p">Add <code class="mono">challenge/&lt;handle&gt;/metrics.json</code>. That PR enrolls you. We merge and update this board.</p>
+      <h3 data-i18n="s3t">PR → live board</h3>
+      <p data-i18n="s3p">Open a metrics PR to enroll. The public ranking updates as valid PRs arrive. We may ask for a Meet presentation.</p>
     </div>
   </div>
   <div class="levels">
@@ -284,14 +284,14 @@ <h3 data-i18n="s3t">Open the PR</h3>
     </div>
   </div>
   <p class="note" style="margin-top:1.5rem" data-i18n-html="honestyNote">
-    <strong>Honesty.</strong> Do not invent results. Fabrizio or the review team may request a live review (e.g. Google Meet) to verify your runs.
+    <strong>Honesty.</strong> Do not invent results or fake projects. Fabrizio or the review team may require a conference call + presentation to verify your project and test process.
     When possible we will also host public lives discussing submitted tests.
   </p>
 </section>
 
 <section class="section" id="board">
   <h2 data-i18n="boardTitle">Public board</h2>
-  <p class="section-lead" data-i18n="boardLead">Language + framework, project size and file count are required so comparisons stay honest.</p>
+  <p class="section-lead" data-i18n="boardLead">Language + framework + install path and project size are required. Ranking updates as valid PRs arrive and stays public.</p>
 
   <div class="board-toolbar">
     <div class="board-updated" id="board-updated">updated —</div>
@@ -342,28 +342,28 @@ <h2 data-i18n="boardTitle">Public board</h2>
     navHome:'Home', navChallenge:'Challenge', navInstall:'Install',
     eyebrow:'Public evaluation · €200',
     h1:'CodeDNA Challenge<br><em>€200</em>',
-    lead:'Test CodeDNA on your own project. Open a metrics PR to enroll — no separate signup. We update this board when submissions are merged.',
+    lead:'Test CodeDNA on your own real project. Declare stack + install. Same tasks with vs without CodeDNA. Open a metrics PR to enroll — the public board updates as valid PRs arrive.',
     ctaExample:'Copy metrics.json →', ctaRules:'Full rules (EN)', ctaRulesIt:'Regolamento (IT)',
-    experimentalNote:'<strong>Experimental.</strong> Results can vary by AI coding agent and language. If annotations or the CLI misbehave on your stack, open a <a href="https://github.com/Larens94/codedna/issues/new">GitHub issue</a> or a fix <a href="https://github.com/Larens94/codedna/compare">pull request</a> so we can harden support. That feedback is part of the challenge.',
+    experimentalNote:'<strong>Experimental.</strong> Results can vary by AI coding agent and language (Claude Code, OpenCode, Cursor, Codex, …). Publish how you installed CodeDNA. If annotations or the CLI misbehave, open a <a href="https://github.com/Larens94/codedna/issues/new">GitHub issue</a> or a fix <a href="https://github.com/Larens94/codedna/compare">pull request</a>. That feedback is part of the challenge.',
     statusTitle:'Challenge status',
-    statusLead:'Maintainer-updated board. Source file: docs/challenge-board.json.',
+    statusLead:'Public ranking updated as valid PRs arrive. Source file: docs/challenge-board.json.',
     metaPrize:'Prize', metaPrizeSub:'paid after review',
     metaSubs:'Submissions', metaStatus:'Status',
     metaEnroll:'Enrollment', metaEnrollVal:'PR = in', metaEnrollSub:'no official signup',
     howTitle:'How it works',
-    howLead:'Own project · ≥10 tasks · with vs without CodeDNA · fair L0/L1/L2 stack · metrics.json PR.',
-    s1t:'Run on your repo', s1p:'Declare languages, frameworks, file count and size band. Easy / medium / hard mix.',
-    s2t:'Keep the stack fair', s2p:'If you use wiki, skills or Graphify, both conditions get them — unless you declare codedna-only mode.',
-    s3t:'Open the PR', s3p:'Add challenge/<handle>/metrics.json. That PR enrolls you. We merge and update this board.',
+    howLead:'Real working project (≥25 files) · declare languages/frameworks/agent · ≥10 same tasks with vs without CodeDNA · publish install · report bugs · metrics.json PR.',
+    s1t:'Real project + stack', s1p:'No toy sites. Declare languages, frameworks, file count, size band, and how you installed CodeDNA.',
+    s2t:'Same tasks, fair stack', s2p:'Run the same ≥10 tasks with and without CodeDNA. Keep L1/L2 parity unless you declare codedna-only mode.',
+    s3t:'PR → live board', s3p:'Open a metrics PR to enroll. The public ranking updates as valid PRs arrive. We may ask for a Meet presentation.',
     l0:'In-source headers — the layer under test.',
     l1:'LLM wiki, skill packs, instruction files — parity required.',
     l2:'Graph / structural memory — parity required.',
-    honestyNote:'<strong>Honesty.</strong> Do not invent results. Fabrizio or the review team may request a live review (e.g. Google Meet) to verify your runs. When possible we will also host public lives discussing submitted tests.',
+    honestyNote:'<strong>Honesty.</strong> Do not invent results or fake projects. Fabrizio or the review team may require a conference call + presentation to verify your project and test process. When possible we will also host public lives discussing submitted tests.',
     boardTitle:'Public board',
-    boardLead:'Language + framework, project size and file count are required so comparisons stay honest.',
+    boardLead:'Language + framework + install path and project size are required. Ranking updates as valid PRs arrive and stays public.',
     emptyTitle:'No submissions on the board yet',
-    emptyBody:'When a metrics PR is reviewed and merged, we add a row here by editing challenge-board.json.',
-    boardNote:'This board is static and curated. Opening a PR enrolls you automatically; appearing here means maintainers accepted the submission format. Agent × language gaps are expected — file issues or fix PRs.',
+    emptyBody:'As soon as a valid metrics PR is reviewed, we add a row here by editing challenge-board.json.',
+    boardNote:'Live ranking, curated by maintainers. Opening a PR enrolls you; appearing here means the submission format and stack declaration were accepted. Agent × language gaps are expected — file issues or fix PRs.',
     thHandle:'Entrant', thStack:'Language / framework', thSize:'Size', thFiles:'Files',
     thMode:'Mode', thScore:'Pass C / DNA', thFavors:'Favors', thPr:'PR',
     unlock:'min {n} to unlock', windowFallback:'dates TBD'
@@ -372,28 +372,28 @@ <h2 data-i18n="boardTitle">Public board</h2>
     navHome:'Home', navChallenge:'Challenge', navInstall:'Install',
     eyebrow:'Valutazione pubblica · €200',
     h1:'CodeDNA Challenge<br><em>€200</em>',
-    lead:'Prova CodeDNA sul tuo progetto. Apri una PR con le metriche per iscriverti — niente signup separato. Aggiorniamo questa bacheca quando le submission vengono mergiate.',
+    lead:'Prova CodeDNA sul tuo progetto reale. Dichiara stack + install. Stessi task con vs senza CodeDNA. Apri una PR metrics per iscriverti — la bacheca pubblica si aggiorna quando arrivano PR valide.',
     ctaExample:'Copia metrics.json →', ctaRules:'Rules (EN)', ctaRulesIt:'Regolamento (IT)',
-    experimentalNote:'<strong>Sperimentale.</strong> I risultati possono variare per agente AI di coding e linguaggio. Se annotazioni o CLI si comportano male sul tuo stack, apri una <a href="https://github.com/Larens94/codedna/issues/new">issue</a> o una <a href="https://github.com/Larens94/codedna/compare">pull request</a> di fix così possiamo rafforzare il supporto. Quel feedback fa parte della challenge.',
+    experimentalNote:'<strong>Sperimentale.</strong> I risultati possono variare per agente AI e linguaggio (Claude Code, OpenCode, Cursor, Codex, …). Pubblica come hai installato CodeDNA. Se annotazioni o CLI si comportano male, apri una <a href="https://github.com/Larens94/codedna/issues/new">issue</a> o una <a href="https://github.com/Larens94/codedna/compare">pull request</a> di fix. Quel feedback fa parte della challenge.',
     statusTitle:'Stato challenge',
-    statusLead:'Bacheca aggiornata dai maintainer. File sorgente: docs/challenge-board.json.',
+    statusLead:'Classifica pubblica aggiornata quando arrivano PR valide. File sorgente: docs/challenge-board.json.',
     metaPrize:'Premio', metaPrizeSub:'pagato dopo review',
     metaSubs:'Submission', metaStatus:'Stato',
     metaEnroll:'Iscrizione', metaEnrollVal:'PR = dentro', metaEnrollSub:'niente signup ufficiale',
     howTitle:'Come funziona',
-    howLead:'Progetto tuo · ≥10 task · con vs senza CodeDNA · stack L0/L1/L2 equo · PR metrics.json.',
-    s1t:'Sul tuo repo', s1p:'Dichiara linguaggi, framework, numero file e size band. Mix facili / medi / difficili.',
-    s2t:'Stack equo', s2p:'Se usi wiki, skill o Graphify, ci devono essere in entrambe le condizioni — salvo modalità codedna-only dichiarata.',
-    s3t:'Apri la PR', s3p:'Aggiungi challenge/<handle>/metrics.json. Quella PR ti iscrive. Merge e aggiornamento bacheca a cura nostra.',
+    howLead:'Progetto reale funzionante (≥25 file) · dichiara linguaggi/framework/agente · ≥10 stessi task con vs senza CodeDNA · pubblica install · segnala bug · PR metrics.json.',
+    s1t:'Progetto reale + stack', s1p:'Niente siti giocattolo. Dichiara linguaggi, framework, numero file, size band e come hai installato CodeDNA.',
+    s2t:'Stessi task, stack equo', s2p:'Esegui gli stessi ≥10 task con e senza CodeDNA. Tieni la parity L1/L2 salvo modalità codedna-only dichiarata.',
+    s3t:'PR → bacheca live', s3p:'Apri una PR metrics per iscriverti. La classifica pubblica si aggiorna quando arrivano PR valide. Possiamo chiedere una presentazione Meet.',
     l0:'Header in-source — layer sotto test.',
     l1:'Wiki LLM, skill pack, instruction file — parity obbligatoria.',
     l2:'Memoria graph / strutturale — parity obbligatoria.',
-    honestyNote:'<strong>Onestà.</strong> Non inventare i risultati. Fabrizio o il team di review possono chiedere una review live (es. Google Meet) per verificare le run. Quando possibile faremo anche live pubbliche sui test inviati.',
+    honestyNote:'<strong>Onestà.</strong> Non inventare risultati o progetti finti. Fabrizio o il team di review possono richiedere una call + presentazione per verificare progetto e processo di test. Quando possibile faremo anche live pubbliche sui test inviati.',
     boardTitle:'Bacheca pubblica',
-    boardLead:'Linguaggio + framework, dimensione progetto e numero file sono obbligatori per confronti onesti.',
+    boardLead:'Linguaggio + framework + percorso di install e dimensione progetto sono obbligatori. La classifica si aggiorna quando arrivano PR valide e resta pubblica.',
     emptyTitle:'Ancora nessuna submission in bacheca',
-    emptyBody:'Quando una PR metrics viene accettata e mergiata, aggiungiamo una riga modificando challenge-board.json.',
-    boardNote:'Bacheca statica e curata. Aprire una PR ti iscrive automaticamente; comparire qui significa che i maintainer hanno accettato il formato. Gap agente × linguaggio sono attesi — apri issue o PR di fix.',
+    emptyBody:'Appena una PR metrics valida viene reviewata, aggiungiamo una riga modificando challenge-board.json.',
+    boardNote:'Classifica live, curata dai maintainer. Aprire una PR ti iscrive; comparire qui significa che formato e dichiarazione di stack sono stati accettati. Gap agente × linguaggio sono attesi — apri issue o PR di fix.',
     thHandle:'Partecipante', thStack:'Linguaggio / framework', thSize:'Size', thFiles:'File',
     thMode:'Mode', thScore:'Pass C / DNA', thFavors:'Favors', thPr:'PR',
     unlock:'min {n} per sbloccare', windowFallback:'date TBD'
diff --git a/docs/challenge.it.md b/docs/challenge.it.md
index 4dd57e4..220dedb 100644
--- a/docs/challenge.it.md
+++ b/docs/challenge.it.md
@@ -18,13 +18,15 @@ Questa challenge pone una sola domanda:
 
 ## In sintesi
 
-1. Lavori sul **tuo** repo (non sui nostri vecchi benchmark).
-2. Esegui **almeno 10 task** (mix facili / medi / difficili).
-3. Confronti **con CodeDNA** vs **senza CodeDNA**.
+1. Lavori sul **tuo progetto reale e funzionante** (non sui nostri vecchi benchmark — non un sito giocattolo).
+2. Dichiari lo **stack tecnologico** completo (linguaggi, framework, agente, percorso di install).
+3. Esegui **almeno 10 task** (mix facili / medi / difficili) — gli **stessi task due volte**: una **senza** CodeDNA e una **con** CodeDNA.
 4. Tieni lo stack AI **equo** (vedi Livelli sotto).
-5. Apri una **Pull Request** con `metrics.json` — anche se CodeDNA risulta peggiore. **Quella PR è la tua iscrizione.**
-6. Fabrizio + team valutano complessità, protocollo ed evidenze; la bacheca pubblica si aggiorna a mano.
-7. Il premio si sblocca solo se si raggiunge il **minimo di partecipanti**.
+5. Documenti **come hai installato CodeDNA**; segnali bug / apri PR di fix quando qualcosa si rompe (atteso — è ancora sperimentale).
+6. Apri una **Pull Request** con `metrics.json` — anche se CodeDNA risulta peggiore. **Quella PR è la tua iscrizione.**
+7. La **bacheca / classifica pubblica si aggiorna quando arrivano PR valide** e resta visibile sul sito docs.
+8. Il progetto può essere chiamato a una **review live / presentazione** (call) così verifichiamo che i test siano reali.
+9. Il premio si sblocca solo se si raggiunge il **minimo di partecipanti**.
 
 ---
 
@@ -36,24 +38,30 @@ Questa challenge pone una sola domanda:
 | Submission valide minime | **5** per sbloccare il premio (obiettivo stretch: 10) |
 | Se ci sono meno di 5 PR valide | I risultati si pubblicano comunque; il premio **non** viene assegnato (o passa all’edizione successiva) |
 | Vincitore | Scelto da Fabrizio Corpora + team di review (non voto community) |
-| Onestà | Le affermazioni devono essere vere e supportate da evidenze — metriche o narrative inventate = squalifica |
-| Verifica | Fabrizio o il team di review possono chiedere una review live (Google Meet / simile) per ripercorrere run, log e setup del repo |
+| Onestà | Le affermazioni devono essere vere e supportate da evidenze — metriche inventate, progetti finti o narrative false = squalifica |
+| Verifica | Fabrizio o il team di review possono richiedere una **call + presentazione** (Google Meet / simile): mostrare il progetto reale, spiegare come hai fatto i test, ripercorrere install + `metrics.json` |
+| Bacheca pubblica | La classifica su [challenge.html](challenge.html) si aggiorna **appena arrivano PR valide** (review maintainer → riga in bacheca) e resta visibile |
 | Live | Quando possibile ospiteremo live pubbliche sui test inviati (con consenso del partecipante dove serve) |
 
 ### Cosa valutiamo (in ordine)
 
-1. **Onestà del protocollo** — stack equo, modalità dichiarata, note riproducibili; **niente risultati inventati**  
-2. **Qualità dei task** — mix reale facili/medi/difficili su una codebase vera  
-3. **Evidenze** — metriche + breve narrativa; a favore **o** contro CodeDNA va bene  
-4. **Bug report** — issue actionable aperte upstream contano positivamente  
+1. **Onestà del protocollo** — progetto reale, stack equo, modalità dichiarata, note riproducibili; **niente risultati inventati**  
+2. **Qualità dei task** — mix reale facili/medi/difficili su una codebase funzionante  
+3. **Evidenze** — metriche + percorso di install + breve narrativa; a favore **o** contro CodeDNA va bene  
+4. **Bug report / PR di fix** — obbligatori quando CodeDNA si rompe sul tuo agente o linguaggio; le issue actionable contano positivamente  
 5. **Chiarezza** — un altro engineer può ripetere il confronto  
 
 ### Onestà, call di review e live
 
 - **Non puoi inventare** esiti pass/fail, tempi, dettagli di stack o claim narrativi. Se non riesci a riprodurre un numero, marca il task come inconclusive e dillo.
-- Fabrizio Corpora e/o il team di review **possono chiedere una video call** (es. Google Meet) per verificare la submission: screen-share del progetto, riesecuzione di un task campione, walkthrough di `metrics.json` / note.
+- **Progetti finti o throwaway non sono ammessi.** Possiamo richiedere una **conferenza live + breve presentazione** in cui:
+  - dimostri che il progetto è reale e funzionante;
+  - spieghi lo stack tecnologico;
+  - mostri come hai installato CodeDNA;
+  - mostri come hai eseguito gli **stessi task** con e senza CodeDNA;
+  - rispondi su `metrics.json` / note.
 - Rifiutare senza motivo valido una richiesta ragionevole di verifica può annullare l’eleggibilità al premio per quella entry.
-- Quando possibile faremo anche **live pubbliche** sui test della challenge (metodologia, sorprese, gap agente × linguaggio). Partecipare a una live è opzionale, salvo finalisti chiamati in verifica.
+- Quando possibile faremo anche **live pubbliche** sui test della challenge (metodologia, sorprese, gap agente × linguaggio). Partecipare a una live è opzionale, salvo richiesta di verifica.
 
 ---
 
@@ -62,10 +70,11 @@ Questa challenge pone una sola domanda:
 | Fase | Quando |
 |---|---|
 | Annuncio + video | **TBD** |
-| Apertura iscrizioni | **TBD** |
+| Apertura submission (PR = iscrizione) | **TBD** |
 | Finestra challenge | **1 mese** dalla data di inizio |
 | Scadenza submission | fine della finestra (le PR devono essere aperte) |
-| Review | ~1–2 settimane dopo la scadenza |
+| Bacheca / classifica | aggiornata in continuo quando arrivano PR valide |
+| Review | ~1–2 settimane dopo la scadenza (più eventuali presentazioni Meet) |
 | Annuncio vincitore | **TBD** |
 
 ---
@@ -80,20 +89,26 @@ Questa challenge pone una sola domanda:
 
 ## Protocollo centrale (obbligatorio)
 
-### 1. Il tuo progetto
+### 1. Il tuo progetto (deve essere reale)
 
-Usa un repo che **mantieni tu** (lavoro o personale). Linea guida dimensione: abbastanza superficie perché 10 task siano significativi (indicativamente ≥20 file sorgente).
+Usa un repo che **mantieni tu** (lavoro o personale). Deve essere un **progetto reale e funzionante** — anche piccolo va bene; una pagina vetrina, un hello-world o una demo throwaway **no**.
 
-In `metrics.json` **devi** dichiarare:
+**Esempi rifiutati:** sito vetrina monostrato, CRUD vuoto senza logica di dominio, repo sintetici costruiti solo per la challenge.
+
+**Esempi accettati:** un servizio che usi o shippi davvero, un tool interno, una libreria con caller reali, un prodotto piccolo ma completo (≥ **25** file sorgente come soglia dura).
+
+In `metrics.json` **devi** dichiarare lo **stack tecnologico**:
 
 | Campo | Esempio |
 |---|---|
 | `languages` | `["TypeScript", "Python"]` |
-| `frameworks` | `["NestJS", "FastAPI"]` — il solo linguaggio non basta |
-| `approx_source_files` | `120` |
+| `frameworks` | `["NestJS", "FastAPI"]` — il solo linguaggio **non** basta |
+| `approx_source_files` | `120` (minimo **25**) |
 | `size_band` | `S` (<50) · `M` (50–199) · `L` (200–999) · `XL` (1000+) |
+| `tech_stack_notes` | opzionale — DB, layout monorepo, infra, lib principali |
+| `install.agent` + `install.steps` | quale agente AI + comandi esatti di install/init CodeDNA |
 
-### 2. Almeno dieci task
+### 2. Almeno dieci task — gli stessi in entrambe le condizioni
 
 Definisci **≥10** task **prima** delle run cronometrate (o congela l’elenco nella PR).
 
@@ -107,9 +122,11 @@ Mix suggerito:
 
 Per ogni task registra: obiettivo, difficoltà, agente/tool, successo/fallimento, note.
 
+**Critico:** ogni task va eseguito in **entrambe** le condizioni. Non inventare liste di task diverse per Control vs CodeDNA.
+
 ### 3. Due condizioni: senza vs con CodeDNA
 
-Per ogni task (o batch accoppiati) esegui:
+Per **ciascuno degli stessi task** esegui:
 
 | Condizione | Setup |
 |---|---|
@@ -155,12 +172,29 @@ Obbligatorio per ogni task (Control + CodeDNA):
 | `human_interventions` | Quante volte hai dovuto guidare |
 | `confidence_1_to_5` | Opzionale ma utile |
 
-Compila anche `summary.favors`: `codedna` | `control` | `tie` | `inconclusive`.
+Compila anche:
+
+- `summary.favors`: `codedna` | `control` | `tie` | `inconclusive`
+- `install` — **obbligatorio**: agente + passi esatti di install/init (e se ha funzionato)
+- `bugs_reported` — array **obbligatorio** (vuoto se nessuno); se qualcosa si è rotto, apri issue o PR di fix e elencala qui
 
 I numeri grezzi possono favorire CodeDNA **oppure no**. L’onestà batte il tifo.  
 `notes.md` è narrativa opzionale — **non sostituisce** `metrics.json`.
 
-### 6. Submission = Pull Request
+### 6. Percorso di install + bug report (obbligatori quando rilevanti)
+
+Pubblica **come hai installato CodeDNA** in `metrics.json` → `install` (comandi, flag `--tools`, agente).  
+CodeDNA è ancora **sperimentale**: può fallire o comportarsi male con alcuni tool agentici (Claude Code, OpenCode, Cursor, Codex, …) o linguaggi.
+
+Se install, annotation, hook o refresh si comportano male:
+
+1. Riproduci una volta (agente + linguaggio/framework + comando).
+2. Apri una **GitHub issue** o una **PR di fix**.
+3. Elencala in `bugs_reported` e menzionala in `notes.md`.
+
+Tooling rotto su un certo stack **non** ti squalifica — **nasconderlo** sì.
+
+### 7. Submission = Pull Request
 
 Apri una PR su `Larens94/codedna`. Copia la checklist nel body della PR:
 
@@ -184,13 +218,21 @@ challenge: <handle> — CodeDNA Challenge submission
 ```
 
 **Perché JSON:** ogni partecipante ha la sua cartella, le PR si mergiano senza conflitti, e a fine challenge aggreghiamo tutti i `metrics.json`.
+
+### 8. Bacheca / classifica pubblica
+
+La classifica live è [challenge.html](challenge.html) (dati: [`challenge-board.json`](challenge-board.json)).
+
+- Appena una PR metrics valida viene reviewata, aggiungiamo/aggiorniamo una riga — la bacheca resta **visibile pubblicamente**.
+- Aprire una PR ti iscrive; comparire in bacheca significa che i maintainer hanno accettato formato e dichiarazione di stack.
+
 ---
 
 ## Come entrare (niente signup ufficiale)
 
 **Non c’è iscrizione separata.** Quando apri una PR valida di challenge con `metrics.json`, sei iscritto.
 
-1. Installa CodeDNA e annota il progetto:
+1. Installa CodeDNA e annota il progetto (registra i passi esatti in `install`):
 
 ```bash
 pipx install git+https://github.com/Larens94/codedna.git
@@ -198,21 +240,21 @@ codedna install --path . --tools <tuo-agente>
 codedna init . --no-llm   # oppure con LLM per le rules:
 ```
 
-2. Esegui i ≥10 task (Control vs CodeDNA, stack equo).
-3. Apri una PR con `challenge/<handle>/metrics.json`.
-4. Facciamo review, merge e aggiorniamo la [bacheca pubblica](challenge.html).
+2. Esegui i ≥10 **stessi** task due volte (Control vs CodeDNA, stack equo).
+3. Apri una PR con `challenge/<handle>/metrics.json` (includi `install` + `bugs_reported`).
+4. Facciamo review, aggiorniamo la [bacheca pubblica](challenge.html) e possiamo invitarti a una presentazione Meet.
 
 Solo domande (opzionale): GitHub Discussions / Discord — il vecchio template “entry issue” non è obbligatorio.
 
 ### Se qualcosa si rompe sul tuo agente o linguaggio
 
-CodeDNA è ancora sperimentale tra agenti e linguaggi. Percorso consigliato:
+CodeDNA è ancora sperimentale tra agenti e linguaggi (Claude Code, OpenCode, Cursor, Codex, …). Percorso consigliato:
 
 1. Riproduci una volta (agente + linguaggio/framework + comando).
 2. Apri una **issue** (bug) o una **PR** con fix minimo / test di regressione.
 3. Continua la challenge se puoi; annota l’incidente in `metrics.json` → `bugs_reported` e in `notes.md`.
 
-Tooling rotto su un certo stack **non** ti squalifica — segnalarlo è utile.
+Tooling rotto su un certo stack **non** ti squalifica — **nasconderlo** sì.
 
 ---
 
@@ -220,7 +262,8 @@ Tooling rotto su un certo stack **non** ti squalifica — segnalarlo è utile.
 
 - Non è un rerun di SWE-bench / delle nostre tabelle F1 storiche  
 - Non è “annota solo i nostri repo di fixture”  
-- Non è consulenza gratis per noi — tieni l’IP del tuo progetto; noi reviewiamo solo la PR delle metriche  
+- Non è un sito throwaway o hello-world costruito solo per il premio  
+- Non è consulenza gratis per noi — tieni l’IP del tuo progetto; noi reviewiamo solo la PR delle metriche (+ eventuale presentazione Meet)  
 
 ---
 
@@ -229,7 +272,7 @@ Tooling rotto su un certo stack **non** ti squalifica — segnalarlo è utile.
 - Issue / bug: GitHub Issues  
 - Q&A challenge: GitHub Discussions (Announcements / Q&A)  
 - Community: Discord (vedi badge in README)  
-- Call di verifica: Google Meet (o simile) su richiesta dei maintainer  
+- Call di verifica / presentazione: Google Meet (o simile) su richiesta dei maintainer  
 - Live pubbliche: annunciate su Discussions / Discord quando programmate  
 - Maintainer: Fabrizio Corpora  
 
diff --git a/docs/challenge.md b/docs/challenge.md
index 65a5c9d..1d4fed8 100644
--- a/docs/challenge.md
+++ b/docs/challenge.md
@@ -18,13 +18,15 @@ This challenge asks one question:
 
 ## TL;DR
 
-1. Work on **your** repo (not our old benchmarks).
-2. Run **at least 10 tasks** (easy / medium / hard mix).
-3. Compare **with CodeDNA** vs **without CodeDNA**.
+1. Work on **your real, working project** (not our old benchmarks — not a toy site).
+2. Declare your full **tech stack** (languages, frameworks, agent, install path).
+3. Run **at least 10 tasks** (easy / medium / hard mix) — the **same tasks twice**: once **without** CodeDNA, once **with** CodeDNA.
 4. Keep your AI stack **fair** (see Levels below).
-5. Open a **Pull Request** with `metrics.json` — even if CodeDNA looks worse. **That PR is your enrollment.**
-6. Fabrizio + team review complexity, protocol compliance, and evidence; the public board is updated manually.
-7. Prize unlocks only if the **minimum participant count** is met.
+5. Document **how you installed CodeDNA**; report bugs / open fix PRs when something breaks (expected — still experimental).
+6. Open a **Pull Request** with `metrics.json` — even if CodeDNA looks worse. **That PR is your enrollment.**
+7. The **public board / ranking updates as valid PRs arrive** and stays visible on the docs site.
+8. Your project may be called to a **live review / presentation** (conference call) so we can verify the tests are real.
+9. Prize unlocks only if the **minimum participant count** is met.
 
 ---
 
@@ -36,24 +38,30 @@ This challenge asks one question:
 | Minimum valid submissions | **5** to unlock the prize (stretch goal: 10) |
 | If fewer than 5 valid PRs | Challenge still publishes results; prize is **not** awarded (or rolled to the next edition) |
 | Winner | Selected by Fabrizio Corpora + review team (not community vote) |
-| Honesty | Claims must be true and evidence-backed — fabricated metrics or narratives = disqualification |
-| Verification | Fabrizio or the review team may request a live review (Google Meet / similar) to walk through runs, logs, and repo setup |
+| Honesty | Claims must be true and evidence-backed — fabricated metrics, fake projects, or invented narratives = disqualification |
+| Verification | Fabrizio or the review team may require a **conference call + presentation** (Google Meet / similar): show the real project, explain how you ran the tests, walk through install + `metrics.json` |
+| Public board | Ranking on [challenge.html](challenge.html) is updated **as soon as valid PRs arrive** (maintainer review → board row) and remains publicly visible |
 | Lives | When possible we will host public live sessions discussing submitted tests (with entrant consent where needed) |
 
 ### What we score (in order)
 
-1. **Protocol honesty** — fair stack, declared mode, reproducible notes; **no invented results**  
-2. **Task quality** — real easy/medium/hard mix on a real codebase  
-3. **Evidence** — metrics + short narrative; pro **or** against CodeDNA is fine  
-4. **Bug reports** — actionable issues filed upstream count positively  
+1. **Protocol honesty** — real project, fair stack, declared mode, reproducible notes; **no invented results**  
+2. **Task quality** — real easy/medium/hard mix on a real working codebase  
+3. **Evidence** — metrics + install path + short narrative; pro **or** against CodeDNA is fine  
+4. **Bug reports / fix PRs** — required when CodeDNA breaks on your agent or language; actionable upstream issues count positively  
 5. **Clarity** — another engineer can re-run your comparison  
 
 ### Honesty, review calls, and lives
 
 - You **must not invent** pass/fail outcomes, timings, stack details, or narrative claims. If you cannot reproduce a number, mark the task inconclusive and say so.
-- Fabrizio Corpora and/or the review team **may ask for a video call** (e.g. Google Meet) to verify your submission: screen-share the project, rerun a sample task, and walk through `metrics.json` / notes.
+- **Fake or throwaway projects are not allowed.** We may require a **live conference + short presentation** where you:
+  - show that the project is real and working;
+  - explain your tech stack;
+  - walk through how you installed CodeDNA;
+  - show how you ran the **same tasks** with and without CodeDNA;
+  - answer questions on `metrics.json` / notes.
 - Refusing a reasonable verification request without a good reason can void prize eligibility for that entry.
-- When feasible we will also run **public live sessions** covering interesting challenge runs (methodology, surprises, agent × language gaps). Participation in a live is optional unless you are a finalist asked for verification.
+- When feasible we will also run **public live sessions** covering interesting challenge runs (methodology, surprises, agent × language gaps). Participation in a live is optional unless you are asked for verification.
 
 ---
 
@@ -62,10 +70,11 @@ This challenge asks one question:
 | Phase | When |
 |---|---|
 | Announcement + video | **TBD** |
-| Signup opens | **TBD** |
+| Submissions open (PR = enroll) | **TBD** |
 | Challenge window | **1 month** from start date |
 | Submission deadline | end of challenge window (PRs must be open) |
-| Review | ~1–2 weeks after deadline |
+| Board / ranking | updated continuously as valid PRs arrive |
+| Review | ~1–2 weeks after deadline (plus any Meet presentations) |
 | Winner announced | **TBD** |
 
 ---
@@ -80,20 +89,26 @@ This challenge asks one question:
 
 ## Core protocol (mandatory)
 
-### 1. Your project
+### 1. Your project (must be real)
 
-Use a repo **you** maintain (work or personal). Size guideline: enough surface that 10 tasks are meaningful (roughly ≥20 source files).
+Use a repo **you** maintain (work or personal). It must be a **real, working project** — small is fine; a tiny marketing page, hello-world, or throwaway demo is **not**.
 
-In `metrics.json` you **must** declare:
+**Rejected examples:** one-page “sito vetrina”, empty CRUD scaffold with no domain logic, synthetic repos built only for the challenge.
+
+**Accepted examples:** a service you actually use or ship, an internal tool, a library with real callers, a small but complete product (≥ **25** source files as a hard floor).
+
+In `metrics.json` you **must** declare your **tech stack**:
 
 | Field | Example |
 |---|---|
 | `languages` | `["TypeScript", "Python"]` |
-| `frameworks` | `["NestJS", "FastAPI"]` — language alone is not enough |
-| `approx_source_files` | `120` |
+| `frameworks` | `["NestJS", "FastAPI"]` — language alone is **not** enough |
+| `approx_source_files` | `120` (minimum **25**) |
 | `size_band` | `S` (<50) · `M` (50–199) · `L` (200–999) · `XL` (1000+) |
+| `tech_stack_notes` | optional — DB, monorepo layout, infra, major libs |
+| `install.agent` + `install.steps` | which AI agent + exact CodeDNA install/init commands |
 
-### 2. Ten tasks minimum
+### 2. Ten tasks minimum — same tasks both ways
 
 Define **≥10** tasks **before** you start the timed runs (or freeze the list in the PR).
 
@@ -107,9 +122,11 @@ Suggested mix:
 
 Record for each task: goal, difficulty, agent/tool, success/fail, notes.
 
+**Critical:** every task runs in **both** conditions. Do not invent a different task list for Control vs CodeDNA.
+
 ### 3. Two conditions: without vs with CodeDNA
 
-For each task (or paired batches), run:
+For **each of the same tasks**, run:
 
 | Condition | Setup |
 |---|---|
@@ -155,12 +172,29 @@ Required per task (Control + CodeDNA):
 | `human_interventions` | How often you had to steer |
 | `confidence_1_to_5` | Optional but useful |
 
-Also fill `summary.favors`: `codedna` | `control` | `tie` | `inconclusive`.
+Also fill:
+
+- `summary.favors`: `codedna` | `control` | `tie` | `inconclusive`
+- `install` — **required**: agent + exact install/init steps (and whether it worked)
+- `bugs_reported` — **required** array (empty if none); if something broke, file an issue or fix PR and list it here
 
 Raw numbers can favor CodeDNA **or not**. Honesty beats cheerleading.  
 Markdown notes (`notes.md`) are optional narrative — **do not replace** `metrics.json`.
 
-### 6. Submission = Pull Request
+### 6. Install path + bug reports (mandatory when relevant)
+
+Publish **how you installed CodeDNA** in `metrics.json` → `install` (commands, `--tools` flag, agent).  
+CodeDNA is still **experimental**: it may fail or behave poorly with some agentic tools (Claude Code, OpenCode, Cursor, Codex, …) or languages.
+
+If install, annotation, hooks, or refresh misbehaves:
+
+1. Reproduce once (agent + language/framework + command).
+2. Open a **GitHub issue** or a **fix PR**.
+3. List it in `bugs_reported` and mention it in `notes.md`.
+
+Broken tooling on a given stack does **not** disqualify you — **hiding** it does.
+
+### 7. Submission = Pull Request
 
 Open a PR against `Larens94/codedna`. Copy the checklist into the PR body:
 
@@ -184,13 +218,21 @@ challenge: <handle> — CodeDNA Challenge submission
 ```
 
 **Why JSON:** each entrant lands in their own folder, PRs merge cleanly, and we can aggregate all `metrics.json` files after the window closes.
+
+### 8. Public board / ranking
+
+The live ranking is [challenge.html](challenge.html) (data: [`challenge-board.json`](challenge-board.json)).
+
+- As soon as a valid metrics PR is reviewed, we add/update a row — the board stays **publicly visible**.
+- Opening a PR enrolls you; appearing on the board means maintainers accepted the submission format and stack declaration.
+
 ---
 
 ## How to enter (no official signup)
 
 There is **no separate registration**. When you open a valid challenge PR with `metrics.json`, you are enrolled.
 
-1. Install CodeDNA and annotate your project:
+1. Install CodeDNA and annotate your project (record the exact steps in `install`):
 
 ```bash
 pipx install git+https://github.com/Larens94/codedna.git
@@ -198,21 +240,21 @@ codedna install --path . --tools <your-agent>
 codedna init . --no-llm   # or with an LLM for rules:
 ```
 
-2. Run your ≥10 tasks (Control vs CodeDNA, fair stack).
-3. Open a PR adding `challenge/<handle>/metrics.json`.
-4. We review, merge, and update the [public board](challenge.html).
+2. Run your ≥10 **same** tasks twice (Control vs CodeDNA, fair stack).
+3. Open a PR adding `challenge/<handle>/metrics.json` (include `install` + `bugs_reported`).
+4. We review, update the [public board](challenge.html), and may invite you to a Meet presentation.
 
 Questions only (optional): GitHub Discussions / Discord — the old “entry issue” template is not required.
 
 ### If something breaks on your agent or language
 
-CodeDNA is still experimental across agents and languages. Prefer this path:
+CodeDNA is still experimental across agents and languages (Claude Code, OpenCode, Cursor, Codex, …). Prefer this path:
 
 1. Reproduce once (agent + language/framework + command).
 2. Open an **issue** (bug) or a **PR** with a minimal fix / regression test.
 3. Continue the challenge if you can; note the incident in `metrics.json` → `bugs_reported` and in `notes.md`.
 
-Broken tooling on a given stack does **not** disqualify you — reporting it is valuable.
+Broken tooling on a given stack does **not** disqualify you — **hiding** it does.
 
 ---
 
@@ -220,7 +262,8 @@ Broken tooling on a given stack does **not** disqualify you — reporting it is
 
 - Not a rerun of SWE-bench / our historical F1 tables  
 - Not “annotate our fixture repos only”  
-- Not unpaid consulting for us — you keep your project IP; we only review the metrics PR  
+- Not a tiny throwaway site or hello-world built only for the prize  
+- Not unpaid consulting for us — you keep your project IP; we only review the metrics PR (+ optional Meet presentation)  
 
 ---
 
@@ -229,7 +272,7 @@ Broken tooling on a given stack does **not** disqualify you — reporting it is
 - Issues / bugs: GitHub Issues  
 - Challenge Q&A: GitHub Discussions (Announcements / Q&A)  
 - Community: Discord (see README badge)  
-- Verification calls: Google Meet (or similar) when requested by maintainers  
+- Verification / presentation calls: Google Meet (or similar) when requested by maintainers  
 - Public lives: announced on Discussions / Discord when scheduled  
 - Maintainer: Fabrizio Corpora  
 

From c38fc02ec50c9cac1ef9c527e093d53707206cbf Mon Sep 17 00:00:00 2001
From: Cursor Agent <cursoragent@cursor.com>
Date: Tue, 8 Sep 2026 20:38:17 +0000
Subject: [PATCH 03/13] docs(challenge): polish public ranking page for launch
 video
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Rename board to ranking, add #ranking deep link + OG tags, improve empty
state for social screenshots, and soften window note to “coming soon”
until official dates are set.

AI-Agent: composer
AI-Provider: cursor
AI-Session: s_20260908_challenge_rules
AI-Visited: docs/challenge.html,docs/challenge-board.json,README.md
AI-Message: Ranking page ready for video; merge #19 to refresh GitHub Pages

Co-authored-by: Fabrizio Corpora <Larens94@users.noreply.github.com>
---
 README-it.md              |  2 +-
 README.md                 |  2 +-
 docs/challenge-board.json |  2 +-
 docs/challenge.html       | 73 +++++++++++++++++++++++++++------------
 4 files changed, 54 insertions(+), 25 deletions(-)

diff --git a/README-it.md b/README-it.md
index 826c4c4..77b4abb 100644
--- a/README-it.md
+++ b/README-it.md
@@ -50,7 +50,7 @@ Nessuna infrastruttura. Nessuna pipeline di retrieval. Nessuna memoria esterna.
 
 Valutazione pubblica sul **tuo** progetto (non SWE-bench). Progetto reale funzionante (≥25 file), stack tecnologico + install CodeDNA dichiarati, ≥10 **stessi** task con/senza CodeDNA, parity L0/L1/L2, PR con metriche — anche se vanno contro CodeDNA. I bug vanno segnalati (sperimentale tra agenti). Possibile presentazione Meet; la bacheca pubblica si aggiorna quando arrivano PR valide.
 
-- Bacheca: [larens94.github.io/codedna/challenge.html](https://larens94.github.io/codedna/challenge.html)
+- Classifica / bacheca: [larens94.github.io/codedna/challenge.html](https://larens94.github.io/codedna/challenge.html)
 - Regolamento: [`docs/challenge.it.md`](docs/challenge.it.md) (IT) · [`docs/challenge.md`](docs/challenge.md) (EN)
 - Submission: `challenge/<handle>/metrics.json` ([esempio](challenge/metrics.example.json)) — **aprire la PR ti iscrive**
 
diff --git a/README.md b/README.md
index 801989e..c7ad48b 100644
--- a/README.md
+++ b/README.md
@@ -50,7 +50,7 @@ No infrastructure. No retrieval pipeline. No external memory. The code carries i
 
 Public **own-project** evaluation (not SWE-bench). Real working project (≥25 files), declared tech stack + CodeDNA install, ≥10 **same** tasks with/without CodeDNA, fair L0/L1/L2 parity, metrics PR — even if results go against CodeDNA. Bugs must be reported (experimental across agents). Meet presentation may be required; public board updates as valid PRs arrive.
 
-- Board: [larens94.github.io/codedna/challenge.html](https://larens94.github.io/codedna/challenge.html)
+- Ranking / board: [larens94.github.io/codedna/challenge.html](https://larens94.github.io/codedna/challenge.html)
 - Rules: [`docs/challenge.md`](docs/challenge.md) (EN) · [`docs/challenge.it.md`](docs/challenge.it.md) (IT)
 - Submission: `challenge/<handle>/metrics.json` ([example](challenge/metrics.example.json)) — **opening the PR enrolls you**
 
diff --git a/docs/challenge-board.json b/docs/challenge-board.json
index a62d9bf..dcdb6ca 100644
--- a/docs/challenge-board.json
+++ b/docs/challenge-board.json
@@ -7,7 +7,7 @@
   "window": {
     "start": null,
     "end": null,
-    "note": "Dates TBD — set before public launch"
+    "note": "Coming soon — dates announced with the launch video"
   },
   "submissions": []
 }
diff --git a/docs/challenge.html b/docs/challenge.html
index d943874..56563a9 100644
--- a/docs/challenge.html
+++ b/docs/challenge.html
@@ -3,8 +3,14 @@
 <head>
 <meta charset="UTF-8">
 <meta name="viewport" content="width=device-width, initial-scale=1.0">
-<title>CodeDNA Challenge — €200</title>
-<meta name="description" content="Public CodeDNA Challenge: test on your own project, open a metrics PR to enroll, €200 prize. Live board on the docs site.">
+<title>CodeDNA Challenge — €200 · Public ranking</title>
+<meta name="description" content="CodeDNA Challenge €200: real-project evaluation with a live public ranking. Same tasks with vs without CodeDNA. Open a metrics PR to enroll.">
+<meta property="og:title" content="CodeDNA Challenge — €200">
+<meta property="og:description" content="Public ranking for the CodeDNA Challenge. Real projects · same tasks with/without CodeDNA · €200 prize.">
+<meta property="og:type" content="website">
+<meta property="og:url" content="https://larens94.github.io/codedna/challenge.html">
+<meta property="og:image" content="https://larens94.github.io/codedna/logo.png">
+<meta name="twitter:card" content="summary">
 <link rel="icon" href="logo.png" type="image/png">
 <link rel="preconnect" href="https://fonts.googleapis.com">
 <link href="https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700&family=JetBrains+Mono:wght@400;500&display=swap" rel="stylesheet">
@@ -150,10 +156,21 @@
 .favors-control { color: #d97706; font-weight: 600; }
 .favors-tie, .favors-inconclusive { color: var(--gray-600); font-weight: 600; }
 .empty {
-  padding: 3rem 1.5rem; text-align: center; color: var(--gray-600);
+  padding: 3.25rem 1.75rem; text-align: center; color: var(--gray-600);
   border: 1px dashed var(--gray-200); border-radius: var(--radius);
+  background:
+    linear-gradient(180deg, #fafafa 0%, #fff 100%);
+}
+.empty strong { display: block; color: var(--black); font-size: 1.05rem; margin-bottom: 0.55rem; }
+.empty span { display: block; max-width: 420px; margin: 0 auto 1.25rem; line-height: 1.65; }
+.empty-meta {
+  display: inline-flex; flex-wrap: wrap; gap: 0.5rem; justify-content: center;
+  font-family: 'JetBrains Mono', monospace; font-size: 11px; color: var(--gray-600);
+}
+.empty-meta span {
+  display: inline-block; margin: 0; padding: 0.35rem 0.65rem;
+  border: 1px solid var(--gray-200); border-radius: var(--radius); background: var(--white);
 }
-.empty strong { display: block; color: var(--black); margin-bottom: 0.4rem; }
 
 .note {
   margin-top: 1.25rem; font-size: 0.85rem; color: var(--gray-600);
@@ -208,7 +225,8 @@
     <h1 data-i18n-html="h1">CodeDNA Challenge<br><em>€200</em></h1>
     <p class="lead" data-i18n="lead">Test CodeDNA on your own real project. Declare stack + install. Same tasks with vs without CodeDNA. Open a metrics PR to enroll — the public board updates as valid PRs arrive.</p>
     <div class="hero-actions">
-      <a class="btn-primary" href="https://github.com/Larens94/codedna/blob/main/challenge/metrics.example.json" target="_blank" data-i18n="ctaExample">Copy metrics.json →</a>
+      <a class="btn-primary" href="#ranking" data-i18n="ctaRanking">View ranking →</a>
+      <a class="btn-secondary" href="https://github.com/Larens94/codedna/blob/main/challenge/metrics.example.json" target="_blank" data-i18n="ctaExample">Copy metrics.json →</a>
       <a class="btn-secondary" href="challenge.md" data-i18n="ctaRules">Full rules (EN)</a>
       <a class="btn-secondary" href="challenge.it.md" data-i18n="ctaRulesIt">Regolamento (IT)</a>
     </div>
@@ -224,7 +242,7 @@ <h1 data-i18n-html="h1">CodeDNA Challenge<br><em>€200</em></h1>
 
 <section class="section" id="status">
   <h2 data-i18n="statusTitle">Challenge status</h2>
-  <p class="section-lead" data-i18n="statusLead">Maintainer-updated board. Source file: <code class="mono">docs/challenge-board.json</code>.</p>
+  <p class="section-lead" data-i18n="statusLead">Public ranking updated as valid PRs arrive. Source file: <code class="mono">docs/challenge-board.json</code>.</p>
   <div class="meta-strip">
     <div class="meta-cell">
       <div class="meta-label" data-i18n="metaPrize">Prize</div>
@@ -239,7 +257,7 @@ <h2 data-i18n="statusTitle">Challenge status</h2>
     <div class="meta-cell">
       <div class="meta-label" data-i18n="metaStatus">Status</div>
       <div class="meta-value" id="meta-status">open</div>
-      <div class="meta-sub" id="meta-window">dates TBD</div>
+      <div class="meta-sub" id="meta-window">coming soon</div>
     </div>
     <div class="meta-cell">
       <div class="meta-label" data-i18n="metaEnroll">Enrollment</div>
@@ -289,8 +307,8 @@ <h3 data-i18n="s3t">PR → live board</h3>
   </p>
 </section>
 
-<section class="section" id="board">
-  <h2 data-i18n="boardTitle">Public board</h2>
+<section class="section" id="ranking">
+  <h2 data-i18n="boardTitle">Public ranking</h2>
   <p class="section-lead" data-i18n="boardLead">Language + framework + install path and project size are required. Ranking updates as valid PRs arrive and stays public.</p>
 
   <div class="board-toolbar">
@@ -315,12 +333,17 @@ <h2 data-i18n="boardTitle">Public board</h2>
 
   <div id="board-root">
     <div class="empty">
-      <strong data-i18n="emptyTitle">No submissions on the board yet</strong>
-      <span data-i18n="emptyBody">When a metrics PR is reviewed and merged, we add a row here by editing challenge-board.json.</span>
+      <strong data-i18n="emptyTitle">Ranking ready — waiting for first PRs</strong>
+      <span data-i18n="emptyBody">This is the live public leaderboard. As soon as valid metrics PRs arrive and are reviewed, rows appear here.</span>
+      <div class="empty-meta" aria-hidden="true">
+        <span>€200</span>
+        <span>min 5 to unlock</span>
+        <span>PR = enroll</span>
+      </div>
     </div>
   </div>
 
-  <p class="note" data-i18n="boardNote">This board is static and curated. Opening a PR enrolls you automatically; appearing here means maintainers accepted the submission format.</p>
+  <p class="note" data-i18n="boardNote">Live ranking, curated by maintainers. Opening a PR enrolls you; appearing here means the submission format and stack declaration were accepted. Agent × language gaps are expected — file issues or fix PRs.</p>
 </section>
 
 <footer class="site-footer">
@@ -343,7 +366,7 @@ <h2 data-i18n="boardTitle">Public board</h2>
     eyebrow:'Public evaluation · €200',
     h1:'CodeDNA Challenge<br><em>€200</em>',
     lead:'Test CodeDNA on your own real project. Declare stack + install. Same tasks with vs without CodeDNA. Open a metrics PR to enroll — the public board updates as valid PRs arrive.',
-    ctaExample:'Copy metrics.json →', ctaRules:'Full rules (EN)', ctaRulesIt:'Regolamento (IT)',
+    ctaRanking:'View ranking →', ctaExample:'Copy metrics.json →', ctaRules:'Full rules (EN)', ctaRulesIt:'Regolamento (IT)',
     experimentalNote:'<strong>Experimental.</strong> Results can vary by AI coding agent and language (Claude Code, OpenCode, Cursor, Codex, …). Publish how you installed CodeDNA. If annotations or the CLI misbehave, open a <a href="https://github.com/Larens94/codedna/issues/new">GitHub issue</a> or a fix <a href="https://github.com/Larens94/codedna/compare">pull request</a>. That feedback is part of the challenge.',
     statusTitle:'Challenge status',
     statusLead:'Public ranking updated as valid PRs arrive. Source file: docs/challenge-board.json.',
@@ -359,21 +382,21 @@ <h2 data-i18n="boardTitle">Public board</h2>
     l1:'LLM wiki, skill packs, instruction files — parity required.',
     l2:'Graph / structural memory — parity required.',
     honestyNote:'<strong>Honesty.</strong> Do not invent results or fake projects. Fabrizio or the review team may require a conference call + presentation to verify your project and test process. When possible we will also host public lives discussing submitted tests.',
-    boardTitle:'Public board',
+    boardTitle:'Public ranking',
     boardLead:'Language + framework + install path and project size are required. Ranking updates as valid PRs arrive and stays public.',
-    emptyTitle:'No submissions on the board yet',
+    emptyTitle:'Ranking ready — waiting for first PRs',
     emptyBody:'As soon as a valid metrics PR is reviewed, we add a row here by editing challenge-board.json.',
     boardNote:'Live ranking, curated by maintainers. Opening a PR enrolls you; appearing here means the submission format and stack declaration were accepted. Agent × language gaps are expected — file issues or fix PRs.',
     thHandle:'Entrant', thStack:'Language / framework', thSize:'Size', thFiles:'Files',
     thMode:'Mode', thScore:'Pass C / DNA', thFavors:'Favors', thPr:'PR',
-    unlock:'min {n} to unlock', windowFallback:'dates TBD'
+    unlock:'min {n} to unlock', windowFallback:'coming soon'
   },
   it: {
     navHome:'Home', navChallenge:'Challenge', navInstall:'Install',
     eyebrow:'Valutazione pubblica · €200',
     h1:'CodeDNA Challenge<br><em>€200</em>',
     lead:'Prova CodeDNA sul tuo progetto reale. Dichiara stack + install. Stessi task con vs senza CodeDNA. Apri una PR metrics per iscriverti — la bacheca pubblica si aggiorna quando arrivano PR valide.',
-    ctaExample:'Copia metrics.json →', ctaRules:'Rules (EN)', ctaRulesIt:'Regolamento (IT)',
+    ctaRanking:'Vedi classifica →', ctaExample:'Copia metrics.json →', ctaRules:'Rules (EN)', ctaRulesIt:'Regolamento (IT)',
     experimentalNote:'<strong>Sperimentale.</strong> I risultati possono variare per agente AI e linguaggio (Claude Code, OpenCode, Cursor, Codex, …). Pubblica come hai installato CodeDNA. Se annotazioni o CLI si comportano male, apri una <a href="https://github.com/Larens94/codedna/issues/new">issue</a> o una <a href="https://github.com/Larens94/codedna/compare">pull request</a> di fix. Quel feedback fa parte della challenge.',
     statusTitle:'Stato challenge',
     statusLead:'Classifica pubblica aggiornata quando arrivano PR valide. File sorgente: docs/challenge-board.json.',
@@ -389,14 +412,14 @@ <h2 data-i18n="boardTitle">Public board</h2>
     l1:'Wiki LLM, skill pack, instruction file — parity obbligatoria.',
     l2:'Memoria graph / strutturale — parity obbligatoria.',
     honestyNote:'<strong>Onestà.</strong> Non inventare risultati o progetti finti. Fabrizio o il team di review possono richiedere una call + presentazione per verificare progetto e processo di test. Quando possibile faremo anche live pubbliche sui test inviati.',
-    boardTitle:'Bacheca pubblica',
+    boardTitle:'Classifica pubblica',
     boardLead:'Linguaggio + framework + percorso di install e dimensione progetto sono obbligatori. La classifica si aggiorna quando arrivano PR valide e resta pubblica.',
-    emptyTitle:'Ancora nessuna submission in bacheca',
-    emptyBody:'Appena una PR metrics valida viene reviewata, aggiungiamo una riga modificando challenge-board.json.',
+    emptyTitle:'Classifica pronta — in attesa delle prime PR',
+    emptyBody:'Questa è la leaderboard pubblica live. Appena arrivano PR metrics valide e vengono reviewate, le righe compaiono qui.',
     boardNote:'Classifica live, curata dai maintainer. Aprire una PR ti iscrive; comparire qui significa che formato e dichiarazione di stack sono stati accettati. Gap agente × linguaggio sono attesi — apri issue o PR di fix.',
     thHandle:'Partecipante', thStack:'Linguaggio / framework', thSize:'Size', thFiles:'File',
     thMode:'Mode', thScore:'Pass C / DNA', thFavors:'Favors', thPr:'PR',
-    unlock:'min {n} per sbloccare', windowFallback:'date TBD'
+    unlock:'min {n} per sbloccare', windowFallback:'in arrivo'
   }
 };
 
@@ -449,7 +472,13 @@ <h2 data-i18n="boardTitle">Public board</h2>
 
   const root = document.getElementById('board-root');
   if (!rows.length) {
-    root.innerHTML = `<div class="empty"><strong>${dict.emptyTitle}</strong><span>${dict.emptyBody}</span></div>`;
+    root.innerHTML = `<div class="empty"><strong>${dict.emptyTitle}</strong><span>${dict.emptyBody}</span>
+      <div class="empty-meta" aria-hidden="true">
+        <span>€${data.prize_eur ?? 200}</span>
+        <span>${dict.unlock.replace('{n}', data.min_submissions_to_unlock ?? 5)}</span>
+        <span>${dict.metaEnrollVal}</span>
+      </div>
+    </div>`;
     return;
   }
 

From 71de2bab9e7ed3f97b81a593b2cca1413a3e21f2 Mon Sep 17 00:00:00 2001
From: Cursor Agent <cursoragent@cursor.com>
Date: Tue, 8 Sep 2026 20:38:28 +0000
Subject: [PATCH 04/13] docs(challenge): point headers at public ranking page

AI-Agent: composer
AI-Provider: cursor
AI-Session: s_20260908_challenge_rules
AI-Visited: docs/challenge.md,docs/challenge.it.md
AI-Message: Header labels match ranking page for video launch

Co-authored-by: Fabrizio Corpora <Larens94@users.noreply.github.com>
---
 docs/challenge.it.md | 2 +-
 docs/challenge.md    | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/docs/challenge.it.md b/docs/challenge.it.md
index 220dedb..a98f80a 100644
--- a/docs/challenge.it.md
+++ b/docs/challenge.it.md
@@ -6,7 +6,7 @@
 > **Durata:** 1 mese dalla data ufficiale di inizio  
 > **Iscrizione:** aprire la PR con le metriche = sei iscritto (niente signup ufficiale)  
 > **Non è un rerun di SWE-bench.** Provi CodeDNA sul **tuo** progetto.  
-> **Bacheca pubblica:** [challenge.html](challenge.html) sul sito docs
+> **Classifica pubblica:** [challenge.html](challenge.html) sul sito docs
 
 Questa challenge pone una sola domanda:
 
diff --git a/docs/challenge.md b/docs/challenge.md
index 1d4fed8..15e1eaa 100644
--- a/docs/challenge.md
+++ b/docs/challenge.md
@@ -6,7 +6,7 @@
 > **Duration:** 1 month from the official start date  
 > **Enrollment:** opening a metrics PR = you are enrolled (no separate signup)  
 > **Not a SWE-bench rerun.** You test CodeDNA on **your own project**.  
-> **Public board:** [challenge.html](challenge.html) on the docs site
+> **Public ranking:** [challenge.html](challenge.html) on the docs site
 
 This challenge asks one question:
 

From cd2ef8aeb74232344458b001ff14d24bb6b889b6 Mon Sep 17 00:00:00 2001
From: Cursor Agent <cursoragent@cursor.com>
Date: Tue, 8 Sep 2026 20:38:39 +0000
Subject: [PATCH 05/13] docs(challenge): align EN emptyBody copy on ranking
 page

AI-Agent: composer
AI-Provider: cursor
AI-Session: s_20260908_challenge_rules
AI-Visited: docs/challenge.html
AI-Message: EN empty-state I18N matches static HTML for video screenshots

Co-authored-by: Fabrizio Corpora <Larens94@users.noreply.github.com>
---
 docs/challenge.html | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/docs/challenge.html b/docs/challenge.html
index 56563a9..5c89ef2 100644
--- a/docs/challenge.html
+++ b/docs/challenge.html
@@ -385,7 +385,7 @@ <h2 data-i18n="boardTitle">Public ranking</h2>
     boardTitle:'Public ranking',
     boardLead:'Language + framework + install path and project size are required. Ranking updates as valid PRs arrive and stays public.',
     emptyTitle:'Ranking ready — waiting for first PRs',
-    emptyBody:'As soon as a valid metrics PR is reviewed, we add a row here by editing challenge-board.json.',
+    emptyBody:'This is the live public leaderboard. As soon as valid metrics PRs arrive and are reviewed, rows appear here.',
     boardNote:'Live ranking, curated by maintainers. Opening a PR enrolls you; appearing here means the submission format and stack declaration were accepted. Agent × language gaps are expected — file issues or fix PRs.',
     thHandle:'Entrant', thStack:'Language / framework', thSize:'Size', thFiles:'Files',
     thMode:'Mode', thScore:'Pass C / DNA', thFavors:'Favors', thPr:'PR',

From 0b621ae91a5e97df9d987a9c53647f0287f8ac84 Mon Sep 17 00:00:00 2001
From: Cursor Agent <cursoragent@cursor.com>
Date: Tue, 8 Sep 2026 20:42:42 +0000
Subject: [PATCH 06/13] fix(docs): challenge page mobile padding and nav
 spacing

site.css forced .section horizontal padding to 0; restore side gutters
on the challenge page and separate the mobile GitHub CTA from the
language toggle.

AI-Agent: composer
AI-Provider: cursor
AI-Session: s_20260908_challenge_rules
AI-Visited: docs/challenge.html,docs/site.css
AI-Message: Mobile gutters + GitHub/IT gap fixed via challenge-page overrides

Co-authored-by: Fabrizio Corpora <Larens94@users.noreply.github.com>
---
 docs/challenge.html | 69 +++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 66 insertions(+), 3 deletions(-)

diff --git a/docs/challenge.html b/docs/challenge.html
index 5c89ef2..9132c2c 100644
--- a/docs/challenge.html
+++ b/docs/challenge.html
@@ -189,17 +189,80 @@
   .meta-cell { border-right: none; border-bottom: 1px solid var(--gray-200); }
 }
 @media (max-width: 720px) {
-  .nav-links { display: none; position: absolute; top: 60px; left: 0; right: 0; flex-direction: column; gap: 0; background: var(--white); border-bottom: 1px solid var(--gray-200); padding: 0.75rem 1rem 1rem; }
+  .nav-links { display: none; position: absolute; top: 68px; left: 0; right: 0; flex-direction: column; gap: 0.5rem; background: var(--white); border-bottom: 1px solid var(--gray-200); padding: 0.85rem 1.1rem 1.15rem; }
   .nav-links.open { display: flex; }
-  .nav-links a { padding: 0.65rem 0; }
+  .nav-links a { padding: 0.7rem 0; }
+  .nav-links a.nav-cta { margin-top: 0.35rem; margin-bottom: 0.35rem; }
+  .nav-links .lang-toggle { margin-top: 0.15rem; align-self: stretch; }
   .nav-hamburger { display: flex; }
   .meta-strip, .steps, .levels { grid-template-columns: 1fr; }
   .page-hero h1 { max-width: none; }
+  .page-hero, .section, footer.site-footer { padding-left: 20px; padding-right: 20px; }
 }
 </style>
 <link rel="stylesheet" href="site.css">
+<style>
+/* Challenge page overrides — site.css zeros horizontal .section padding */
+body.challenge-page .page-hero {
+  padding: 120px clamp(18px, 5vw, 28px) 56px !important;
+}
+body.challenge-page .section {
+  max-width: 1100px !important;
+  margin-left: auto !important;
+  margin-right: auto !important;
+  padding: 64px clamp(18px, 5vw, 28px) !important;
+}
+body.challenge-page footer.site-footer {
+  padding: 2rem clamp(18px, 5vw, 28px) !important;
+}
+body.challenge-page .note {
+  margin-left: 0 !important;
+  margin-right: 0 !important;
+}
+body.challenge-page .nav-links .lang-toggle {
+  margin-top: 0;
+  align-self: flex-start;
+}
+@media (max-width: 720px) {
+  body.challenge-page .page-hero {
+    padding: 108px 20px 40px !important;
+  }
+  body.challenge-page .section {
+    padding: 48px 20px !important;
+  }
+  body.challenge-page footer.site-footer {
+    padding: 1.75rem 20px !important;
+  }
+  body.challenge-page .nav-links {
+    gap: 0.35rem !important;
+    padding: 0.85rem 1.1rem 1.15rem !important;
+    top: 68px !important;
+  }
+  body.challenge-page .nav-links a {
+    padding: 0.7rem 0 !important;
+  }
+  body.challenge-page .nav-links a.nav-cta {
+    margin-top: 0.55rem !important;
+    margin-bottom: 0.65rem !important;
+    width: 100%;
+    justify-content: center;
+  }
+  body.challenge-page .nav-links .lang-toggle {
+    margin-top: 0.15rem !important;
+    align-self: stretch;
+    min-height: 40px;
+  }
+  body.challenge-page .hero-actions {
+    width: 100%;
+  }
+  body.challenge-page .hero-actions .btn-primary,
+  body.challenge-page .hero-actions .btn-secondary {
+    width: 100%;
+  }
+}
+</style>
 </head>
-<body>
+<body class="challenge-page">
 
 <nav>
   <a class="nav-logo" href="index.html">

From 847247a9cbecc62d71eacfcecda83b056e8e56c1 Mon Sep 17 00:00:00 2001
From: Cursor Agent <cursoragent@cursor.com>
Date: Tue, 8 Sep 2026 20:54:17 +0000
Subject: [PATCH 07/13] docs(challenge): default public ranking page to Italian

Set html lang, JS default, and first-paint copy to Italian so the
challenge page opens in IT; language toggle still switches to English.

AI-Agent: composer
AI-Provider: cursor
AI-Session: s_20260908_challenge_rules
AI-Visited: docs/challenge.html,docs/challenge-board.json
AI-Message: Challenge page defaults to IT; EN available via toggle

Co-authored-by: Fabrizio Corpora <Larens94@users.noreply.github.com>
---
 docs/challenge.html | 89 +++++++++++++++++++++++----------------------
 1 file changed, 45 insertions(+), 44 deletions(-)

diff --git a/docs/challenge.html b/docs/challenge.html
index 9132c2c..5f2dc70 100644
--- a/docs/challenge.html
+++ b/docs/challenge.html
@@ -1,10 +1,10 @@
 <!DOCTYPE html>
-<html lang="en">
+<html lang="it">
 <head>
 <meta charset="UTF-8">
 <meta name="viewport" content="width=device-width, initial-scale=1.0">
-<title>CodeDNA Challenge — €200 · Public ranking</title>
-<meta name="description" content="CodeDNA Challenge €200: real-project evaluation with a live public ranking. Same tasks with vs without CodeDNA. Open a metrics PR to enroll.">
+<title>CodeDNA Challenge — €200 · Classifica pubblica</title>
+<meta name="description" content="CodeDNA Challenge €200: valutazione sul tuo progetto reale con classifica pubblica live. Stessi task con e senza CodeDNA. Apri una PR metrics per iscriverti.">
 <meta property="og:title" content="CodeDNA Challenge — €200">
 <meta property="og:description" content="Public ranking for the CodeDNA Challenge. Real projects · same tasks with/without CodeDNA · €200 prize.">
 <meta property="og:type" content="website">
@@ -275,7 +275,7 @@
     <a href="challenge.html" class="active" data-i18n="navChallenge">Challenge</a>
     <a href="install.html" data-i18n="navInstall">Install</a>
     <a href="https://github.com/Larens94/codedna" target="_blank" class="nav-cta">GitHub →</a>
-    <button class="lang-toggle" id="lang-toggle" onclick="toggleLang()">IT</button>
+    <button class="lang-toggle" id="lang-toggle" onclick="toggleLang()">EN</button>
   </div>
   <button class="nav-hamburger" onclick="document.getElementById('nav-links').classList.toggle('open')" aria-label="Menu">
     <span></span><span></span><span></span>
@@ -284,95 +284,95 @@
 
 <header class="page-hero">
   <div class="page-hero-inner">
-    <div class="eyebrow"><span class="eyebrow-dot"></span><span data-i18n="eyebrow">Public evaluation · €200</span></div>
+    <div class="eyebrow"><span class="eyebrow-dot"></span><span data-i18n="eyebrow">Valutazione pubblica · €200</span></div>
     <h1 data-i18n-html="h1">CodeDNA Challenge<br><em>€200</em></h1>
-    <p class="lead" data-i18n="lead">Test CodeDNA on your own real project. Declare stack + install. Same tasks with vs without CodeDNA. Open a metrics PR to enroll — the public board updates as valid PRs arrive.</p>
+    <p class="lead" data-i18n="lead">Prova CodeDNA sul tuo progetto reale. Dichiara stack + install. Stessi task con vs senza CodeDNA. Apri una PR metrics per iscriverti — la bacheca pubblica si aggiorna quando arrivano PR valide.</p>
     <div class="hero-actions">
-      <a class="btn-primary" href="#ranking" data-i18n="ctaRanking">View ranking →</a>
-      <a class="btn-secondary" href="https://github.com/Larens94/codedna/blob/main/challenge/metrics.example.json" target="_blank" data-i18n="ctaExample">Copy metrics.json →</a>
-      <a class="btn-secondary" href="challenge.md" data-i18n="ctaRules">Full rules (EN)</a>
+      <a class="btn-primary" href="#ranking" data-i18n="ctaRanking">Vedi classifica →</a>
+      <a class="btn-secondary" href="https://github.com/Larens94/codedna/blob/main/challenge/metrics.example.json" target="_blank" data-i18n="ctaExample">Copia metrics.json →</a>
+      <a class="btn-secondary" href="challenge.md" data-i18n="ctaRules">Rules (EN)</a>
       <a class="btn-secondary" href="challenge.it.md" data-i18n="ctaRulesIt">Regolamento (IT)</a>
     </div>
     <p class="note" style="margin-top:1.75rem;max-width:640px" data-i18n-html="experimentalNote">
-      <strong>Experimental.</strong> Results can vary by AI coding agent and language (Claude Code, OpenCode, Cursor, Codex, …).
-      Publish how you installed CodeDNA. If annotations or the CLI misbehave, open a
-      <a href="https://github.com/Larens94/codedna/issues/new">GitHub issue</a> or a fix
-      <a href="https://github.com/Larens94/codedna/compare">pull request</a>.
-      That feedback is part of the challenge.
+      <strong>Sperimentale.</strong> I risultati possono variare per agente AI e linguaggio (Claude Code, OpenCode, Cursor, Codex, …).
+      Pubblica come hai installato CodeDNA. Se annotazioni o CLI si comportano male, apri una
+      <a href="https://github.com/Larens94/codedna/issues/new">issue</a> o una
+      <a href="https://github.com/Larens94/codedna/compare">pull request</a> di fix.
+      Quel feedback fa parte della challenge.
     </p>
   </div>
 </header>
 
 <section class="section" id="status">
-  <h2 data-i18n="statusTitle">Challenge status</h2>
-  <p class="section-lead" data-i18n="statusLead">Public ranking updated as valid PRs arrive. Source file: <code class="mono">docs/challenge-board.json</code>.</p>
+  <h2 data-i18n="statusTitle">Stato challenge</h2>
+  <p class="section-lead" data-i18n="statusLead">Classifica pubblica aggiornata quando arrivano PR valide. File sorgente: <code class="mono">docs/challenge-board.json</code>.</p>
   <div class="meta-strip">
     <div class="meta-cell">
-      <div class="meta-label" data-i18n="metaPrize">Prize</div>
+      <div class="meta-label" data-i18n="metaPrize">Premio</div>
       <div class="meta-value" id="meta-prize">€200</div>
-      <div class="meta-sub" data-i18n="metaPrizeSub">paid after review</div>
+      <div class="meta-sub" data-i18n="metaPrizeSub">pagato dopo review</div>
     </div>
     <div class="meta-cell">
-      <div class="meta-label" data-i18n="metaSubs">Submissions</div>
+      <div class="meta-label" data-i18n="metaSubs">Submission</div>
       <div class="meta-value" id="meta-subs">0</div>
-      <div class="meta-sub" id="meta-unlock">min 5 to unlock</div>
+      <div class="meta-sub" id="meta-unlock">min 5 per sbloccare</div>
     </div>
     <div class="meta-cell">
-      <div class="meta-label" data-i18n="metaStatus">Status</div>
+      <div class="meta-label" data-i18n="metaStatus">Stato</div>
       <div class="meta-value" id="meta-status">open</div>
-      <div class="meta-sub" id="meta-window">coming soon</div>
+      <div class="meta-sub" id="meta-window">in arrivo</div>
     </div>
     <div class="meta-cell">
-      <div class="meta-label" data-i18n="metaEnroll">Enrollment</div>
-      <div class="meta-value" style="font-size:1.05rem" data-i18n="metaEnrollVal">PR = in</div>
-      <div class="meta-sub" data-i18n="metaEnrollSub">no official signup</div>
+      <div class="meta-label" data-i18n="metaEnroll">Iscrizione</div>
+      <div class="meta-value" style="font-size:1.05rem" data-i18n="metaEnrollVal">PR = dentro</div>
+      <div class="meta-sub" data-i18n="metaEnrollSub">niente signup ufficiale</div>
     </div>
   </div>
 </section>
 
 <section class="section" id="how">
-  <h2 data-i18n="howTitle">How it works</h2>
-  <p class="section-lead" data-i18n="howLead">Real working project (≥25 files) · declare languages/frameworks/agent · ≥10 same tasks with vs without CodeDNA · publish install · report bugs · metrics.json PR.</p>
+  <h2 data-i18n="howTitle">Come funziona</h2>
+  <p class="section-lead" data-i18n="howLead">Progetto reale funzionante (≥25 file) · dichiara linguaggi/framework/agente · ≥10 stessi task con vs senza CodeDNA · pubblica install · segnala bug · PR metrics.json.</p>
   <div class="steps">
     <div class="step">
       <div class="step-num">01</div>
-      <h3 data-i18n="s1t">Real project + stack</h3>
-      <p data-i18n="s1p">No toy sites. Declare languages, frameworks, file count, size band, and how you installed CodeDNA.</p>
+      <h3 data-i18n="s1t">Progetto reale + stack</h3>
+      <p data-i18n="s1p">Niente siti giocattolo. Dichiara linguaggi, framework, numero file, size band e come hai installato CodeDNA.</p>
     </div>
     <div class="step">
       <div class="step-num">02</div>
-      <h3 data-i18n="s2t">Same tasks, fair stack</h3>
-      <p data-i18n="s2p">Run the same ≥10 tasks with and without CodeDNA. Keep L1/L2 parity unless you declare codedna-only mode.</p>
+      <h3 data-i18n="s2t">Stessi task, stack equo</h3>
+      <p data-i18n="s2p">Esegui gli stessi ≥10 task con e senza CodeDNA. Tieni la parity L1/L2 salvo modalità codedna-only dichiarata.</p>
     </div>
     <div class="step">
       <div class="step-num">03</div>
-      <h3 data-i18n="s3t">PR → live board</h3>
-      <p data-i18n="s3p">Open a metrics PR to enroll. The public ranking updates as valid PRs arrive. We may ask for a Meet presentation.</p>
+      <h3 data-i18n="s3t">PR → bacheca live</h3>
+      <p data-i18n="s3p">Apri una PR metrics per iscriverti. La classifica pubblica si aggiorna quando arrivano PR valide. Possiamo chiedere una presentazione Meet.</p>
     </div>
   </div>
   <div class="levels">
     <div class="level">
       <div class="level-tag">L0 · CodeDNA</div>
-      <p data-i18n="l0">In-source headers — the layer under test.</p>
+      <p data-i18n="l0">Header in-source — layer sotto test.</p>
     </div>
     <div class="level">
       <div class="level-tag">L1 · Wiki / skills</div>
-      <p data-i18n="l1">LLM wiki, skill packs, instruction files — parity required.</p>
+      <p data-i18n="l1">Wiki LLM, skill pack, instruction file — parity obbligatoria.</p>
     </div>
     <div class="level">
       <div class="level-tag">L2 · Graphify</div>
-      <p data-i18n="l2">Graph / structural memory — parity required.</p>
+      <p data-i18n="l2">Memoria graph / strutturale — parity obbligatoria.</p>
     </div>
   </div>
   <p class="note" style="margin-top:1.5rem" data-i18n-html="honestyNote">
-    <strong>Honesty.</strong> Do not invent results or fake projects. Fabrizio or the review team may require a conference call + presentation to verify your project and test process.
-    When possible we will also host public lives discussing submitted tests.
+    <strong>Onestà.</strong> Non inventare risultati o progetti finti. Fabrizio o il team di review possono richiedere una call + presentazione per verificare progetto e processo di test.
+    Quando possibile faremo anche live pubbliche sui test inviati.
   </p>
 </section>
 
 <section class="section" id="ranking">
-  <h2 data-i18n="boardTitle">Public ranking</h2>
-  <p class="section-lead" data-i18n="boardLead">Language + framework + install path and project size are required. Ranking updates as valid PRs arrive and stays public.</p>
+  <h2 data-i18n="boardTitle">Classifica pubblica</h2>
+  <p class="section-lead" data-i18n="boardLead">Linguaggio + framework + percorso di install e dimensione progetto sono obbligatori. La classifica si aggiorna quando arrivano PR valide e resta pubblica.</p>
 
   <div class="board-toolbar">
     <div class="board-updated" id="board-updated">updated —</div>
@@ -396,8 +396,8 @@ <h2 data-i18n="boardTitle">Public ranking</h2>
 
   <div id="board-root">
     <div class="empty">
-      <strong data-i18n="emptyTitle">Ranking ready — waiting for first PRs</strong>
-      <span data-i18n="emptyBody">This is the live public leaderboard. As soon as valid metrics PRs arrive and are reviewed, rows appear here.</span>
+      <strong data-i18n="emptyTitle">Classifica pronta — in attesa delle prime PR</strong>
+      <span data-i18n="emptyBody">Questa è la leaderboard pubblica live. Appena arrivano PR metrics valide e vengono reviewate, le righe compaiono qui.</span>
       <div class="empty-meta" aria-hidden="true">
         <span>€200</span>
         <span>min 5 to unlock</span>
@@ -406,7 +406,7 @@ <h2 data-i18n="boardTitle">Public ranking</h2>
     </div>
   </div>
 
-  <p class="note" data-i18n="boardNote">Live ranking, curated by maintainers. Opening a PR enrolls you; appearing here means the submission format and stack declaration were accepted. Agent × language gaps are expected — file issues or fix PRs.</p>
+  <p class="note" data-i18n="boardNote">Classifica live, curata dai maintainer. Aprire una PR ti iscrive; comparire qui significa che formato e dichiarazione di stack sono stati accettati. Gap agente × linguaggio sono attesi — apri issue o PR di fix.</p>
 </section>
 
 <footer class="site-footer">
@@ -486,11 +486,12 @@ <h2 data-i18n="boardTitle">Public ranking</h2>
   }
 };
 
-let currentLang = 'en';
+let currentLang = 'it';
 let boardData = null;
 
 function applyI18n() {
   const dict = I18N[currentLang];
+  document.documentElement.lang = currentLang;
   document.querySelectorAll('[data-i18n]').forEach(el => {
     const k = el.getAttribute('data-i18n');
     if (dict[k] != null) el.textContent = dict[k];

From fe052094800f6603382949e41d3c7634cfec7e47 Mon Sep 17 00:00:00 2001
From: Cursor Agent <cursoragent@cursor.com>
Date: Tue, 8 Sep 2026 20:54:27 +0000
Subject: [PATCH 08/13] docs(challenge): Italian coming-soon note on board JSON

AI-Agent: composer
AI-Provider: cursor
AI-Session: s_20260908_challenge_rules
AI-Visited: docs/challenge-board.json
AI-Message: Board window note defaults to Italian for IT-first page

Co-authored-by: Fabrizio Corpora <Larens94@users.noreply.github.com>
---
 docs/challenge-board.json | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/docs/challenge-board.json b/docs/challenge-board.json
index dcdb6ca..a46e423 100644
--- a/docs/challenge-board.json
+++ b/docs/challenge-board.json
@@ -7,7 +7,7 @@
   "window": {
     "start": null,
     "end": null,
-    "note": "Coming soon — dates announced with the launch video"
+    "note": "In arrivo — date annunciate con il video di lancio"
   },
   "submissions": []
 }

From d3960884621866efcedaaeb38c8771062ab017c4 Mon Sep 17 00:00:00 2001
From: Cursor Agent <cursoragent@cursor.com>
Date: Tue, 8 Sep 2026 20:54:41 +0000
Subject: [PATCH 09/13] docs(challenge): put Italian rules CTA before English

AI-Agent: composer
AI-Provider: cursor
AI-Session: s_20260908_challenge_rules
AI-Visited: docs/challenge.html
AI-Message: IT-first CTA order on challenge hero

Co-authored-by: Fabrizio Corpora <Larens94@users.noreply.github.com>
---
 docs/challenge.html | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/docs/challenge.html b/docs/challenge.html
index 5f2dc70..26934bf 100644
--- a/docs/challenge.html
+++ b/docs/challenge.html
@@ -289,9 +289,9 @@ <h1 data-i18n-html="h1">CodeDNA Challenge<br><em>€200</em></h1>
     <p class="lead" data-i18n="lead">Prova CodeDNA sul tuo progetto reale. Dichiara stack + install. Stessi task con vs senza CodeDNA. Apri una PR metrics per iscriverti — la bacheca pubblica si aggiorna quando arrivano PR valide.</p>
     <div class="hero-actions">
       <a class="btn-primary" href="#ranking" data-i18n="ctaRanking">Vedi classifica →</a>
-      <a class="btn-secondary" href="https://github.com/Larens94/codedna/blob/main/challenge/metrics.example.json" target="_blank" data-i18n="ctaExample">Copia metrics.json →</a>
-      <a class="btn-secondary" href="challenge.md" data-i18n="ctaRules">Rules (EN)</a>
       <a class="btn-secondary" href="challenge.it.md" data-i18n="ctaRulesIt">Regolamento (IT)</a>
+      <a class="btn-secondary" href="challenge.md" data-i18n="ctaRules">Rules (EN)</a>
+      <a class="btn-secondary" href="https://github.com/Larens94/codedna/blob/main/challenge/metrics.example.json" target="_blank" data-i18n="ctaExample">Copia metrics.json →</a>
     </div>
     <p class="note" style="margin-top:1.75rem;max-width:640px" data-i18n-html="experimentalNote">
       <strong>Sperimentale.</strong> I risultati possono variare per agente AI e linguaggio (Claude Code, OpenCode, Cursor, Codex, …).

From c8cd9a172864dc2de6bc30dd7f2b57deb352cd1f Mon Sep 17 00:00:00 2001
From: Cursor Agent <cursoragent@cursor.com>
Date: Tue, 8 Sep 2026 21:03:25 +0000
Subject: [PATCH 10/13] feat(challenge): agent copy-pack, shared methodology,
 task list
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Add paste-ready IT/EN agent prompts with activity skeleton, allow
Control vs CodeDNA via two branches / checkouts / twin projects
(setup.layout), and wire “Copia per il tuo agente” on the ranking page.

AI-Agent: composer
AI-Provider: cursor
AI-Session: s_20260908_challenge_rules
AI-Visited: docs/challenge.html,docs/challenge.md,docs/challenge.it.md,challenge/metrics.schema.json
AI-Message: Agent prompt + two-branch methodology added; same protocol for all

Co-authored-by: Fabrizio Corpora <Larens94@users.noreply.github.com>
---
 README.md                           |  2 +-
 challenge/AGENT_PROMPT.md           | 13 ++++
 challenge/README.md                 | 17 +++--
 challenge/SUBMISSION_TEMPLATE.it.md | 10 +++
 challenge/SUBMISSION_TEMPLATE.md    | 10 +++
 challenge/TASKS_TEMPLATE.it.md      | 35 ++++++++++
 challenge/TASKS_TEMPLATE.md         | 35 ++++++++++
 challenge/metrics.example.json      |  6 ++
 challenge/metrics.schema.json       | 37 +++++++++++
 docs/challenge-agent-prompt.it.md   | 99 +++++++++++++++++++++++++++++
 docs/challenge-agent-prompt.md      | 99 +++++++++++++++++++++++++++++
 docs/challenge.html                 | 59 ++++++++++++-----
 docs/challenge.it.md                | 32 ++++++++--
 docs/challenge.md                   | 32 ++++++++--
 14 files changed, 452 insertions(+), 34 deletions(-)
 create mode 100644 challenge/AGENT_PROMPT.md
 create mode 100644 challenge/TASKS_TEMPLATE.it.md
 create mode 100644 challenge/TASKS_TEMPLATE.md
 create mode 100644 docs/challenge-agent-prompt.it.md
 create mode 100644 docs/challenge-agent-prompt.md

diff --git a/README.md b/README.md
index c7ad48b..f308de8 100644
--- a/README.md
+++ b/README.md
@@ -48,7 +48,7 @@ No infrastructure. No retrieval pipeline. No external memory. The code carries i
 
 ## CodeDNA Challenge — €200
 
-Public **own-project** evaluation (not SWE-bench). Real working project (≥25 files), declared tech stack + CodeDNA install, ≥10 **same** tasks with/without CodeDNA, fair L0/L1/L2 parity, metrics PR — even if results go against CodeDNA. Bugs must be reported (experimental across agents). Meet presentation may be required; public board updates as valid PRs arrive.
+Public **own-project** evaluation (not SWE-bench). Real working project (≥25 files), declared tech stack + CodeDNA install, same methodology for everyone (two branches / checkouts / twin projects), ≥10 **same** tasks with/without CodeDNA, copy-paste [agent prompt](docs/challenge-agent-prompt.md), metrics PR — even if results go against CodeDNA. Bugs must be reported (experimental across agents). Meet presentation may be required; public board updates as valid PRs arrive.
 
 - Ranking / board: [larens94.github.io/codedna/challenge.html](https://larens94.github.io/codedna/challenge.html)
 - Rules: [`docs/challenge.md`](docs/challenge.md) (EN) · [`docs/challenge.it.md`](docs/challenge.it.md) (IT)
diff --git a/challenge/AGENT_PROMPT.md b/challenge/AGENT_PROMPT.md
new file mode 100644
index 0000000..9899e2e
--- /dev/null
+++ b/challenge/AGENT_PROMPT.md
@@ -0,0 +1,13 @@
+# Agent prompts (copy-paste)
+
+Canonical prompts (also served on GitHub Pages):
+
+- Italiano: [`docs/challenge-agent-prompt.it.md`](../docs/challenge-agent-prompt.it.md)
+- English: [`docs/challenge-agent-prompt.md`](../docs/challenge-agent-prompt.md)
+
+Task checklist:
+
+- Italiano: [`TASKS_TEMPLATE.it.md`](./TASKS_TEMPLATE.it.md)
+- English: [`TASKS_TEMPLATE.md`](./TASKS_TEMPLATE.md)
+
+On the public page use **Copia per il tuo agente** → https://larens94.github.io/codedna/challenge.html
diff --git a/challenge/README.md b/challenge/README.md
index 3101fb8..f33f0c6 100644
--- a/challenge/README.md
+++ b/challenge/README.md
@@ -4,21 +4,28 @@
 
 ```text
 challenge/<github-handle>/
-  metrics.json    # REQUIRED (stack + install + same-task control/codedna results)
+  metrics.json    # REQUIRED (stack + install + setup.layout + same-task control/codedna results)
   README.md       # optional
   notes.md        # optional
 ```
 
-No official signup. Public board / ranking (updated as valid PRs arrive): [`docs/challenge.html`](../docs/challenge.html) · data [`docs/challenge-board.json`](../docs/challenge-board.json)
+No official signup. Public ranking (updated as valid PRs arrive): [`docs/challenge.html`](../docs/challenge.html) · data [`docs/challenge-board.json`](../docs/challenge-board.json)
+
+**Copy for your agent**
+
+- Italiano: [`docs/challenge-agent-prompt.it.md`](../docs/challenge-agent-prompt.it.md)
+- English: [`docs/challenge-agent-prompt.md`](../docs/challenge-agent-prompt.md)
+- Task lists: [`TASKS_TEMPLATE.it.md`](./TASKS_TEMPLATE.it.md) · [`TASKS_TEMPLATE.md`](./TASKS_TEMPLATE.md)
+- Pointer: [`AGENT_PROMPT.md`](./AGENT_PROMPT.md)
+
+Methodology (same for everyone): same ≥10 tasks with vs without CodeDNA via `two_branches` / `two_checkouts` / `two_projects`.
 
 **Metrics format**
 
 - Example: [`metrics.example.json`](./metrics.example.json)
 - Schema: [`metrics.schema.json`](./metrics.schema.json)
 
-Required: `languages`, `frameworks`, `approx_source_files` (≥25), `size_band` (`S`/`M`/`L`/`XL`), `install` (agent + steps), `bugs_reported` (array; empty OK).
-
-Real working projects only — toy sites / hello-world demos are rejected. Same tasks must be run with and without CodeDNA. Meet presentation may be required.
+Required: `languages`, `frameworks`, `approx_source_files` (≥25), `size_band`, `install`, `setup.layout`, `bugs_reported`.
 
 **Rules**
 
diff --git a/challenge/SUBMISSION_TEMPLATE.it.md b/challenge/SUBMISSION_TEMPLATE.it.md
index 82adfbf..27a0962 100644
--- a/challenge/SUBMISSION_TEMPLATE.it.md
+++ b/challenge/SUBMISSION_TEMPLATE.it.md
@@ -29,6 +29,15 @@
 - Valore di `codedna install --tools` (se usato):
 - Install + annotation ok? sì / no — note:
 
+### Setup Control vs CodeDNA (obbligatorio)
+
+- [ ] `two_branches` (es. `challenge/control` + `challenge/codedna`)
+- [ ] `two_checkouts` (due cartelle)
+- [ ] `two_projects` (progetti gemelli)
+- Ref Control / CodeDNA:
+- [ ] Prompt agente usato: [`docs/challenge-agent-prompt.it.md`](../docs/challenge-agent-prompt.it.md)
+- [ ] Lista task congelata: [`TASKS_TEMPLATE.it.md`](./TASKS_TEMPLATE.it.md)
+
 ### Stack sotto test (L0/L1/L2)
 
 | Livello | Tool / file | Presente in Control? | Presente in CodeDNA? |
@@ -44,6 +53,7 @@
 - [ ] `schema_version` = `"1.0"`
 - [ ] Compilati `project.languages` + `project.frameworks` + `approx_source_files` (≥25) + `size_band`
 - [ ] Compilati `install.agent` + `install.steps`
+- [ ] Compilato `setup.layout` (`two_branches` | `two_checkouts` | `two_projects`)
 - [ ] Presente `bugs_reported` (array vuoto OK se nessuno)
 - [ ] ≥10 task con mix `difficulty` (easy ≥3, medium ≥3, hard ≥2)
 - [ ] Gli **stessi task** hanno risultati `control` e `codedna`
diff --git a/challenge/SUBMISSION_TEMPLATE.md b/challenge/SUBMISSION_TEMPLATE.md
index 6d1e8b9..e89271a 100644
--- a/challenge/SUBMISSION_TEMPLATE.md
+++ b/challenge/SUBMISSION_TEMPLATE.md
@@ -29,6 +29,15 @@
 - `codedna install --tools` value (if any):
 - Did install + annotation work? yes / no — notes:
 
+### Control vs CodeDNA setup (required)
+
+- [ ] `two_branches` (e.g. `challenge/control` + `challenge/codedna`)
+- [ ] `two_checkouts` (two folders)
+- [ ] `two_projects` (twin projects)
+- Control / CodeDNA refs:
+- [ ] Agent prompt used: [`docs/challenge-agent-prompt.md`](../docs/challenge-agent-prompt.md)
+- [ ] Frozen task list: [`TASKS_TEMPLATE.md`](./TASKS_TEMPLATE.md)
+
 ### Stack under test (L0/L1/L2)
 
 | Level | Tools / files | Present in Control? | Present in CodeDNA? |
@@ -44,6 +53,7 @@
 - [ ] `schema_version` = `"1.0"`
 - [ ] `project.languages` + `project.frameworks` + `approx_source_files` (≥25) + `size_band` filled
 - [ ] `install.agent` + `install.steps` filled
+- [ ] `setup.layout` filled (`two_branches` | `two_checkouts` | `two_projects`)
 - [ ] `bugs_reported` present (empty array OK if none)
 - [ ] ≥10 tasks with `difficulty` mix (easy ≥3, medium ≥3, hard ≥2)
 - [ ] **Same tasks** each have both `control` and `codedna` results
diff --git a/challenge/TASKS_TEMPLATE.it.md b/challenge/TASKS_TEMPLATE.it.md
new file mode 100644
index 0000000..ac391c9
--- /dev/null
+++ b/challenge/TASKS_TEMPLATE.it.md
@@ -0,0 +1,35 @@
+# CodeDNA Challenge — template lista task
+
+> Congela questa lista **prima** delle run cronometrate. Esegui gli **stessi** ID su Control e su CodeDNA.  
+> Prompt completo per l’agente: [`docs/challenge-agent-prompt.it.md`](../docs/challenge-agent-prompt.it.md)
+
+## Setup (scegline uno)
+
+- [ ] `two_branches` — es. `challenge/control` + `challenge/codedna`
+- [ ] `two_checkouts` — due cartelle dello stesso progetto
+- [ ] `two_projects` — progetti gemelli dallo stesso codice di partenza
+
+Agente / modello: _______________  
+Note layout: _______________
+
+## Task (≥10 — easy ≥3, medium ≥3, hard ≥2)
+
+| ID | Difficoltà | Titolo (specifico del progetto) | Control ok? | CodeDNA ok? | Note |
+|---|---|---|---|---|---|
+| E1 | easy | | | | |
+| E2 | easy | | | | |
+| E3 | easy | | | | |
+| M1 | medium | | | | |
+| M2 | medium | | | | |
+| M3 | medium | | | | |
+| H1 | hard | | | | |
+| H2 | hard | | | | |
+| T9 | easy/medium/hard | | | | |
+| T10 | easy/medium/hard | | | | |
+
+Extra opzionali: E4, M4, H3…
+
+## Dopo le run
+
+- [ ] `metrics.json` compilato (`install`, `setup.layout`, `bugs_reported`, `summary.favors`)
+- [ ] PR aperta con submission template
diff --git a/challenge/TASKS_TEMPLATE.md b/challenge/TASKS_TEMPLATE.md
new file mode 100644
index 0000000..7e4a0b6
--- /dev/null
+++ b/challenge/TASKS_TEMPLATE.md
@@ -0,0 +1,35 @@
+# CodeDNA Challenge — task list template
+
+> Freeze this list **before** timed runs. Run the **same** IDs on Control and on CodeDNA.  
+> Italiano: adatta i titoli al tuo repo; scheletro anche in [`docs/challenge-agent-prompt.it.md`](../docs/challenge-agent-prompt.it.md).
+
+## Setup (pick one)
+
+- [ ] `two_branches` — e.g. `challenge/control` + `challenge/codedna`
+- [ ] `two_checkouts` — two folders of the same project
+- [ ] `two_projects` — twin projects from the same starting code
+
+Agent / model: _______________  
+Layout notes: _______________
+
+## Tasks (≥10 — easy ≥3, medium ≥3, hard ≥2)
+
+| ID | Difficulty | Title (project-specific) | Control pass? | CodeDNA pass? | Notes |
+|---|---|---|---|---|---|
+| E1 | easy | | | | |
+| E2 | easy | | | | |
+| E3 | easy | | | | |
+| M1 | medium | | | | |
+| M2 | medium | | | | |
+| M3 | medium | | | | |
+| H1 | hard | | | | |
+| H2 | hard | | | | |
+| T9 | easy/medium/hard | | | | |
+| T10 | easy/medium/hard | | | | |
+
+Optional extras: E4, M4, H3…
+
+## After runs
+
+- [ ] `metrics.json` filled (`install`, `setup.layout`, `bugs_reported`, `summary.favors`)
+- [ ] PR opened with submission template
diff --git a/challenge/metrics.example.json b/challenge/metrics.example.json
index 38a1e81..6da9603 100644
--- a/challenge/metrics.example.json
+++ b/challenge/metrics.example.json
@@ -321,5 +321,11 @@
     "steps": "pipx install git+https://github.com/Larens94/codedna.git\ncodedna install --path . --tools cursor\ncodedna init . --no-llm",
     "worked": true,
     "notes": "Install OK on Cursor; no adapter issues on this NestJS/FastAPI stack."
+  },
+  "setup": {
+    "layout": "two_branches",
+    "control_ref": "challenge/control",
+    "codedna_ref": "challenge/codedna",
+    "notes": "Same commit base; CodeDNA branch has codedna init annotations"
   }
 }
diff --git a/challenge/metrics.schema.json b/challenge/metrics.schema.json
index a1eaf0e..2f588e1 100644
--- a/challenge/metrics.schema.json
+++ b/challenge/metrics.schema.json
@@ -11,6 +11,7 @@
     "stack",
     "project",
     "install",
+    "setup",
     "tasks",
     "summary",
     "bugs_reported"
@@ -381,6 +382,42 @@
           "description": "Problems, workarounds, agent-specific gaps."
         }
       }
+    },
+    "setup": {
+      "type": "object",
+      "additionalProperties": false,
+      "required": [
+        "layout"
+      ],
+      "description": "How Control vs CodeDNA workspaces were separated.",
+      "properties": {
+        "layout": {
+          "type": "string",
+          "enum": [
+            "two_branches",
+            "two_checkouts",
+            "two_projects"
+          ],
+          "description": "two_branches | two_checkouts | two_projects"
+        },
+        "control_ref": {
+          "type": [
+            "string",
+            "null"
+          ],
+          "description": "Branch name, path, or URL for Control"
+        },
+        "codedna_ref": {
+          "type": [
+            "string",
+            "null"
+          ],
+          "description": "Branch name, path, or URL for CodeDNA"
+        },
+        "notes": {
+          "type": "string"
+        }
+      }
     }
   },
   "$defs": {
diff --git a/docs/challenge-agent-prompt.it.md b/docs/challenge-agent-prompt.it.md
new file mode 100644
index 0000000..5eea746
--- /dev/null
+++ b/docs/challenge-agent-prompt.it.md
@@ -0,0 +1,99 @@
+# Prompt agente — CodeDNA Challenge (€200)
+
+> Incolla questo intero messaggio nel tuo agente AI (Cursor, Claude Code, OpenCode, Codex, …).  
+> Pagina classifica: https://larens94.github.io/codedna/challenge.html  
+> Regolamento: https://github.com/Larens94/codedna/blob/main/docs/challenge.it.md  
+> English: [challenge-agent-prompt.md](challenge-agent-prompt.md)
+
+---
+
+## Ruolo
+
+Aiutami a eseguire la **CodeDNA Challenge** sul **mio progetto reale** in modo onesto e riproducibile. Non inventare metriche. Se qualcosa non è misurabile, marca `inconclusive`.
+
+## Metodologia (uguale per tutti)
+
+Confrontiamo gli **stessi ≥10 task** in due condizioni:
+
+| Condizione | Cosa |
+|---|---|
+| **A — Control** | Workflow AI **senza** CodeDNA (niente header / senza affidarsi a CodeDNA) |
+| **B — CodeDNA** | Stesso agente/modello **con** CodeDNA installato e annotato |
+
+### Setup ammesso (scegline uno e dichiaralo)
+
+1. **Due branch** nello stesso repo — es. `challenge/control` e `challenge/codedna`  
+2. **Due checkout / due cartelle** dello stesso progetto  
+3. **Due progetti gemelli** (stesso codice di partenza)
+
+Regole:
+
+- Gli **stessi task** in A e in B (stessa lista, stesso ordine se possibile).
+- Stesso agente, stesso modello, stessi layer L1/L2 (wiki/skills/Graphify) salvo modalità `codedna_only` dichiarata.
+- Progetto **reale e funzionante** (≥25 file sorgente). Niente sito vetrina / hello-world.
+
+## Stack da dichiarare
+
+Annota e poi metti in `metrics.json`:
+
+- linguaggi, framework (non solo il linguaggio), approx file, size band S/M/L/XL
+- agente + modello
+- come hai installato CodeDNA (`install.steps`)
+- layout usato: `two_branches` | `two_checkouts` | `two_projects`
+
+## Install CodeDNA (condizione B)
+
+```bash
+pipx install git+https://github.com/Larens94/codedna.git
+codedna install --path . --tools <mio-agente>
+codedna init . --no-llm
+```
+
+Se fallisce: riproduci, apri issue o PR di fix su Larens94/codedna, elenca in `bugs_reported`. È sperimentale — gap agente×linguaggio sono attesi.
+
+## Lista attività (adattale al mio repo, poi congelale)
+
+Crea **≥10 task** sul mio codice. Mix obbligatorio: easy ≥3, medium ≥3, hard ≥2.
+
+Usa questa checklist come scheletro (sostituisci i titoli con task **reali** del mio progetto):
+
+### Easy (≥3)
+
+- [ ] **E1** — Rename di un simbolo + aggiornare tutti i caller
+- [ ] **E2** — Aggiungere un campo/DTO/prop con validazione
+- [ ] **E3** — Fix bug chiaro con file già noto
+- [ ] **E4** (opz.) — Aggiornare un test esistente dopo un rename
+
+### Medium (≥3)
+
+- [ ] **M1** — Feature piccola cross-file (API + service + test)
+- [ ] **M2** — Refactor con invariante da non rompere
+- [ ] **M3** — Cambio firma / contratto e aggiornamento caller
+- [ ] **M4** (opz.) — Aggiungere logging/metriche in un percorso esistente
+
+### Hard (≥2)
+
+- [ ] **H1** — Bug multi-modulo (“dove cambio questo in sicurezza?”)
+- [ ] **H2** — Vincolo architetturale / auth / multi-tenant / confini di package
+- [ ] **H3** (opz.) — Migrazione o cambio schema con impatto su più layer
+
+Per **ogni** task registra Control e CodeDNA: `passed`, `minutes`/`turns`/`tool_calls` se disponibili, `wrong_file_or_module`, `human_interventions`, note.
+
+## Deliverable
+
+1. Congela la lista task **prima** delle run cronometrate.
+2. Esegui tutti i task in **A**, poi gli **stessi** in **B** (o interleaved, ma stessi ID).
+3. Compila `challenge/<mio-github-handle>/metrics.json` da  
+   https://github.com/Larens94/codedna/blob/main/challenge/metrics.example.json  
+   (schema: `metrics.schema.json`).
+4. Apri PR su `Larens94/codedna` con titolo:  
+   `challenge: <handle> — CodeDNA Challenge submission`  
+   Checklist: `challenge/SUBMISSION_TEMPLATE.it.md`
+5. Onestà: niente risultati inventati. Possibile Meet di verifica.
+
+## Cosa fare ora
+
+1. Ispeziona il mio repo e proponi la lista ≥10 task adattata (easy/medium/hard).
+2. Chiedimi quale setup uso: **due branch** / **due checkout** / **due progetti**.
+3. Prepara i comandi di install per la condizione B.
+4. Solo dopo conferma mia: esegui i task e compila le metriche.
diff --git a/docs/challenge-agent-prompt.md b/docs/challenge-agent-prompt.md
new file mode 100644
index 0000000..575cb85
--- /dev/null
+++ b/docs/challenge-agent-prompt.md
@@ -0,0 +1,99 @@
+# Agent prompt — CodeDNA Challenge (€200)
+
+> Paste this whole message into your AI coding agent (Cursor, Claude Code, OpenCode, Codex, …).  
+> Ranking page: https://larens94.github.io/codedna/challenge.html  
+> Rules: https://github.com/Larens94/codedna/blob/main/docs/challenge.md  
+> Italiano: [challenge-agent-prompt.it.md](challenge-agent-prompt.it.md)
+
+---
+
+## Role
+
+Help me run the **CodeDNA Challenge** on **my real project** honestly and reproducibly. Do not invent metrics. If something is not measurable, mark it `inconclusive`.
+
+## Methodology (same for everyone)
+
+We compare the **same ≥10 tasks** under two conditions:
+
+| Condition | What |
+|---|---|
+| **A — Control** | Normal AI workflow **without** CodeDNA (no headers / do not rely on CodeDNA) |
+| **B — CodeDNA** | Same agent/model **with** CodeDNA installed and annotated |
+
+### Allowed setup (pick one and declare it)
+
+1. **Two branches** in the same repo — e.g. `challenge/control` and `challenge/codedna`  
+2. **Two checkouts / two folders** of the same project  
+3. **Two twin projects** (same starting code)
+
+Rules:
+
+- The **same tasks** in A and B (same list, same order if possible).
+- Same agent, same model, same L1/L2 layers (wiki/skills/Graphify) unless `codedna_only` mode is declared.
+- **Real working project** (≥25 source files). No toy landing page / hello-world.
+
+## Stack to declare
+
+Capture, then put in `metrics.json`:
+
+- languages, frameworks (not language alone), approx files, size band S/M/L/XL
+- agent + model
+- how CodeDNA was installed (`install.steps`)
+- layout used: `two_branches` | `two_checkouts` | `two_projects`
+
+## Install CodeDNA (condition B)
+
+```bash
+pipx install git+https://github.com/Larens94/codedna.git
+codedna install --path . --tools <my-agent>
+codedna init . --no-llm
+```
+
+If it fails: reproduce, open an issue or fix PR on Larens94/codedna, list it in `bugs_reported`. Experimental — agent×language gaps are expected.
+
+## Activity list (adapt to my repo, then freeze)
+
+Create **≥10 tasks** on my code. Required mix: easy ≥3, medium ≥3, hard ≥2.
+
+Use this checklist as a skeleton (replace titles with **real** tasks from my project):
+
+### Easy (≥3)
+
+- [ ] **E1** — Rename a symbol and update all callers
+- [ ] **E2** — Add a field/DTO/prop with validation
+- [ ] **E3** — Fix a clear bug with a known file
+- [ ] **E4** (opt.) — Update an existing test after a rename
+
+### Medium (≥3)
+
+- [ ] **M1** — Small cross-file feature (API + service + test)
+- [ ] **M2** — Refactor with an invariant that must not break
+- [ ] **M3** — Signature / contract change and update callers
+- [ ] **M4** (opt.) — Add logging/metrics on an existing path
+
+### Hard (≥2)
+
+- [ ] **H1** — Multi-module bug (“where do I change this safely?”)
+- [ ] **H2** — Architectural / auth / multi-tenant / package-boundary constraint
+- [ ] **H3** (opt.) — Migration or schema change across layers
+
+For **each** task record Control and CodeDNA: `passed`, `minutes`/`turns`/`tool_calls` when available, `wrong_file_or_module`, `human_interventions`, notes.
+
+## Deliverable
+
+1. Freeze the task list **before** timed runs.
+2. Run every task in **A**, then the **same** ones in **B** (or interleaved, but same IDs).
+3. Fill `challenge/<my-github-handle>/metrics.json` from  
+   https://github.com/Larens94/codedna/blob/main/challenge/metrics.example.json  
+   (schema: `metrics.schema.json`).
+4. Open a PR on `Larens94/codedna` titled:  
+   `challenge: <handle> — CodeDNA Challenge submission`  
+   Checklist: `challenge/SUBMISSION_TEMPLATE.md`
+5. Honesty: no invented results. A Meet verification may be requested.
+
+## What to do now
+
+1. Inspect my repo and propose an adapted ≥10 task list (easy/medium/hard).
+2. Ask which setup I will use: **two branches** / **two checkouts** / **two projects**.
+3. Prepare install commands for condition B.
+4. Only after I confirm: run the tasks and fill the metrics.
diff --git a/docs/challenge.html b/docs/challenge.html
index 26934bf..6cb754a 100644
--- a/docs/challenge.html
+++ b/docs/challenge.html
@@ -91,10 +91,14 @@
   display: inline-flex; align-items: center; gap: 0.5rem;
   padding: 0.75rem 1.5rem; border-radius: var(--radius); font-size: 0.875rem; font-weight: 600;
   text-decoration: none; border: 1px solid var(--black); transition: opacity .15s, border-color .15s;
+  cursor: pointer; font-family: inherit; background: transparent; color: var(--black);
 }
 .btn-primary { background: var(--black); color: #fff; }
 .btn-secondary { background: transparent; color: var(--black); border-color: var(--gray-200); }
 .btn-secondary:hover { border-color: var(--black); }
+.copy-status {
+  margin-top: 0.75rem; font-size: 0.85rem; color: var(--gray-600);
+}
 
 .section { max-width: 1100px; margin: 0 auto; padding: 64px 28px; }
 .section + .section { border-top: 1px solid var(--gray-200); }
@@ -286,13 +290,15 @@
   <div class="page-hero-inner">
     <div class="eyebrow"><span class="eyebrow-dot"></span><span data-i18n="eyebrow">Valutazione pubblica · €200</span></div>
     <h1 data-i18n-html="h1">CodeDNA Challenge<br><em>€200</em></h1>
-    <p class="lead" data-i18n="lead">Prova CodeDNA sul tuo progetto reale. Dichiara stack + install. Stessi task con vs senza CodeDNA. Apri una PR metrics per iscriverti — la bacheca pubblica si aggiorna quando arrivano PR valide.</p>
+    <p class="lead" data-i18n="lead">Prova CodeDNA sul tuo progetto reale. Metodologia uguale per tutti: due branch o progetti gemelli, stessi task con vs senza CodeDNA. Copia il prompt per l’agente, poi apri la PR metrics.</p>
     <div class="hero-actions">
       <a class="btn-primary" href="#ranking" data-i18n="ctaRanking">Vedi classifica →</a>
+      <button type="button" class="btn-secondary" id="copy-agent-prompt" data-i18n="ctaCopyAgent">Copia per il tuo agente</button>
       <a class="btn-secondary" href="challenge.it.md" data-i18n="ctaRulesIt">Regolamento (IT)</a>
       <a class="btn-secondary" href="challenge.md" data-i18n="ctaRules">Rules (EN)</a>
       <a class="btn-secondary" href="https://github.com/Larens94/codedna/blob/main/challenge/metrics.example.json" target="_blank" data-i18n="ctaExample">Copia metrics.json →</a>
     </div>
+    <p class="copy-status" id="copy-status" hidden></p>
     <p class="note" style="margin-top:1.75rem;max-width:640px" data-i18n-html="experimentalNote">
       <strong>Sperimentale.</strong> I risultati possono variare per agente AI e linguaggio (Claude Code, OpenCode, Cursor, Codex, …).
       Pubblica come hai installato CodeDNA. Se annotazioni o CLI si comportano male, apri una
@@ -332,7 +338,7 @@ <h2 data-i18n="statusTitle">Stato challenge</h2>
 
 <section class="section" id="how">
   <h2 data-i18n="howTitle">Come funziona</h2>
-  <p class="section-lead" data-i18n="howLead">Progetto reale funzionante (≥25 file) · dichiara linguaggi/framework/agente · ≥10 stessi task con vs senza CodeDNA · pubblica install · segnala bug · PR metrics.json.</p>
+  <p class="section-lead" data-i18n="howLead">Metodologia uguale per tutti: progetto reale · due branch/checkout/progetti · ≥10 stessi task con vs senza CodeDNA · copia prompt per l’agente · pubblica install · segnala bug.</p>
   <div class="steps">
     <div class="step">
       <div class="step-num">01</div>
@@ -341,13 +347,13 @@ <h3 data-i18n="s1t">Progetto reale + stack</h3>
     </div>
     <div class="step">
       <div class="step-num">02</div>
-      <h3 data-i18n="s2t">Stessi task, stack equo</h3>
-      <p data-i18n="s2p">Esegui gli stessi ≥10 task con e senza CodeDNA. Tieni la parity L1/L2 salvo modalità codedna-only dichiarata.</p>
+      <h3 data-i18n="s2t">Due branch o due progetti</h3>
+      <p data-i18n="s2p">Stessi ≥10 task su Control e CodeDNA. Separa con due branch, due checkout o due progetti gemelli — dichiara `setup.layout`.</p>
     </div>
     <div class="step">
       <div class="step-num">03</div>
-      <h3 data-i18n="s3t">PR → bacheca live</h3>
-      <p data-i18n="s3p">Apri una PR metrics per iscriverti. La classifica pubblica si aggiorna quando arrivano PR valide. Possiamo chiedere una presentazione Meet.</p>
+      <h3 data-i18n="s3t">Copia prompt → PR</h3>
+      <p data-i18n="s3p">Copia il prompt per il tuo agente, esegui la lista attività, apri la PR metrics. La classifica si aggiorna quando arrivano PR valide.</p>
     </div>
   </div>
   <div class="levels">
@@ -428,8 +434,8 @@ <h2 data-i18n="boardTitle">Classifica pubblica</h2>
     navHome:'Home', navChallenge:'Challenge', navInstall:'Install',
     eyebrow:'Public evaluation · €200',
     h1:'CodeDNA Challenge<br><em>€200</em>',
-    lead:'Test CodeDNA on your own real project. Declare stack + install. Same tasks with vs without CodeDNA. Open a metrics PR to enroll — the public board updates as valid PRs arrive.',
-    ctaRanking:'View ranking →', ctaExample:'Copy metrics.json →', ctaRules:'Full rules (EN)', ctaRulesIt:'Regolamento (IT)',
+    lead:'Test CodeDNA on your own real project. Same methodology for everyone: two branches or twin projects, same tasks with vs without CodeDNA. Copy the agent prompt, then open a metrics PR.',
+    ctaRanking:'View ranking →', ctaCopyAgent:'Copy for your agent', ctaExample:'Copy metrics.json →', ctaRules:'Full rules (EN)', ctaRulesIt:'Regolamento (IT)', ctaCopied:'Copied — paste into your AI agent', ctaCopyFail:'Could not copy — open challenge-agent-prompt.md',
     experimentalNote:'<strong>Experimental.</strong> Results can vary by AI coding agent and language (Claude Code, OpenCode, Cursor, Codex, …). Publish how you installed CodeDNA. If annotations or the CLI misbehave, open a <a href="https://github.com/Larens94/codedna/issues/new">GitHub issue</a> or a fix <a href="https://github.com/Larens94/codedna/compare">pull request</a>. That feedback is part of the challenge.',
     statusTitle:'Challenge status',
     statusLead:'Public ranking updated as valid PRs arrive. Source file: docs/challenge-board.json.',
@@ -437,10 +443,10 @@ <h2 data-i18n="boardTitle">Classifica pubblica</h2>
     metaSubs:'Submissions', metaStatus:'Status',
     metaEnroll:'Enrollment', metaEnrollVal:'PR = in', metaEnrollSub:'no official signup',
     howTitle:'How it works',
-    howLead:'Real working project (≥25 files) · declare languages/frameworks/agent · ≥10 same tasks with vs without CodeDNA · publish install · report bugs · metrics.json PR.',
+    howLead:'Same methodology for everyone: real project · two branches/checkouts/projects · ≥10 same tasks with vs without CodeDNA · copy agent prompt · publish install · report bugs.',
     s1t:'Real project + stack', s1p:'No toy sites. Declare languages, frameworks, file count, size band, and how you installed CodeDNA.',
-    s2t:'Same tasks, fair stack', s2p:'Run the same ≥10 tasks with and without CodeDNA. Keep L1/L2 parity unless you declare codedna-only mode.',
-    s3t:'PR → live board', s3p:'Open a metrics PR to enroll. The public ranking updates as valid PRs arrive. We may ask for a Meet presentation.',
+    s2t:'Two branches or twin projects', s2p:'Same ≥10 tasks on Control and CodeDNA. Split with two branches, two checkouts, or twin projects — declare setup.layout.',
+    s3t:'Copy prompt → PR', s3p:'Copy the agent prompt, run the activity list, open the metrics PR. Ranking updates as valid PRs arrive.',
     l0:'In-source headers — the layer under test.',
     l1:'LLM wiki, skill packs, instruction files — parity required.',
     l2:'Graph / structural memory — parity required.',
@@ -458,8 +464,8 @@ <h2 data-i18n="boardTitle">Classifica pubblica</h2>
     navHome:'Home', navChallenge:'Challenge', navInstall:'Install',
     eyebrow:'Valutazione pubblica · €200',
     h1:'CodeDNA Challenge<br><em>€200</em>',
-    lead:'Prova CodeDNA sul tuo progetto reale. Dichiara stack + install. Stessi task con vs senza CodeDNA. Apri una PR metrics per iscriverti — la bacheca pubblica si aggiorna quando arrivano PR valide.',
-    ctaRanking:'Vedi classifica →', ctaExample:'Copia metrics.json →', ctaRules:'Rules (EN)', ctaRulesIt:'Regolamento (IT)',
+    lead:'Prova CodeDNA sul tuo progetto reale. Metodologia uguale per tutti: due branch o progetti gemelli, stessi task con vs senza CodeDNA. Copia il prompt per l’agente, poi apri la PR metrics.',
+    ctaRanking:'Vedi classifica →', ctaCopyAgent:'Copia per il tuo agente', ctaExample:'Copia metrics.json →', ctaRules:'Rules (EN)', ctaRulesIt:'Regolamento (IT)', ctaCopied:'Copiato — incollalo nel tuo agente AI', ctaCopyFail:'Copia non riuscita — apri challenge-agent-prompt.it.md',
     experimentalNote:'<strong>Sperimentale.</strong> I risultati possono variare per agente AI e linguaggio (Claude Code, OpenCode, Cursor, Codex, …). Pubblica come hai installato CodeDNA. Se annotazioni o CLI si comportano male, apri una <a href="https://github.com/Larens94/codedna/issues/new">issue</a> o una <a href="https://github.com/Larens94/codedna/compare">pull request</a> di fix. Quel feedback fa parte della challenge.',
     statusTitle:'Stato challenge',
     statusLead:'Classifica pubblica aggiornata quando arrivano PR valide. File sorgente: docs/challenge-board.json.',
@@ -467,10 +473,10 @@ <h2 data-i18n="boardTitle">Classifica pubblica</h2>
     metaSubs:'Submission', metaStatus:'Stato',
     metaEnroll:'Iscrizione', metaEnrollVal:'PR = dentro', metaEnrollSub:'niente signup ufficiale',
     howTitle:'Come funziona',
-    howLead:'Progetto reale funzionante (≥25 file) · dichiara linguaggi/framework/agente · ≥10 stessi task con vs senza CodeDNA · pubblica install · segnala bug · PR metrics.json.',
+    howLead:'Metodologia uguale per tutti: progetto reale · due branch/checkout/progetti · ≥10 stessi task con vs senza CodeDNA · copia prompt per l’agente · pubblica install · segnala bug.',
     s1t:'Progetto reale + stack', s1p:'Niente siti giocattolo. Dichiara linguaggi, framework, numero file, size band e come hai installato CodeDNA.',
-    s2t:'Stessi task, stack equo', s2p:'Esegui gli stessi ≥10 task con e senza CodeDNA. Tieni la parity L1/L2 salvo modalità codedna-only dichiarata.',
-    s3t:'PR → bacheca live', s3p:'Apri una PR metrics per iscriverti. La classifica pubblica si aggiorna quando arrivano PR valide. Possiamo chiedere una presentazione Meet.',
+    s2t:'Due branch o due progetti', s2p:'Stessi ≥10 task su Control e CodeDNA. Separa con due branch, due checkout o due progetti gemelli — dichiara setup.layout.',
+    s3t:'Copia prompt → PR', s3p:'Copia il prompt per il tuo agente, esegui la lista attività, apri la PR metrics. La classifica si aggiorna quando arrivano PR valide.',
     l0:'Header in-source — layer sotto test.',
     l1:'Wiki LLM, skill pack, instruction file — parity obbligatoria.',
     l2:'Memoria graph / strutturale — parity obbligatoria.',
@@ -590,8 +596,29 @@ <h2 data-i18n="boardTitle">Classifica pubblica</h2>
 document.getElementById('filter-favors').addEventListener('change', () => boardData && renderBoard(boardData));
 document.getElementById('filter-size').addEventListener('change', () => boardData && renderBoard(boardData));
 
+
+async function copyAgentPrompt() {
+  const dict = I18N[currentLang];
+  const status = document.getElementById('copy-status');
+  const file = currentLang === 'it' ? 'challenge-agent-prompt.it.md' : 'challenge-agent-prompt.md';
+  try {
+    const res = await fetch(file, { cache: 'no-store' });
+    if (!res.ok) throw new Error(res.statusText);
+    const text = await res.text();
+    await navigator.clipboard.writeText(text);
+    status.hidden = false;
+    status.textContent = dict.ctaCopied;
+  } catch (e) {
+    status.hidden = false;
+    status.textContent = dict.ctaCopyFail;
+    window.open(file, '_blank');
+  }
+}
+document.getElementById('copy-agent-prompt').addEventListener('click', copyAgentPrompt);
+
 applyI18n();
 loadBoard();
+
 </script>
 </body>
 </html>
diff --git a/docs/challenge.it.md b/docs/challenge.it.md
index a98f80a..895d371 100644
--- a/docs/challenge.it.md
+++ b/docs/challenge.it.md
@@ -20,13 +20,14 @@ Questa challenge pone una sola domanda:
 
 1. Lavori sul **tuo progetto reale e funzionante** (non sui nostri vecchi benchmark — non un sito giocattolo).
 2. Dichiari lo **stack tecnologico** completo (linguaggi, framework, agente, percorso di install).
-3. Esegui **almeno 10 task** (mix facili / medi / difficili) — gli **stessi task due volte**: una **senza** CodeDNA e una **con** CodeDNA.
+3. Usi la **stessa metodologia per tutti**: **≥10 stessi task** due volte — senza CodeDNA e con CodeDNA — tramite **due branch**, **due checkout** o **due progetti gemelli**.
 4. Tieni lo stack AI **equo** (vedi Livelli sotto).
 5. Documenti **come hai installato CodeDNA**; segnali bug / apri PR di fix quando qualcosa si rompe (atteso — è ancora sperimentale).
-6. Apri una **Pull Request** con `metrics.json` — anche se CodeDNA risulta peggiore. **Quella PR è la tua iscrizione.**
-7. La **bacheca / classifica pubblica si aggiorna quando arrivano PR valide** e resta visibile sul sito docs.
-8. Il progetto può essere chiamato a una **review live / presentazione** (call) così verifichiamo che i test siano reali.
-9. Il premio si sblocca solo se si raggiunge il **minimo di partecipanti**.
+6. Opzionale: incolla il **[prompt agente](challenge-agent-prompt.it.md)** nel tuo agente AI così segue protocollo + lista attività.
+7. Apri una **Pull Request** con `metrics.json` — anche se CodeDNA risulta peggiore. **Quella PR è la tua iscrizione.**
+8. La **bacheca / classifica pubblica si aggiorna quando arrivano PR valide** e resta visibile sul sito docs.
+9. Il progetto può essere chiamato a una **review live / presentazione** (call) così verifichiamo che i test siano reali.
+10. Il premio si sblocca solo se si raggiunge il **minimo di partecipanti**.
 
 ---
 
@@ -126,15 +127,33 @@ Per ogni task registra: obiettivo, difficoltà, agente/tool, successo/fallimento
 
 ### 3. Due condizioni: senza vs con CodeDNA
 
-Per **ciascuno degli stessi task** esegui:
+**La metodologia è uguale per tutti.** Per **ciascuno degli stessi task** esegui:
 
 | Condizione | Setup |
 |---|---|
 | **A — Control** | Il tuo workflow AI normale **senza** annotation CodeDNA / senza affidarti agli header CodeDNA |
 | **B — CodeDNA** | Stesso workflow **con** CodeDNA installato e annotato (`codedna init` / header mantenuti) |
 
+#### Come separare A e B (scegline uno — dichiaralo in `metrics.json` → `setup.layout`)
+
+| Layout | `setup.layout` | Esempio |
+|---|---|---|
+| Due branch nello stesso repo | `two_branches` | `challenge/control` + `challenge/codedna` |
+| Due checkout / cartelle | `two_checkouts` | due clone locali dello stesso progetto |
+| Due progetti gemelli | `two_projects` | due repo/cartelle dallo stesso codice di partenza |
+
 Mantieni **agente, modello e layer superiori identici** tra A e B, tranne CodeDNA stesso (salvo modalità dichiarata — vedi sotto).
 
+### 3b. Copia per il tuo agente (consigliato)
+
+Incolla il prompt pronto nel tuo agente AI così riceve regole + scheletro attività:
+
+- Italiano: [`challenge-agent-prompt.it.md`](challenge-agent-prompt.it.md)
+- English: [`challenge-agent-prompt.md`](challenge-agent-prompt.md)
+- Checklist task: [`../challenge/TASKS_TEMPLATE.it.md`](../challenge/TASKS_TEMPLATE.it.md) · [EN](../challenge/TASKS_TEMPLATE.md)
+
+Sulla pagina pubblica: **Copia per il tuo agente** copia il prompt nella lingua corrente.
+
 ### 4. Stack equo — Livelli (critico)
 
 Se già usi layer extra per lo sviluppo AI, CodeDNA va testato **sopra lo stesso stack**, non al posto di quelle cose per sbaglio.
@@ -176,6 +195,7 @@ Compila anche:
 
 - `summary.favors`: `codedna` | `control` | `tie` | `inconclusive`
 - `install` — **obbligatorio**: agente + passi esatti di install/init (e se ha funzionato)
+- `setup.layout` — **obbligatorio**: `two_branches` | `two_checkouts` | `two_projects`
 - `bugs_reported` — array **obbligatorio** (vuoto se nessuno); se qualcosa si è rotto, apri issue o PR di fix e elencala qui
 
 I numeri grezzi possono favorire CodeDNA **oppure no**. L’onestà batte il tifo.  
diff --git a/docs/challenge.md b/docs/challenge.md
index 15e1eaa..a74765c 100644
--- a/docs/challenge.md
+++ b/docs/challenge.md
@@ -20,13 +20,14 @@ This challenge asks one question:
 
 1. Work on **your real, working project** (not our old benchmarks — not a toy site).
 2. Declare your full **tech stack** (languages, frameworks, agent, install path).
-3. Run **at least 10 tasks** (easy / medium / hard mix) — the **same tasks twice**: once **without** CodeDNA, once **with** CodeDNA.
+3. Use the **same methodology for everyone**: **≥10 same tasks** twice — without CodeDNA and with CodeDNA — via **two branches**, **two checkouts**, or **two twin projects**.
 4. Keep your AI stack **fair** (see Levels below).
 5. Document **how you installed CodeDNA**; report bugs / open fix PRs when something breaks (expected — still experimental).
-6. Open a **Pull Request** with `metrics.json` — even if CodeDNA looks worse. **That PR is your enrollment.**
-7. The **public board / ranking updates as valid PRs arrive** and stays visible on the docs site.
-8. Your project may be called to a **live review / presentation** (conference call) so we can verify the tests are real.
-9. Prize unlocks only if the **minimum participant count** is met.
+6. Optional: paste the **[agent prompt](challenge-agent-prompt.md)** into your AI agent so it follows the protocol + activity list.
+7. Open a **Pull Request** with `metrics.json` — even if CodeDNA looks worse. **That PR is your enrollment.**
+8. The **public board / ranking updates as valid PRs arrive** and stays visible on the docs site.
+9. Your project may be called to a **live review / presentation** (conference call) so we can verify the tests are real.
+10. Prize unlocks only if the **minimum participant count** is met.
 
 ---
 
@@ -126,15 +127,33 @@ Record for each task: goal, difficulty, agent/tool, success/fail, notes.
 
 ### 3. Two conditions: without vs with CodeDNA
 
-For **each of the same tasks**, run:
+**Methodology is the same for everyone.** For **each of the same tasks**, run:
 
 | Condition | Setup |
 |---|---|
 | **A — Control** | Your normal AI workflow **without** CodeDNA annotations / without relying on CodeDNA headers |
 | **B — CodeDNA** | Same workflow **with** CodeDNA installed + annotated (`codedna init` / maintained headers) |
 
+#### How to separate A and B (pick one — declare in `metrics.json` → `setup.layout`)
+
+| Layout | `setup.layout` | Example |
+|---|---|---|
+| Two branches in your repo | `two_branches` | `challenge/control` + `challenge/codedna` |
+| Two checkouts / folders | `two_checkouts` | two local clones of the same project |
+| Two twin projects | `two_projects` | two repos/folders from the same starting code |
+
 Keep the **agent, model, and higher stack layers identical** between A and B except for CodeDNA itself (unless you declare an explicit mode — see below).
 
+### 3b. Copy for your agent (recommended)
+
+Paste the ready-made prompt into your AI agent so it gets rules + activity skeleton:
+
+- English: [`challenge-agent-prompt.md`](challenge-agent-prompt.md)
+- Italiano: [`challenge-agent-prompt.it.md`](challenge-agent-prompt.it.md)
+- Task checklist: [`../challenge/TASKS_TEMPLATE.md`](../challenge/TASKS_TEMPLATE.md) · [IT](../challenge/TASKS_TEMPLATE.it.md)
+
+On the public page: **Copia per il tuo agente** copies the prompt for the current language.
+
 ### 4. Fair stack — Levels (critical)
 
 If you already use extra AI-dev layers, CodeDNA must be tested **on top of the same stack**, not instead of it by accident.
@@ -176,6 +195,7 @@ Also fill:
 
 - `summary.favors`: `codedna` | `control` | `tie` | `inconclusive`
 - `install` — **required**: agent + exact install/init steps (and whether it worked)
+- `setup.layout` — **required**: `two_branches` | `two_checkouts` | `two_projects`
 - `bugs_reported` — **required** array (empty if none); if something broke, file an issue or fix PR and list it here
 
 Raw numbers can favor CodeDNA **or not**. Honesty beats cheerleading.  

From d8d9282aca300981f6eeb1a67507bbd2d7e27cc2 Mon Sep 17 00:00:00 2001
From: Cursor Agent <cursoragent@cursor.com>
Date: Tue, 8 Sep 2026 21:03:38 +0000
Subject: [PATCH 11/13] docs: note agent prompt methodology in README IT and
 changelog

AI-Agent: composer
AI-Provider: cursor
AI-Session: s_20260908_challenge_rules
AI-Visited: README-it.md,CHANGELOG.md
AI-Message: IT README + changelog mention agent copy-pack

Co-authored-by: Fabrizio Corpora <Larens94@users.noreply.github.com>
---
 CHANGELOG.md | 2 +-
 README-it.md | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 22eec31..e494b2a 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -6,7 +6,7 @@ All notable changes to CodeDNA will be documented in this file.
 
 ### Added
 
-- **CodeDNA Challenge (€200)** — bilingual rules, required `metrics.json` (languages + frameworks + size/files ≥25 + `install` + `bugs_reported`), real-project bar (no toy sites), same-task control vs CodeDNA, Meet presentation verification, docs site page + live public board updated as valid PRs arrive. PR opens enrollment (no separate signup). Separate from historical SWE-bench benchmarks.
+- **CodeDNA Challenge (€200)** — bilingual rules, required `metrics.json` (languages + frameworks + size/files ≥25 + `install` + `setup.layout` + `bugs_reported`), real-project bar (no toy sites), same-task control vs CodeDNA via two branches/checkouts/twin projects, copy-paste agent prompt + task list, Meet presentation verification, docs site page + live public board (IT default) updated as valid PRs arrive. PR opens enrollment (no separate signup). Separate from historical SWE-bench benchmarks.
 
 ### Added
 
diff --git a/README-it.md b/README-it.md
index 77b4abb..9f0ba38 100644
--- a/README-it.md
+++ b/README-it.md
@@ -48,7 +48,7 @@ Nessuna infrastruttura. Nessuna pipeline di retrieval. Nessuna memoria esterna.
 
 ## CodeDNA Challenge — €200
 
-Valutazione pubblica sul **tuo** progetto (non SWE-bench). Progetto reale funzionante (≥25 file), stack tecnologico + install CodeDNA dichiarati, ≥10 **stessi** task con/senza CodeDNA, parity L0/L1/L2, PR con metriche — anche se vanno contro CodeDNA. I bug vanno segnalati (sperimentale tra agenti). Possibile presentazione Meet; la bacheca pubblica si aggiorna quando arrivano PR valide.
+Valutazione pubblica sul **tuo** progetto (non SWE-bench). Progetto reale funzionante (≥25 file), stack tecnologico + install CodeDNA dichiarati, stessa metodologia per tutti (due branch / checkout / progetti gemelli), ≥10 **stessi** task con/senza CodeDNA, [prompt da copiare per l’agente](docs/challenge-agent-prompt.it.md), PR con metriche — anche se vanno contro CodeDNA. I bug vanno segnalati (sperimentale tra agenti). Possibile presentazione Meet; la bacheca pubblica si aggiorna quando arrivano PR valide.
 
 - Classifica / bacheca: [larens94.github.io/codedna/challenge.html](https://larens94.github.io/codedna/challenge.html)
 - Regolamento: [`docs/challenge.it.md`](docs/challenge.it.md) (IT) · [`docs/challenge.md`](docs/challenge.md) (EN)

From 3829cbf99a49eb179b16289caaaa7c40c39bc729 Mon Sep 17 00:00:00 2001
From: Cursor Agent <cursoragent@cursor.com>
Date: Tue, 8 Sep 2026 21:13:32 +0000
Subject: [PATCH 12/13] feat(challenge): optional file metrics + judge-agent
 comparison

Keep files_expected / precision / recall / F1 optional when ground truth
is unknown. Add judge prompt and schema fields so entrants can compare
Control vs CodeDNA sessions with a separate judge agent after both runs.

AI-Agent: composer
AI-Provider: cursor
AI-Session: s_20260908_challenge_rules
AI-Visited: challenge/metrics.schema.json,docs/challenge.md,docs/challenge-judge-prompt.it.md,docs/challenge.html
AI-Message: File metrics optional; judge agent path added for post-run comparison

Co-authored-by: Fabrizio Corpora <Larens94@users.noreply.github.com>
---
 challenge/README.md                 |   4 +-
 challenge/SUBMISSION_TEMPLATE.it.md |   2 +
 challenge/SUBMISSION_TEMPLATE.md    |   2 +
 challenge/metrics.example.json      |  92 ++++++-
 challenge/metrics.schema.json       | 372 ++++++++++++++++++++++++++++
 docs/challenge-agent-prompt.it.md   |  14 +-
 docs/challenge-agent-prompt.md      |  14 +-
 docs/challenge-board.README.md      |   6 +
 docs/challenge-judge-prompt.it.md   |  81 ++++++
 docs/challenge-judge-prompt.md      |  81 ++++++
 docs/challenge.html                 |  14 +-
 docs/challenge.it.md                |  21 ++
 docs/challenge.md                   |  21 ++
 13 files changed, 710 insertions(+), 14 deletions(-)
 create mode 100644 docs/challenge-judge-prompt.it.md
 create mode 100644 docs/challenge-judge-prompt.md

diff --git a/challenge/README.md b/challenge/README.md
index f33f0c6..a617008 100644
--- a/challenge/README.md
+++ b/challenge/README.md
@@ -15,10 +15,12 @@ No official signup. Public ranking (updated as valid PRs arrive): [`docs/challen
 
 - Italiano: [`docs/challenge-agent-prompt.it.md`](../docs/challenge-agent-prompt.it.md)
 - English: [`docs/challenge-agent-prompt.md`](../docs/challenge-agent-prompt.md)
+- Judge (optional, after both sessions): [`docs/challenge-judge-prompt.it.md`](../docs/challenge-judge-prompt.it.md) · [EN](../docs/challenge-judge-prompt.md)
 - Task lists: [`TASKS_TEMPLATE.it.md`](./TASKS_TEMPLATE.it.md) · [`TASKS_TEMPLATE.md`](./TASKS_TEMPLATE.md)
 - Pointer: [`AGENT_PROMPT.md`](./AGENT_PROMPT.md)
 
-Methodology (same for everyone): same ≥10 tasks with vs without CodeDNA via `two_branches` / `two_checkouts` / `two_projects`.
+Methodology (same for everyone): same ≥10 tasks with vs without CodeDNA via `two_branches` / `two_checkouts` / `two_projects`.  
+File-localization fields (`files_expected`, F1, …) are **optional** — use when known; otherwise a judge agent can still compare sessions.
 
 **Metrics format**
 
diff --git a/challenge/SUBMISSION_TEMPLATE.it.md b/challenge/SUBMISSION_TEMPLATE.it.md
index 27a0962..b529cf7 100644
--- a/challenge/SUBMISSION_TEMPLATE.it.md
+++ b/challenge/SUBMISSION_TEMPLATE.it.md
@@ -58,6 +58,8 @@
 - [ ] ≥10 task con mix `difficulty` (easy ≥3, medium ≥3, hard ≥2)
 - [ ] Gli **stessi task** hanno risultati `control` e `codedna`
 - [ ] `summary.favors` impostato (`codedna` | `control` | `tie` | `inconclusive`)
+- [ ] (Opzionale) `files_expected` / file F1 solo se conosciuti — non inventati
+- [ ] (Opzionale) giudice usato: [`docs/challenge-judge-prompt.it.md`](../docs/challenge-judge-prompt.it.md) → `judge` / `tasks[].judge`
 - [ ] I risultati possono favorire **oppure** sfavorire CodeDNA (onestà OK)
 
 ### Opzionale
diff --git a/challenge/SUBMISSION_TEMPLATE.md b/challenge/SUBMISSION_TEMPLATE.md
index e89271a..8745274 100644
--- a/challenge/SUBMISSION_TEMPLATE.md
+++ b/challenge/SUBMISSION_TEMPLATE.md
@@ -58,6 +58,8 @@
 - [ ] ≥10 tasks with `difficulty` mix (easy ≥3, medium ≥3, hard ≥2)
 - [ ] **Same tasks** each have both `control` and `codedna` results
 - [ ] `summary.favors` set (`codedna` | `control` | `tie` | `inconclusive`)
+- [ ] (Optional) `files_expected` / file F1 only when known — never invented
+- [ ] (Optional) judge used: [`docs/challenge-judge-prompt.md`](../docs/challenge-judge-prompt.md) → `judge` / `tasks[].judge`
 - [ ] Results may favor **or** disfavor CodeDNA (honesty OK)
 
 ### Optional
diff --git a/challenge/metrics.example.json b/challenge/metrics.example.json
index 6da9603..7f3be9e 100644
--- a/challenge/metrics.example.json
+++ b/challenge/metrics.example.json
@@ -61,7 +61,30 @@
         "tool_calls": 12,
         "wrong_file_or_module": false,
         "human_interventions": 0,
-        "confidence_1_to_5": 4
+        "confidence_1_to_5": 4,
+        "outcome": "pass",
+        "files_opened": [
+          "apps/api/src/billing/formatAmount.ts",
+          "apps/api/src/billing/invoice.service.ts",
+          "apps/api/src/users/user.service.ts",
+          "apps/api/src/billing/invoice.service.spec.ts"
+        ],
+        "files_edited": [
+          "apps/api/src/billing/formatAmount.ts",
+          "apps/api/src/billing/invoice.service.ts"
+        ],
+        "files_missed": [
+          "apps/api/src/billing/invoice.service.spec.ts"
+        ],
+        "files_extra": [],
+        "file_precision": 1.0,
+        "file_recall": 0.667,
+        "file_f1": 0.8,
+        "first_relevant_file_turn": 3,
+        "opened_before_first_edit": 2,
+        "tests_green": false,
+        "rules_followed": true,
+        "human_rewrote_core": false
       },
       "codedna": {
         "passed": true,
@@ -70,9 +93,46 @@
         "tool_calls": 8,
         "wrong_file_or_module": false,
         "human_interventions": 0,
-        "confidence_1_to_5": 5
+        "confidence_1_to_5": 5,
+        "outcome": "pass",
+        "files_opened": [
+          "apps/api/src/billing/formatAmount.ts",
+          "apps/api/src/billing/invoice.service.ts",
+          "apps/api/src/billing/invoice.service.spec.ts"
+        ],
+        "files_edited": [
+          "apps/api/src/billing/formatAmount.ts",
+          "apps/api/src/billing/invoice.service.ts",
+          "apps/api/src/billing/invoice.service.spec.ts"
+        ],
+        "files_missed": [],
+        "files_extra": [],
+        "file_precision": 1.0,
+        "file_recall": 1.0,
+        "file_f1": 1.0,
+        "first_relevant_file_turn": 1,
+        "opened_before_first_edit": 1,
+        "tests_green": true,
+        "rules_followed": true,
+        "human_rewrote_core": false
       },
-      "notes": ""
+      "notes": "",
+      "files_expected": [
+        "apps/api/src/billing/formatAmount.ts",
+        "apps/api/src/billing/invoice.service.ts",
+        "apps/api/src/billing/invoice.service.spec.ts"
+      ],
+      "judge": {
+        "agent": "cursor",
+        "prompt_ref": "docs/challenge-judge-prompt.it.md",
+        "favors": "codedna",
+        "score_control_1_to_5": 3,
+        "score_codedna_1_to_5": 5,
+        "navigation_notes": "Control opened an unrelated user service file first.",
+        "correctness_notes": "Control missed updating the unit test.",
+        "files_assessment": "CodeDNA edited all three expected files; Control missed the spec.",
+        "raw_verdict": "CodeDNA clearer navigation and complete caller/test updates."
+      }
     },
     {
       "id": "T02",
@@ -312,7 +372,18 @@
     "control_total_turns": 212,
     "codedna_total_turns": 125,
     "favors": "codedna",
-    "one_line_verdict": "CodeDNA reduced wrong-file mistakes on hard tasks; still honest about one medium task needing a steer."
+    "one_line_verdict": "CodeDNA reduced wrong-file mistakes on hard tasks; still honest about one medium task needing a steer.",
+    "control_pass_rate": 0.7,
+    "codedna_pass_rate": 0.9,
+    "delta_pass_rate": 0.2,
+    "control_file_f1_avg": 0.8,
+    "codedna_file_f1_avg": 1.0,
+    "delta_file_f1": 0.2,
+    "control_wrong_file_rate": 0.3,
+    "codedna_wrong_file_rate": 0.0,
+    "control_total_interventions": 8,
+    "codedna_total_interventions": 2,
+    "favors_basis": "judge + pass_rate + file_f1"
   },
   "bugs_reported": [],
   "install": {
@@ -327,5 +398,18 @@
     "control_ref": "challenge/control",
     "codedna_ref": "challenge/codedna",
     "notes": "Same commit base; CodeDNA branch has codedna init annotations"
+  },
+  "judge": {
+    "used": true,
+    "agent": "cursor",
+    "model": "claude-sonnet",
+    "prompt_ref": "docs/challenge-judge-prompt.it.md",
+    "method": "per_task",
+    "favors": "codedna",
+    "tasks_judged": 10,
+    "codedna_wins": 7,
+    "control_wins": 1,
+    "ties": 2,
+    "notes": "Judge compared diffs/logs after both sessions; did not need files_expected on every task."
   }
 }
diff --git a/challenge/metrics.schema.json b/challenge/metrics.schema.json
index 2f588e1..8f9e3a3 100644
--- a/challenge/metrics.schema.json
+++ b/challenge/metrics.schema.json
@@ -251,6 +251,90 @@
           },
           "notes": {
             "type": "string"
+          },
+          "files_expected": {
+            "type": [
+              "array",
+              "null"
+            ],
+            "items": {
+              "type": "string"
+            },
+            "description": "Optional ground-truth files for this task. Omit if unknown before/during the run. Needed only to compute precision/recall/F1."
+          },
+          "judge": {
+            "type": [
+              "object",
+              "null"
+            ],
+            "additionalProperties": false,
+            "description": "Optional judge-agent comparison of Control vs CodeDNA for this task.",
+            "properties": {
+              "agent": {
+                "type": "string",
+                "description": "Judge agent/model used."
+              },
+              "prompt_ref": {
+                "type": [
+                  "string",
+                  "null"
+                ],
+                "description": "Link or path to the judge prompt used."
+              },
+              "favors": {
+                "type": "string",
+                "enum": [
+                  "codedna",
+                  "control",
+                  "tie",
+                  "inconclusive"
+                ]
+              },
+              "score_control_1_to_5": {
+                "type": [
+                  "integer",
+                  "null"
+                ],
+                "minimum": 1,
+                "maximum": 5
+              },
+              "score_codedna_1_to_5": {
+                "type": [
+                  "integer",
+                  "null"
+                ],
+                "minimum": 1,
+                "maximum": 5
+              },
+              "navigation_notes": {
+                "type": [
+                  "string",
+                  "null"
+                ]
+              },
+              "correctness_notes": {
+                "type": [
+                  "string",
+                  "null"
+                ]
+              },
+              "files_assessment": {
+                "type": [
+                  "string",
+                  "null"
+                ],
+                "description": "Judge notes on right/wrong/missed files even without a predeclared expected list."
+              },
+              "raw_verdict": {
+                "type": [
+                  "string",
+                  "null"
+                ]
+              }
+            },
+            "required": [
+              "favors"
+            ]
           }
         }
       }
@@ -324,6 +408,89 @@
         },
         "one_line_verdict": {
           "type": "string"
+        },
+        "control_pass_rate": {
+          "type": [
+            "number",
+            "null"
+          ],
+          "minimum": 0,
+          "maximum": 1
+        },
+        "codedna_pass_rate": {
+          "type": [
+            "number",
+            "null"
+          ],
+          "minimum": 0,
+          "maximum": 1
+        },
+        "delta_pass_rate": {
+          "type": [
+            "number",
+            "null"
+          ]
+        },
+        "control_file_f1_avg": {
+          "type": [
+            "number",
+            "null"
+          ],
+          "minimum": 0,
+          "maximum": 1,
+          "description": "Optional average file_f1 over tasks where measurable."
+        },
+        "codedna_file_f1_avg": {
+          "type": [
+            "number",
+            "null"
+          ],
+          "minimum": 0,
+          "maximum": 1
+        },
+        "delta_file_f1": {
+          "type": [
+            "number",
+            "null"
+          ]
+        },
+        "control_wrong_file_rate": {
+          "type": [
+            "number",
+            "null"
+          ],
+          "minimum": 0,
+          "maximum": 1,
+          "description": "Optional share of tasks with missed/extra or wrong_file_or_module."
+        },
+        "codedna_wrong_file_rate": {
+          "type": [
+            "number",
+            "null"
+          ],
+          "minimum": 0,
+          "maximum": 1
+        },
+        "control_total_interventions": {
+          "type": [
+            "integer",
+            "null"
+          ],
+          "minimum": 0
+        },
+        "codedna_total_interventions": {
+          "type": [
+            "integer",
+            "null"
+          ],
+          "minimum": 0
+        },
+        "favors_basis": {
+          "type": [
+            "string",
+            "null"
+          ],
+          "description": "Optional note: human / judge / pass_rate+file_f1 / etc."
         }
       }
     },
@@ -418,6 +585,90 @@
           "type": "string"
         }
       }
+    },
+    "judge": {
+      "type": [
+        "object",
+        "null"
+      ],
+      "additionalProperties": false,
+      "description": "Optional overall judge-agent review across sessions (Control vs CodeDNA).",
+      "properties": {
+        "used": {
+          "type": "boolean"
+        },
+        "agent": {
+          "type": "string"
+        },
+        "model": {
+          "type": [
+            "string",
+            "null"
+          ]
+        },
+        "prompt_ref": {
+          "type": [
+            "string",
+            "null"
+          ]
+        },
+        "method": {
+          "type": "string",
+          "enum": [
+            "per_task",
+            "batch",
+            "spot_check"
+          ],
+          "description": "How the judge reviewed the runs."
+        },
+        "favors": {
+          "type": "string",
+          "enum": [
+            "codedna",
+            "control",
+            "tie",
+            "inconclusive"
+          ]
+        },
+        "tasks_judged": {
+          "type": [
+            "integer",
+            "null"
+          ],
+          "minimum": 0
+        },
+        "codedna_wins": {
+          "type": [
+            "integer",
+            "null"
+          ],
+          "minimum": 0
+        },
+        "control_wins": {
+          "type": [
+            "integer",
+            "null"
+          ],
+          "minimum": 0
+        },
+        "ties": {
+          "type": [
+            "integer",
+            "null"
+          ],
+          "minimum": 0
+        },
+        "notes": {
+          "type": [
+            "string",
+            "null"
+          ]
+        }
+      },
+      "required": [
+        "used",
+        "favors"
+      ]
     }
   },
   "$defs": {
@@ -468,6 +719,127 @@
           ],
           "minimum": 1,
           "maximum": 5
+        },
+        "outcome": {
+          "description": "Optional finer outcome; if set, passed should match outcome==pass.",
+          "anyOf": [
+            {
+              "type": "string",
+              "enum": [
+                "pass",
+                "partial",
+                "fail",
+                "inconclusive"
+              ]
+            },
+            {
+              "type": "null"
+            }
+          ]
+        },
+        "files_opened": {
+          "type": [
+            "array",
+            "null"
+          ],
+          "items": {
+            "type": "string"
+          },
+          "description": "Optional. Files opened/read during the run."
+        },
+        "files_edited": {
+          "type": [
+            "array",
+            "null"
+          ],
+          "items": {
+            "type": "string"
+          },
+          "description": "Optional. Files actually modified."
+        },
+        "files_missed": {
+          "type": [
+            "array",
+            "null"
+          ],
+          "items": {
+            "type": "string"
+          },
+          "description": "Optional. Expected files not edited (only if files_expected known)."
+        },
+        "files_extra": {
+          "type": [
+            "array",
+            "null"
+          ],
+          "items": {
+            "type": "string"
+          },
+          "description": "Optional. Edited files not in expected (only if files_expected known)."
+        },
+        "file_precision": {
+          "type": [
+            "number",
+            "null"
+          ],
+          "minimum": 0,
+          "maximum": 1,
+          "description": "Optional. |edited \u2229 expected| / |edited|."
+        },
+        "file_recall": {
+          "type": [
+            "number",
+            "null"
+          ],
+          "minimum": 0,
+          "maximum": 1,
+          "description": "Optional. |edited \u2229 expected| / |expected|."
+        },
+        "file_f1": {
+          "type": [
+            "number",
+            "null"
+          ],
+          "minimum": 0,
+          "maximum": 1,
+          "description": "Optional. Harmonic mean of file_precision and file_recall."
+        },
+        "first_relevant_file_turn": {
+          "type": [
+            "integer",
+            "null"
+          ],
+          "minimum": 0,
+          "description": "Optional. Turn when the first relevant/expected file was opened."
+        },
+        "opened_before_first_edit": {
+          "type": [
+            "integer",
+            "null"
+          ],
+          "minimum": 0,
+          "description": "Optional. Files opened before the first useful edit."
+        },
+        "tests_green": {
+          "type": [
+            "boolean",
+            "null"
+          ],
+          "description": "Optional. Relevant tests green after the change."
+        },
+        "rules_followed": {
+          "type": [
+            "boolean",
+            "null"
+          ],
+          "description": "Optional. Applicable CodeDNA Rules: constraints followed."
+        },
+        "human_rewrote_core": {
+          "type": [
+            "boolean",
+            "null"
+          ],
+          "description": "Optional. Human had to rewrite core logic after the agent."
         }
       }
     }
diff --git a/docs/challenge-agent-prompt.it.md b/docs/challenge-agent-prompt.it.md
index 5eea746..fafaf3b 100644
--- a/docs/challenge-agent-prompt.it.md
+++ b/docs/challenge-agent-prompt.it.md
@@ -79,17 +79,24 @@ Usa questa checklist come scheletro (sostituisci i titoli con task **reali** del
 
 Per **ogni** task registra Control e CodeDNA: `passed`, `minutes`/`turns`/`tool_calls` se disponibili, `wrong_file_or_module`, `human_interventions`, note.
 
+**Opzionale (non obbligatorio):**
+
+- `files_expected` solo se li conosci; altrimenti ometti (niente precision/recall inventate)
+- `files_opened` / `files_edited` se riesci a traccarli
+- Dopo entrambe le sessioni: usa un **agente giudice** con [`challenge-judge-prompt.it.md`](challenge-judge-prompt.it.md) e salva `tasks[].judge` + eventuale `judge` top-level
+
 ## Deliverable
 
 1. Congela la lista task **prima** delle run cronometrate.
 2. Esegui tutti i task in **A**, poi gli **stessi** in **B** (o interleaved, ma stessi ID).
-3. Compila `challenge/<mio-github-handle>/metrics.json` da  
+3. (Consigliato) Fai giudicare le due sessioni da un agente separato col prompt giudice.
+4. Compila `challenge/<mio-github-handle>/metrics.json` da  
    https://github.com/Larens94/codedna/blob/main/challenge/metrics.example.json  
    (schema: `metrics.schema.json`).
-4. Apri PR su `Larens94/codedna` con titolo:  
+5. Apri PR su `Larens94/codedna` con titolo:  
    `challenge: <handle> — CodeDNA Challenge submission`  
    Checklist: `challenge/SUBMISSION_TEMPLATE.it.md`
-5. Onestà: niente risultati inventati. Possibile Meet di verifica.
+6. Onestà: niente risultati inventati. Possibile Meet di verifica.
 
 ## Cosa fare ora
 
@@ -97,3 +104,4 @@ Per **ogni** task registra Control e CodeDNA: `passed`, `minutes`/`turns`/`tool_
 2. Chiedimi quale setup uso: **due branch** / **due checkout** / **due progetti**.
 3. Prepara i comandi di install per la condizione B.
 4. Solo dopo conferma mia: esegui i task e compila le metriche.
+5. Dopo le run: proponi il passaggio giudice (prompt già pronto) se voglio confrontare le sessioni.
diff --git a/docs/challenge-agent-prompt.md b/docs/challenge-agent-prompt.md
index 575cb85..30ee114 100644
--- a/docs/challenge-agent-prompt.md
+++ b/docs/challenge-agent-prompt.md
@@ -79,17 +79,24 @@ Use this checklist as a skeleton (replace titles with **real** tasks from my pro
 
 For **each** task record Control and CodeDNA: `passed`, `minutes`/`turns`/`tool_calls` when available, `wrong_file_or_module`, `human_interventions`, notes.
 
+**Optional (not required):**
+
+- `files_expected` only if known; otherwise omit (do not invent precision/recall)
+- `files_opened` / `files_edited` if you can track them
+- After both sessions: run a **judge agent** with [`challenge-judge-prompt.md`](challenge-judge-prompt.md) and save `tasks[].judge` + optional top-level `judge`
+
 ## Deliverable
 
 1. Freeze the task list **before** timed runs.
 2. Run every task in **A**, then the **same** ones in **B** (or interleaved, but same IDs).
-3. Fill `challenge/<my-github-handle>/metrics.json` from  
+3. (Recommended) Have a separate agent judge both sessions with the judge prompt.
+4. Fill `challenge/<my-github-handle>/metrics.json` from  
    https://github.com/Larens94/codedna/blob/main/challenge/metrics.example.json  
    (schema: `metrics.schema.json`).
-4. Open a PR on `Larens94/codedna` titled:  
+5. Open a PR on `Larens94/codedna` titled:  
    `challenge: <handle> — CodeDNA Challenge submission`  
    Checklist: `challenge/SUBMISSION_TEMPLATE.md`
-5. Honesty: no invented results. A Meet verification may be requested.
+6. Honesty: no invented results. A Meet verification may be requested.
 
 ## What to do now
 
@@ -97,3 +104,4 @@ For **each** task record Control and CodeDNA: `passed`, `minutes`/`turns`/`tool_
 2. Ask which setup I will use: **two branches** / **two checkouts** / **two projects**.
 3. Prepare install commands for condition B.
 4. Only after I confirm: run the tasks and fill the metrics.
+5. After the runs: offer the judge pass (prompt ready) if I want a structured comparison.
diff --git a/docs/challenge-board.README.md b/docs/challenge-board.README.md
index 87a6d1e..be93b12 100644
--- a/docs/challenge-board.README.md
+++ b/docs/challenge-board.README.md
@@ -22,11 +22,17 @@ It reads [`docs/challenge-board.json`](../docs/challenge-board.json) (static —
   "tasks_total": 10,
   "control_passed": 7,
   "codedna_passed": 9,
+  "control_file_f1_avg": 0.72,
+  "codedna_file_f1_avg": 0.91,
+  "delta_file_f1": 0.19,
+  "judge_favors": "codedna",
   "favors": "codedna",
+  "favors_basis": "judge + pass_rate",
   "merged": true
 }
 ```
 
+Optional board fields (`control_file_f1_avg`, `codedna_file_f1_avg`, `delta_file_f1`, `judge_favors`, `favors_basis`) are shown when present — file lists / F1 are **not** required from entrants.
 3. Bump `updated_at` (ISO date).
 4. Commit + push `main` so GitHub Pages refreshes.
 
diff --git a/docs/challenge-judge-prompt.it.md b/docs/challenge-judge-prompt.it.md
new file mode 100644
index 0000000..f36dc0a
--- /dev/null
+++ b/docs/challenge-judge-prompt.it.md
@@ -0,0 +1,81 @@
+# Prompt giudice — CodeDNA Challenge
+
+> Usa questo prompt **dopo** le due sessioni (Control e CodeDNA), non durante.  
+> Serve a confrontare le run in modo strutturato.  
+> English: [challenge-judge-prompt.md](challenge-judge-prompt.md)
+
+---
+
+## Ruolo
+
+Sei un **agente giudice imparziale**. Confronti due esecuzioni dello **stesso task** (o dello stesso set di task):
+
+- **A — Control:** senza CodeDNA  
+- **B — CodeDNA:** con CodeDNA  
+
+Non inventare file o esiti. Se l’evidenza manca, marca `inconclusive`.
+
+## Input che ti fornisco
+
+Per ogni task:
+
+1. Titolo / descrizione del task  
+2. Diff / file modificati della sessione Control  
+3. Diff / file modificati della sessione CodeDNA  
+4. (Opzionale) log turni, tool calls, interventi umani  
+5. (Opzionale) `files_expected` — **solo se li conosco**; altrimenti valuta senza ground truth
+
+## Cosa valutare (in ordine)
+
+1. **Completezza** — ha risolto il task? (pass / partial / fail)  
+2. **File toccati** — ha modificato i posti giusti? Ha dimenticato caller/test? Ha toccato file irrilevanti?  
+3. **Navigazione** — ha girato a vuoto o è andato subito ai file rilevanti?  
+4. **Vincoli** — ha rispettato regole di dominio / `Rules:` se evidenti dal diff o dal contesto  
+5. **Intervento umano** — quanto ha dovuto correggere la persona  
+
+Se **non** c’è `files_expected`:
+
+- Non inventare precision/recall/F1  
+- Descrivi comunque `files_assessment` in linguaggio naturale (completo / incompleto / fuori target)  
+- Puoi confrontare le due liste `files_edited` tra loro (quale sembra più coerente col task)
+
+Se **c’è** `files_expected`:
+
+- Calcola o stima missed / extra / precision / recall / F1 per entrambe le condizioni
+
+## Output richiesto (per task)
+
+```json
+{
+  "task_id": "T01",
+  "favors": "codedna | control | tie | inconclusive",
+  "score_control_1_to_5": 1,
+  "score_codedna_1_to_5": 1,
+  "navigation_notes": "",
+  "correctness_notes": "",
+  "files_assessment": "",
+  "raw_verdict": ""
+}
+```
+
+## Output complessivo (opzionale)
+
+```json
+{
+  "used": true,
+  "method": "per_task | batch | spot_check",
+  "favors": "codedna | control | tie | inconclusive",
+  "tasks_judged": 10,
+  "codedna_wins": 0,
+  "control_wins": 0,
+  "ties": 0,
+  "notes": ""
+}
+```
+
+## Regole
+
+- Non favorire CodeDNA a prescindere: se Control è meglio, dillo.  
+- Non usare conoscenza esterna inventata sul repo.  
+- Se confronti solo un campione di task, usa `method: spot_check` e dichiaralo.  
+- Il giudice **non sostituisce** `metrics.json`: arricchisce `tasks[].judge` e/o `judge` top-level.
diff --git a/docs/challenge-judge-prompt.md b/docs/challenge-judge-prompt.md
new file mode 100644
index 0000000..ddb2123
--- /dev/null
+++ b/docs/challenge-judge-prompt.md
@@ -0,0 +1,81 @@
+# Judge prompt — CodeDNA Challenge
+
+> Use this prompt **after** both sessions (Control and CodeDNA), not during them.  
+> It structures a comparison of the two runs.  
+> Italiano: [challenge-judge-prompt.it.md](challenge-judge-prompt.it.md)
+
+---
+
+## Role
+
+You are an **impartial judge agent**. Compare two executions of the **same task** (or task set):
+
+- **A — Control:** without CodeDNA  
+- **B — CodeDNA:** with CodeDNA  
+
+Do not invent files or outcomes. If evidence is missing, mark `inconclusive`.
+
+## Inputs I will provide
+
+Per task:
+
+1. Task title / description  
+2. Diff / edited files from Control  
+3. Diff / edited files from CodeDNA  
+4. (Optional) turn logs, tool calls, human interventions  
+5. (Optional) `files_expected` — **only if known**; otherwise judge without ground truth
+
+## What to score (in order)
+
+1. **Completeness** — did it solve the task? (pass / partial / fail)  
+2. **Files touched** — right places? missed callers/tests? irrelevant files?  
+3. **Navigation** — wandered vs went to relevant files quickly?  
+4. **Constraints** — domain / `Rules:` respected when visible from diff/context  
+5. **Human steering** — how much did the person have to fix  
+
+If there is **no** `files_expected`:
+
+- Do **not** invent precision/recall/F1  
+- Still write a natural-language `files_assessment` (complete / incomplete / off-target)  
+- You may compare the two `files_edited` lists for coherence with the task
+
+If there **is** `files_expected`:
+
+- Compute or estimate missed / extra / precision / recall / F1 for both conditions
+
+## Required per-task output
+
+```json
+{
+  "task_id": "T01",
+  "favors": "codedna | control | tie | inconclusive",
+  "score_control_1_to_5": 1,
+  "score_codedna_1_to_5": 1,
+  "navigation_notes": "",
+  "correctness_notes": "",
+  "files_assessment": "",
+  "raw_verdict": ""
+}
+```
+
+## Optional overall output
+
+```json
+{
+  "used": true,
+  "method": "per_task | batch | spot_check",
+  "favors": "codedna | control | tie | inconclusive",
+  "tasks_judged": 10,
+  "codedna_wins": 0,
+  "control_wins": 0,
+  "ties": 0,
+  "notes": ""
+}
+```
+
+## Rules
+
+- Do not bias toward CodeDNA: if Control is better, say so.  
+- Do not invent external repo knowledge.  
+- If you only sample tasks, use `method: spot_check` and declare it.  
+- The judge **does not replace** `metrics.json`: it fills `tasks[].judge` and/or top-level `judge`.
diff --git a/docs/challenge.html b/docs/challenge.html
index 6cb754a..8f9b2ef 100644
--- a/docs/challenge.html
+++ b/docs/challenge.html
@@ -457,7 +457,7 @@ <h2 data-i18n="boardTitle">Classifica pubblica</h2>
     emptyBody:'This is the live public leaderboard. As soon as valid metrics PRs arrive and are reviewed, rows appear here.',
     boardNote:'Live ranking, curated by maintainers. Opening a PR enrolls you; appearing here means the submission format and stack declaration were accepted. Agent × language gaps are expected — file issues or fix PRs.',
     thHandle:'Entrant', thStack:'Language / framework', thSize:'Size', thFiles:'Files',
-    thMode:'Mode', thScore:'Pass C / DNA', thFavors:'Favors', thPr:'PR',
+    thMode:'Mode', thScore:'Pass C / DNA', thF1:'File-F1 C / DNA', thFavors:'Favors', thJudge:'Judge', thPr:'PR',
     unlock:'min {n} to unlock', windowFallback:'coming soon'
   },
   it: {
@@ -487,7 +487,7 @@ <h2 data-i18n="boardTitle">Classifica pubblica</h2>
     emptyBody:'Questa è la leaderboard pubblica live. Appena arrivano PR metrics valide e vengono reviewate, le righe compaiono qui.',
     boardNote:'Classifica live, curata dai maintainer. Aprire una PR ti iscrive; comparire qui significa che formato e dichiarazione di stack sono stati accettati. Gap agente × linguaggio sono attesi — apri issue o PR di fix.',
     thHandle:'Partecipante', thStack:'Linguaggio / framework', thSize:'Size', thFiles:'File',
-    thMode:'Mode', thScore:'Pass C / DNA', thFavors:'Favors', thPr:'PR',
+    thMode:'Mode', thScore:'Pass C / DNA', thF1:'File-F1 C / DNA', thFavors:'Favors', thJudge:'Giudice', thPr:'PR',
     unlock:'min {n} per sbloccare', windowFallback:'in arrivo'
   }
 };
@@ -556,6 +556,12 @@ <h2 data-i18n="boardTitle">Classifica pubblica</h2>
     const langs = (s.languages || []).map(x => `<span class="pill">${escapeHtml(x)}</span>`).join('');
     const fws = (s.frameworks || []).map(x => `<span class="pill">${escapeHtml(x)}</span>`).join('');
     const pr = s.pr_url ? `<a href="${escapeAttr(s.pr_url)}" target="_blank" rel="noopener">#${escapeHtml(String(s.pr_number || 'PR'))}</a>` : '—';
+    const f1c = s.control_file_f1_avg;
+    const f1d = s.codedna_file_f1_avg;
+    const f1 = (f1c != null || f1d != null)
+      ? `${f1c != null ? Number(f1c).toFixed(2) : '—'} / ${f1d != null ? Number(f1d).toFixed(2) : '—'}`
+      : '—';
+    const judge = s.judge_favors || '—';
     return `<tr>
       <td><strong>${escapeHtml(s.handle || '—')}</strong></td>
       <td>${langs}<div style="margin-top:4px">${fws || '<span class="pill">—</span>'}</div></td>
@@ -563,7 +569,9 @@ <h2 data-i18n="boardTitle">Classifica pubblica</h2>
       <td class="mono">${s.approx_source_files != null ? escapeHtml(String(s.approx_source_files)) : '—'}</td>
       <td class="mono">${escapeHtml(s.mode || '—')}</td>
       <td class="mono">${escapeHtml(String(s.control_passed ?? '—'))} / ${escapeHtml(String(s.codedna_passed ?? '—'))}</td>
+      <td class="mono">${escapeHtml(f1)}</td>
       <td class="${favorsClass(s.favors)}">${escapeHtml(s.favors || '—')}</td>
+      <td class="${favorsClass(s.judge_favors)}">${escapeHtml(String(judge))}</td>
       <td>${pr}</td>
     </tr>`;
   }).join('');
@@ -571,7 +579,7 @@ <h2 data-i18n="boardTitle">Classifica pubblica</h2>
   root.innerHTML = `<div class="table-wrap"><table class="board">
     <thead><tr>
       <th>${dict.thHandle}</th><th>${dict.thStack}</th><th>${dict.thSize}</th><th>${dict.thFiles}</th>
-      <th>${dict.thMode}</th><th>${dict.thScore}</th><th>${dict.thFavors}</th><th>${dict.thPr}</th>
+      <th>${dict.thMode}</th><th>${dict.thScore}</th><th>${dict.thF1}</th><th>${dict.thFavors}</th><th>${dict.thJudge}</th><th>${dict.thPr}</th>
     </tr></thead>
     <tbody>${body}</tbody>
   </table></div>`;
diff --git a/docs/challenge.it.md b/docs/challenge.it.md
index 895d371..04b4335 100644
--- a/docs/challenge.it.md
+++ b/docs/challenge.it.md
@@ -191,16 +191,37 @@ Obbligatorio per ogni task (Control + CodeDNA):
 | `human_interventions` | Quante volte hai dovuto guidare |
 | `confidence_1_to_5` | Opzionale ma utile |
 
+Opzionali (consigliati quando disponibili — **non obbligatori**):
+
+| Campo | Note |
+|---|---|
+| `files_expected` (a livello task) | Solo se conosci/congeli i file; ometti se sconosciuti |
+| `files_opened` / `files_edited` | Impronta di navigazione e edit |
+| `files_missed` / `files_extra` / `file_precision` / `file_recall` / `file_f1` | Solo se esiste `files_expected` |
+| `first_relevant_file_turn` / `opened_before_first_edit` | Efficienza di navigazione |
+| `outcome` / `tests_green` / `rules_followed` / `human_rewrote_core` | Qualità più fine |
+| `tasks[].judge` + `judge` top-level | Opzionale **agente giudice** che confronta Control vs CodeDNA dopo entrambe le sessioni — vedi [`challenge-judge-prompt.it.md`](challenge-judge-prompt.it.md) |
+
 Compila anche:
 
 - `summary.favors`: `codedna` | `control` | `tie` | `inconclusive`
 - `install` — **obbligatorio**: agente + passi esatti di install/init (e se ha funzionato)
 - `setup.layout` — **obbligatorio**: `two_branches` | `two_checkouts` | `two_projects`
 - `bugs_reported` — array **obbligatorio** (vuoto se nessuno); se qualcosa si è rotto, apri issue o PR di fix e elencala qui
+- Extra summary opzionali: `*_file_f1_avg`, `delta_file_f1`, `*_wrong_file_rate`, `favors_basis` (`human` / `judge` / misto)
 
 I numeri grezzi possono favorire CodeDNA **oppure no**. L’onestà batte il tifo.  
 `notes.md` è narrativa opzionale — **non sostituisce** `metrics.json`.
 
+### 5b. Agente giudice opzionale (consigliato)
+
+**Non** serve avere `files_expected` per confrontare le run. Dopo entrambe le sessioni puoi incollare il [prompt giudice](challenge-judge-prompt.it.md) in un agente separato con i due diff/log e registrare:
+
+- per task `tasks[].judge` (favors, score, assessment file in linguaggio naturale)
+- opzionale summary top-level `judge` (`method`: `per_task` | `batch` | `spot_check`)
+
+Il giudice arricchisce la storia in classifica; non sostituisce i campi obbligatori.
+
 ### 6. Percorso di install + bug report (obbligatori quando rilevanti)
 
 Pubblica **come hai installato CodeDNA** in `metrics.json` → `install` (comandi, flag `--tools`, agente).  
diff --git a/docs/challenge.md b/docs/challenge.md
index a74765c..611bfd0 100644
--- a/docs/challenge.md
+++ b/docs/challenge.md
@@ -191,16 +191,37 @@ Required per task (Control + CodeDNA):
 | `human_interventions` | How often you had to steer |
 | `confidence_1_to_5` | Optional but useful |
 
+Optional (recommended when available — **not required**):
+
+| Field | Notes |
+|---|---|
+| `files_expected` (task-level) | Only if you know/ground the files beforehand or after; omit if unknown |
+| `files_opened` / `files_edited` | Navigation & edit footprint |
+| `files_missed` / `files_extra` / `file_precision` / `file_recall` / `file_f1` | Only meaningful when `files_expected` exists |
+| `first_relevant_file_turn` / `opened_before_first_edit` | Navigation efficiency |
+| `outcome` / `tests_green` / `rules_followed` / `human_rewrote_core` | Finer quality signals |
+| `tasks[].judge` + top-level `judge` | Optional **judge agent** comparing Control vs CodeDNA after both sessions — see [`challenge-judge-prompt.md`](challenge-judge-prompt.md) |
+
 Also fill:
 
 - `summary.favors`: `codedna` | `control` | `tie` | `inconclusive`
 - `install` — **required**: agent + exact install/init steps (and whether it worked)
 - `setup.layout` — **required**: `two_branches` | `two_checkouts` | `two_projects`
 - `bugs_reported` — **required** array (empty if none); if something broke, file an issue or fix PR and list it here
+- Optional summary extras: `*_file_f1_avg`, `delta_file_f1`, `*_wrong_file_rate`, `favors_basis` (`human` / `judge` / mixed)
 
 Raw numbers can favor CodeDNA **or not**. Honesty beats cheerleading.  
 Markdown notes (`notes.md`) are optional narrative — **do not replace** `metrics.json`.
 
+### 5b. Optional judge agent (recommended)
+
+You do **not** need `files_expected` to compare runs. After both sessions you may paste the [judge prompt](challenge-judge-prompt.md) into a separate agent with the two diffs/logs and record:
+
+- per-task `tasks[].judge` (favors, scores, file assessment in natural language)
+- optional top-level `judge` summary (`method`: `per_task` | `batch` | `spot_check`)
+
+The judge enriches the ranking story; it does not replace required fields.
+
 ### 6. Install path + bug reports (mandatory when relevant)
 
 Publish **how you installed CodeDNA** in `metrics.json` → `install` (commands, `--tools` flag, agent).  

From 574b705c9a66a19c78786f78249adc62e7b05629 Mon Sep 17 00:00:00 2001
From: Cursor Agent <cursoragent@cursor.com>
Date: Tue, 8 Sep 2026 21:13:44 +0000
Subject: [PATCH 13/13] docs: changelog note for optional file metrics and
 judge

AI-Agent: composer
AI-Provider: cursor
AI-Session: s_20260908_challenge_rules
AI-Visited: CHANGELOG.md
AI-Message: Changelog mentions optional F1 + judge agent

Co-authored-by: Fabrizio Corpora <Larens94@users.noreply.github.com>
---
 CHANGELOG.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index e494b2a..fe6ba70 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -6,7 +6,7 @@ All notable changes to CodeDNA will be documented in this file.
 
 ### Added
 
-- **CodeDNA Challenge (€200)** — bilingual rules, required `metrics.json` (languages + frameworks + size/files ≥25 + `install` + `setup.layout` + `bugs_reported`), real-project bar (no toy sites), same-task control vs CodeDNA via two branches/checkouts/twin projects, copy-paste agent prompt + task list, Meet presentation verification, docs site page + live public board (IT default) updated as valid PRs arrive. PR opens enrollment (no separate signup). Separate from historical SWE-bench benchmarks.
+- **CodeDNA Challenge (€200)** — bilingual rules, required `metrics.json` (languages + frameworks + size/files ≥25 + `install` + `setup.layout` + `bugs_reported`), optional file-localization metrics + optional judge-agent comparison, real-project bar (no toy sites), same-task control vs CodeDNA via two branches/checkouts/twin projects, copy-paste agent/judge prompts + task list, Meet presentation verification, docs site page + live public board (IT default) updated as valid PRs arrive. PR opens enrollment (no separate signup). Separate from historical SWE-bench benchmarks.
 
 ### Added