Refresh public release surface

README.md

@@ -140,7 +140,7 @@ The repository has three audiences:
    under severe memory, storage, and CPU limits.
 2. **AI practitioners** who want a readable, inspectable transformer runtime
    without the usual stack of frameworks and accelerators.
-3. **Retro and embedded-system builders** who want a working assistant runtime,
+3. **Legacy and embedded-system builders** who want a working assistant runtime,
    not just a conceptual port.
 
 The important result is not that this tiny model competes with modern hosted

docs/hardware-validation.md

@@ -152,7 +152,7 @@ mismatches so evidence filenames cannot drift away from the machine notes.
 Start with whatever 486-class hardware is easiest to borrow, buy, or assemble.
 A Pentium can wait until after the 486 baseline is captured. Useful sources are:
 
-- local retro-computing groups or repair shops
+- local legacy-computing groups or repair shops
 - eBay or marketplace listings for tested 486 boards or complete systems
 - university surplus or electronics recyclers
 - friends with old industrial PCs, DOS tooling rigs, or ISA test benches

docs/marketing/promo-kit.md

@@ -64,7 +64,7 @@ for physical DOS systems.
 
 Portable machine intelligence in BASIC: a QEMU-verified DOS/486 transformer and
 assistant runtime with fixed-point GPT inference, hot-swappable local model
-packs, sharded recall indexes, and release bundles for retro and constrained
+packs, sharded recall indexes, and release bundles for legacy and constrained
 systems.
 
 ## Release Announcement Draft
@@ -124,11 +124,11 @@ video tooling are not presented as DOS or Windows 95 runtime features.
 
 ## Thumbnail Text Options
 
-- GPT IN DOS
-- AI ON A 486?
+- LOCAL AI IN DOS
+- 486-CLASS TARGET
 - TRANSFORMERS IN BASIC
 - DOS CHAT MODEL
-- REAL INFERENCE, RETRO HARDWARE
+- REAL INFERENCE, LOCAL HARDWARE
 
 ## Press/Contact Placeholder
 

docs/marketing/video-plan.md

@@ -41,8 +41,9 @@ Target length: 4 to 6 minutes.
 
 5. Evidence segment, 45 seconds
    - Show `qemu/evidence/quality_report_assistant_chat.md`.
-   - Show `48/48`, average `1.000`.
-   - Show model report status OK.
+   - Show CHAT quality `160/160`, average `0.999`.
+   - Show raw/generalist/consistency gates: `83/83`, `24/24`, and `498/498`.
+   - Show assistant stress `50/50` across five packs and model report status OK.
    - Explain QEMU evidence versus pending physical-hardware timing.
 
 6. Release/package segment, 30 seconds
@@ -178,7 +179,7 @@ Use precise language:
 - Say "GPT-style" instead of "GPT-2 scale".
 - Say "QEMU 486 path" unless physical hardware is on screen.
 - Say "small model" and "real fixed-point inference".
-- Say "assistant pack" for CHAT, DOSHELP, and OFFICE.
+- Say "assistant pack" for CHAT, DOSHELP, OFFICE, DEV, and PORTABLE.
 
 Do not hide the latency. The wait is part of the story.
 

docs/public-launch.md

@@ -20,8 +20,12 @@ The CHAT assistant pack currently has:
 
 - Model profile: `486dx2-usable`
 - Shape: `3L 64D 4H ctx192 hidden256 vocab4096`
-- Fixed-point quality gate: `48/48`, average `1.000`
-- Two-sentence CHAT outputs in the generated quality report
+- Fixed-point CHAT pack quality gate: `160/160`, average `0.999`
+- Raw direct prompt gate: `83/83`
+- Generalist conversational prompt gate: `24/24`
+- Consistency gate: `498/498` variants across `83/83` prompt groups
+- Pack retrieval and KDB/KB2 recall gates: `42/42`
+- QEMU assistant stress gate: `50/50` replies across five packs
 - Training performed on Apple Metal/MPS, fixed-point evaluation on CPU
 - Runtime artifacts validated by `scripts/model_report.py --strict`
 

qemu/evidence/assistant_showcase_video.md

@@ -9,14 +9,14 @@ The visible DOS session uses era-accurate DOS commands only; Python, ffmpeg, and
 - Cast: `promo/renders/gpt2_basic_assistant_showcase.cast`
 - GIF intermediate: `promo/renders/gpt2_basic_assistant_showcase.gif`
 - Resolution: `1920x1080`
-- Duration seconds: `173.200000`
+- Duration seconds: `168.566667`
 - Final frame hold seconds: `12`
-- MP4 SHA-256: `54af5236f373e690ea5bbf7a557a3ae2926f4b68033454d29a58ac65bd93ff90`
+- MP4 SHA-256: `e29292f29c5c1ce771014dd85e400aa67e01ad56164d3b4e2d87edafe4b1e362`
 
 Audience:
 
 - Engineers evaluating local language models on constrained systems.
-- Retrocomputing, embedded, industrial, archival, and air-gapped operators.
+- Legacy, embedded, industrial, archival, and air-gapped operators.
 - Pack authors who need fast local recall, small weights, and auditable behavior.
 
 Covered functionality:

qemu/evidence/preview_release_manifest.md

@@ -5,7 +5,7 @@ Generated: `2026-05-12`
 Package tree: `gpt2-basic-preview`
 Package zip: `gpt2-basic-preview.zip`
 Package checksums: `SHA256SUMS.txt`; zip sidecar: `gpt2-basic-preview.zip.sha256`
-Package status: `705 files, 120,009,447 bytes`
+Package status: `705 files, 120,011,903 bytes`
 
 This is an iterative preview payload. It ships only strict-quality release models and assistant packs; rejected repair attempts and old candidates remain repo evidence only.
 

scripts/build_assistant_showcase_video.py

@@ -199,7 +199,7 @@ def write_report(report: Path, paths: dict[str, Path], resolution: str, duration
         "Audience:",
         "",
         "- Engineers evaluating local language models on constrained systems.",
-        "- Retrocomputing, embedded, industrial, archival, and air-gapped operators.",
+        "- Legacy, embedded, industrial, archival, and air-gapped operators.",
         "- Pack authors who need fast local recall, small weights, and auditable behavior.",
         "",
         "Covered functionality:",

scripts/play_assistant_showcase_terminal.py

@@ -176,7 +176,7 @@ def play_showcase(stress_report: Path, capability_report: Path, speed: float) ->
 
     term.title("Who This Is For", "Product demonstration, not personal branding")
     term.line("For engineers evaluating useful local language models on constrained systems.", WHITE, 0.09)
-    term.line("For retrocomputing, embedded, industrial, archival, and air-gapped environments.", WHITE, 0.09)
+    term.line("For legacy, embedded, industrial, archival, and air-gapped environments.", WHITE, 0.09)
     term.line("For pack authors who need fast local recall, small weights, and auditable behavior.", WHITE, 0.09)
     term.line("This is not a cloud demo or a personal release: it shows what the product can do.", YELLOW, 0.09)
     term.sleep(1.2)

tests/test_public_repo_hygiene.py

@@ -18,6 +18,9 @@
     ROOT / "gpt2_basic_documentation.md",
     ROOT / "docs" / "marketing" / "promo-kit.md",
     ROOT / "docs" / "marketing" / "video-plan.md",
+    ROOT / "qemu" / "evidence" / "assistant_showcase_video.md",
+    ROOT / "scripts" / "build_assistant_showcase_video.py",
+    ROOT / "scripts" / "play_assistant_showcase_terminal.py",
 )