Skip to content

brief_patch.md

Latest commit

 

History

History
58 lines (48 loc) · 3.63 KB

brief_patch.md

File metadata and controls

58 lines (48 loc) · 3.63 KB
Raw

Patch brief: fold a gap-round slot into the built graphs

Paths are relative to the guideline folder ai_tooling/dev_guidelines/javascript_guidelines/ unless stated. Your task message gives the absolute root and the slot to fold in.

Why a patch, not a re-run

The stage-7 fixes edited the integrated files (build/distill/skeleton.md, build/distill/claims_ledger.md, build/distill/decisions_for_review.md) directly. Never run build/distill/5c_integrate/integrate.py: it rebuilds those files from the branch folders and would silently undo the review fixes. The integrated files are now the source of truth; the branch folders are history.

Read first

  • build/distill/brief_distill.md, section Shared (rules, ledger format, epistemic states, the max rule) and Step 5b (what a node, a judge record and a decision look like).
  • build/distill/target_hierarchy.md (tree, max rule, judge record).
  • build/review/fixes/result.md, ## L2 claims (the conventions the fixes introduced: unvalidated flags, rules-only constraints, bookkeeping in build/assemble/bookkeeping_log.md) and build/assemble/result.md, ## L2 claims (how the graphs are built and checked).
  • The slot's claim table: build/research/<slot>/result.md, all levels.
  • The nodes you touch, read through ig-tool query -g intent_graphs/js_defaults.ig.yaml --node <UID> (and the philosophy graph), and their rows in skeleton.md and claims_ledger.md.

What to do

  1. Place every claim of the slot exactly once in claims_ledger.md (ledger format of the brief), on the node that owns it: an existing node, a node the slot resolves (an undecided node becomes decided when the slot grounds a position), or a new node. New nodes follow the UID rules, two to six children per parent; add them to skeleton.md with intent, epistemic state and ledger rows, and a judge record when they hold a decision.
  2. Decide by the max rule and record each new decision, and each undecided node now decided, in decisions_for_review.md under a new section named after the slot.
  3. Keep the fix conventions: constraints are rules only; facts go to design notes; single-source, inferred or open grounds carry unvalidated: <why>; no bookkeeping in engineer-facing fields; JavaScript and tool names stay out of the philosophy graph.
  4. Links: add intentlinks the new nodes need (inside and across branches, and to js-philosophy), in the skeleton's link tables.
  5. Rebuild and gate, from the root: python3 build/assemble/apply_flags.py, python3 build/assemble/build_batches.py, zsh build/assemble/build_graphs.sh, python3 build/assemble/check_complete.py (0 failures), python3 build/distill/check_ledger.py (0 errors), and ig-tool validate -g <graph> -m intent_graphs/intent_resources.yaml for both graphs (0 errors).
  6. README: update counts in intent_graphs/README.md and remove from "Not covered" what the slot now covers; keep what it does not.

Result file

Write build/patch/<slot>/result.md with ## L1 headline, ## L2 claims, ## L3 full, each readable alone, items marked grounded, inferred or open. Pin, word for word at every level: claims placed (0 dropped); nodes added and nodes resolved from undecided; decisions added; validate summaries of both graphs; the outputs of check_ledger.py and check_complete.py; what remains under "Not covered".

Write only build/distill/ (the three integrated files), build/assemble/ (logs, if needed), intent_graphs/ (rebuilt graphs and README) and build/patch/<slot>/. Public repository: no absolute or home paths, no private names. Never commit, never git push.