chore(scripts): dashboard regen extracts PR# from idea-only legacy folders

[SoundMindsAI]

Summary

Extends _extract_pr_number in the dashboard regen script with two new cascade priorities (3.5 + 3.6) that mine idea.md content for own-PR assertions. This closes the data gap where ~50 early MVP1 implemented features with only idea.md artifacts (no spec/plan/pipeline_status) render as Complete instead of [PR #N](url) merged YYYY-MM-DD in the regenerated dashboard.

• Story 1.1 — adds 5 module-level re.compile'd patterns (3 line-anchored strict patterns + **PR:** frontmatter pattern + metadata-key matcher) and a new _extract_metadata_block helper that bounds the **PR:** search to the idea's metadata cluster (contiguous metadata-key lines stopping at the first ## heading or non-metadata non-blank line; 30-line cap for headingless edge cases). Each pattern's \b boundary sits immediately after the digit capture per GPT-5.5 cycle-2 fix.
• Story 1.2 — threads idea.md content through _load_implemented + _load_planned via the new 4th arg; adds 27-case test file at backend/tests/unit/scripts/test_dashboard_pr_extraction.py; documents the **PR:** frontmatter convention in architecture.md's new "Dashboard regen" subsection.

Pipeline artifacts

• idea.md — preflighted 2026-05-23 (line-drift fix 476→499, dependency PR# refreshed to PR fix(dashboard): time-order DEPS_ALL_BACKEND expansion so shipped features don't inherit post-shipment peers #208 shipped, "⚠️ Audit finding" block flagging regex-viability issue, 3 open questions with locked defaults)
• feature_spec.md — GPT-5.5 cross-model review converged at cycle 3: 12 findings total (2 H / 7 M / 3 L) across 3 cycles, all 11 actionable accepted and applied; 17 ACs locked
• implementation_plan.md — GPT-5.5 cycle 1: 3 findings (0 H / 1 M / 2 L), all accepted (AC-12 ownership clarified, _extract_metadata_block title-once flag added, pytest collection grep -q fix)

Test coverage

Layer	Files	Tests
Unit	backend/tests/unit/scripts/test_dashboard_pr_extraction.py (new)	27 cases covering all 17 ACs + mutual-exclusion + helper unit cases + regex-constant locks
Integration	none	— (script is hermetic; no DB/network)
Contract	none	— (no API surface)
E2E	none	— (no UI surface)

Existing unit suite remains green: 1190/1190 (1163 baseline + 27 new).

Empirical verification (per spec §14 step 3)

Run make dashboard and compare shipped/planned table PR-number columns:

Forward gain: 3 rows transitioned None → PR#.
  • chore_create_study_modal_e2e_stability → PR #161 (Pattern B precedent)
  • chore_precommit_node_path_resolution → PR #171 (Pattern C precedent)
  • chore_dashboard_pr_extraction_from_idea → PR #4 ← cosmetic anomaly (see note)

Regressions: 0 rows whose PR# changed.

PASS: zero regressions; forward gain documented above.

Forward gain (3) is below the spec's predicted 5–8. During the preflight survey I counted 5–8 candidate features carrying the canonical strict shapes. The regen actually picks up only 3 because the other 2 of the candidates (chore_data_table_columnvisibility_tanstack, feat_contextual_help_mvp2) have their PR# captured earlier in the cascade — by priority 3's fuzzy merged-context match against narrative content in their idea bodies. They were already rendering correctly; the new priority 3.5 doesn't get a chance to fire. Net effect on operator value: the chore still does what it set out to do (legacy idea-only folders now surface their PRs), just slightly less of a forward gain than predicted because the existing priority-3 was doing more work than the preflight assumed.

Cosmetic anomaly for THIS chore's own planned-feature row: the chore appears in the dashboard's Plan section as [PR #4] merged 2026-05-15, which is wrong (PR #4 is infra_foundation, the date is from an AC-1 example body). Root cause: the chore's own spec content contains AC-body examples that quote OTHER features' PR-merge phrases (e.g., AC-4's body **Depends on:** [infra_foundation] — merged via PR #4), which the pre-existing priority-3 fuzzy regex grabs. The anomaly is cosmetic only — resolves automatically when this chore merges and _load_implemented takes over the row rendering. Captured as a separate idea: chore_dashboard_regen_quoted_pr_false_positive for future hardening of the priority-3 fuzzy regex.

Test plan

• make backend-fmt / make backend-lint / make backend-typecheck — green
• .venv/bin/ruff format --check backend/ — green (CI parity)
• make test-unit — 1190 passed in 5.71s (27 new + 1163 baseline)
• Empirical verification — 3 forward gain, 0 regressions (see above)
• make dashboard — clean regen, no errors
• CI green (.github/workflows/pr.yml)
• Gemini Code Assist review adjudicated
• Final GPT-5.5 review on merged diff

Tangential capture

docs/02_product/planned_features/chore_dashboard_regen_quoted_pr_false_positive/idea.md captures the pre-existing priority-3 fuzzy regex weakness surfaced during this chore's empirical verification. Out of scope here; provides a future-work target with three implementation options ranked by surgical-ness.

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request extends the dashboard regeneration script to extract PR numbers from idea.md files for legacy features that lack formal specification or pipeline status documents. It introduces strict regex patterns and a new **PR:** frontmatter convention to identify a feature's own PR while avoiding false positives from dependency references. The changes include updates to scripts/build_mvp1_dashboard.py, a new comprehensive test suite in backend/tests/unit/scripts/test_dashboard_pr_extraction.py, and documentation updates in architecture.md. Review feedback suggests improving the robustness of the new regex patterns by supporting markdown links consistently and refining the one_liner extraction logic for legacy features.

[gemini-code-assist]

Pattern B (_IDEA_STATUS_IMPLEMENTED_RE) should support the same markdown link alternation as Pattern A. This ensures robustness if an implemented status line uses a linked PR number, and maintains consistency across the strict idea-body patterns documented in architecture.md.

Suggested change

	_IDEA_STATUS_IMPLEMENTED_RE = re.compile(
	r"^\*\*Status:\*\*\s+\*\*Implemented\s*[—\-]\s*PR\s*#(\d+)\b",
	re.MULTILINE,
	)
	_IDEA_STATUS_IMPLEMENTED_RE = re.compile(
	r"^\*\*Status:\*\*\s+\*\*Implemented\s*[—\-]\s*PR\s*"
	r"(?:\[#(\d+)\b\]\([^)]*\)|\[#(\d+)\b\]|#(\d+)\b)",
	re.MULTILINE,
	)

[gemini-code-assist]

Pattern C (_IDEA_SHIPPED_DATELINE_RE) should support the same markdown link alternation as Pattern A for consistency and robustness against linked PR references in legacy datelines.

Suggested change

	_IDEA_SHIPPED_DATELINE_RE = re.compile(
	r"^\*\*shipped\s+\d{4}-\d{2}-\d{2}\s+as\s+PR\s*#(\d+)\b",
	re.MULTILINE,
	)
	_IDEA_SHIPPED_DATELINE_RE = re.compile(
	r"^\*\*shipped\s+\d{4}-\d{2}-\d{2}\s+as\s+PR\s*"
	r"(?:\[#(\d+)\b\]\([^)]*\)|\[#(\d+)\b\]|#(\d+)\b)",
	re.MULTILINE,
	)

[gemini-code-assist]

In _load_implemented, the one_liner extraction should fall back to _extract_idea_problem(idea) if spec is missing, consistent with the logic in _load_planned. This ensures that legacy idea-only features display their problem description in the dashboard summary column instead of falling back to the generic "Complete" status line. Additionally, consider including idea in the status_line extraction (at line 791) for better descriptive coverage of legacy folders.

Suggested change

	one_liner = _extract_one_liner(spec, source_dir=folder_path)
	pr = _extract_pr_number(pipe, plan, spec)
	pr = _extract_pr_number(pipe, plan, spec, idea)
	one_liner = (
	_extract_one_liner(spec, source_dir=folder_path)
	if spec
	else _extract_idea_problem(idea, idea_dir=folder_path)
	)
	pr = _extract_pr_number(pipe, plan, spec, idea)

[SoundMindsAI]

Review adjudication (Gemini Code Assist + GPT-5.5 final review)

Commits landing fixes: 5a90d989 (Gemini), d03e3d5f (GPT-5.5 final).

Gemini Code Assist (3 findings)

#	Sev	Location	Verdict	Notes
1	Medium	scripts/build_mvp1_dashboard.py:512 (Pattern B regex)	Accepted	Fixed in 5a90d989 — added the same markdown-link alternation (?:\[#(\d+)\b\]\([^)]*\)|\[#(\d+)\b\]|#(\d+)\b) Pattern A carries. Purely additive; never reduces matches. Symmetric across all three strict patterns now.
2	Medium	scripts/build_mvp1_dashboard.py:520 (Pattern C regex)	Accepted	Same fix as #1 in 5a90d989.
3	Medium	scripts/build_mvp1_dashboard.py:783 (_load_implemented one-liner)	Accepted	Fixed in 5a90d989 — added _extract_idea_problem(idea) fallback (symmetric with _load_planned). Idea-only legacy folders now render meaningful one-liners in the dashboard instead of empty/Complete-only. Material UX improvement on top of the chore's primary PR# extraction value.

GPT-5.5 final review (2 findings)

#	Sev	Location	Verdict	Notes
1	Medium	docs/00_overview/MVP1_DASHBOARD.md:151 (this chore's own planned row shows [PR #4] merged 2026-05-15)	Deferred	Counter-evidence: explicitly documented as cosmetic-only in the PR body's "Cosmetic anomaly" section. The Plan-section row shows misleading PR# (extracted by the pre-existing priority-3 fuzzy regex from AC-body narrative quoting OTHER features' merge events). Resolves automatically when this PR merges — _load_implemented takes over the row rendering and reads the actual PR# from finalization. The underlying priority-3 weakness is captured as chore_dashboard_regen_quoted_pr_false_positive. Tightening the priority-3 fuzzy regex is explicitly out of scope per spec §3 ("Changing _extract_pr_number's priority 1–3 logic" — those paths already work; only the new 3.5/3.6 idea-aware paths are in scope).
2	Medium	scripts/build_mvp1_dashboard.py:559-568 (_extract_metadata_block title-once flag)	Accepted	Real bug. When an idea body begins with metadata-key lines (NO # title at top), a later # within the 30-line cap was treated as the title because title_seen was only ever set on # lines. Fixed in d03e3d5f — replaced title_seen with nonblank_seen. Added regression test test_h1_after_metadata_is_body_heading_not_title locking the exact scenario.

Outcomes

• Applied fixes (4): 5a90d989 (Gemini docs: reorganize into numbered IA, add pipeline skills and feature templates #1/docs: MVP1 architecture + 12 feature specs + GPT-5.5 review fixes #2/chore(pipeline): add project-wide priority-ordered status mode #3), d03e3d5f (GPT-5.5 docs: MVP1 architecture + 12 feature specs + GPT-5.5 review fixes #2). 28 unit tests pass; CI green on the final SHA.
• Rejected with cited counter-evidence (0).
• Deferred as non-regression follow-ups (1): GPT-5.5 docs: reorganize into numbered IA, add pipeline skills and feature templates #1 — captured in chore_dashboard_regen_quoted_pr_false_positive; cosmetic only, resolves on merge.

Ready for human review + merge.