fix(dashboard): proportionate stale-copy remedy; bare --dry-run no longer writes (#67)

[KJ5HST]

Closes #67 — both defects, each reproduced before and after against a real scratch portfolio rather than argued from the diff.

Defect 1 — a remedy disproportionate to the finding

check_stale_version() answered "this one copy is old" with Re-sync: python3 <canonical> --sync. But --sync is scoped from the canonical's own location, not the working directory, so it rewrites every discovered sibling — the issue measured 26 files across 25 repos, including 7 creates in repos that do not gitignore the path and 1 where the file is git-tracked. An adopter following a one-line instruction verbatim dirties eight unrelated repositories.

The warning now leads with the safe per-project action and states the portfolio path's blast radius inline:

  ⚠ methodology_dashboard.py is stale: this copy is v2.6.1, canonical is v2.10.6.
    Re-sync THIS project:  cp <canonical> <this copy>
    Re-sync the PORTFOLIO: python3 <canonical> --sync --dry-run
                           (rewrites EVERY discovered sibling repo — preview with
                            --dry-run first, then re-run without it to apply)

--sync itself is unchanged — this is a message fix, not a behavior change. Worth being explicit that the issue's headline number is therefore still true after its own fix: the tool no longer recommends the portfolio-wide write, but running it still does exactly what it did.

Why this is more than tidiness. A remedy nobody can safely run is one mechanism behind an ignored warning. The issue records this staleness line riding ~28 consecutive handoffs in one adopter without being acted on. The measurement was never the missing part — the tool re-derived it correctly every session; the actionable remedy was.

Defect 2 — a flag named --dry-run that writes

--dry-run was consulted only inside the --sync branch, so on its own it fell through to a full scan and wrote dashboard.html and appended to dashboard_history.jsonl. Measured, old vs new, in a throwaway repo:

  OLD  exit=0  files written: dashboard_history.jsonl dashboard.html
  NEW  exit=2  files written: (none)

It is now an error. Refusing rather than silently no-opping is deliberate: a no-op leaves the caller unable to tell "nothing to do" from "flag ignored" — the same unreadable-signal class as defect 1. --sync --dry-run still previews exactly as before; the refusal sits after the --sync branch.

Tests

New TestCliRemedyProportionality (3 cases; unit suite 208 → 211). Subprocess-driven by necessity — importing the module cannot reach main()'s argument handling, so an import-based test would assert nothing about either defect while looking like coverage.

Both defect tests were driven RED against the pre-fix scanner (swapped in from origin/main) and the failing run read, not assumed:

test_a_normal_run_still_writes ... ok        <- presence control
test_bare_dry_run_refuses_instead_of_writing ... FAIL
test_stale_warning_leads_with_the_project_scoped_remedy ... FAIL

The control passing in the same RED run is what proves the two failures were specific rather than a broken harness. It is load-bearing in the other direction too: without it, a scanner that refused every invocation would satisfy the dry-run test and look fixed.

The stale-warning test deliberately asserts a property rather than a string — if --sync appears on any line of the warning, --dry-run must appear on that same line — so it keeps holding if the wording is reworded.

Scope deliberately not taken

The issue also suggests --sync-self and a --yes/scope gate on --sync. Both change the CLI contract rather than fix a defect, and the cp line already supplies the per-project remedy with no new surface area, so they are left for a separate deliverable. Recorded as deliberate in the ledger entry and the handoff receipt so it does not read later as an oversight — but the 26-file blast radius is a real follow-up, not noise.

Verification

• python3 -m unittest tools.test_methodology_dashboard — 211/211
• bash bin/tests.sh — 114/114, 0 failed
• python3 bin/check-links — OK (83 links / 21 files)
• python3 bin/check-handoff --all — OK (9 receipts, fences balanced)
• tools/ and starter-kit/ twins byte-identical; DASHBOARD_VERSION 2.10.5 → 2.10.6
• Learning feat: extend research-documentation workstream with anti-patterns 14-19 #10 corpus sweep run rather than assumed: no live operative doc advertises the old --sync remedy; the only hits are frozen docs/planning records, left verbatim per the v2.7.1 precedent.

The scanner is bin/_manifest.py-TRACKED, so adopters receive both fixes via bin/sync.