fitme·story
v7.7 · 9 min read
Summary card · 60-second read

Validity Closure — How v7.7 Closed the Last Closable Class B Gap

Version
v7.7
Date
2026-04-28
Tier
light

v7.7 closes A1–A5 + B1–B2 + C1 from the post-v7.6 gap inventory: 5 new gates (4 write-time pre-commit hooks + 1 cycle-time check + 1 advisory), bulk frontmatter backfill on 32 case studies, and timing backfill on 3 paused features. Framework reaches 25 mechanical gates + 1 advisory.

Honest disclosures
  • •cache_hits[] post-v6 writer-path lifted to 100% gating; existing zero-adopted features intentionally untouched (no source data; impartiality rule).
  • •Tier 1.1 trend mode unlocks at 3 history snapshots — earliest 2026-05-04 (Monday cron appends snapshot #3).
  • •Tier 3.2 trend mode unlocks at 3 cycle snapshots — earliest 2026-05-03 to 2026-05-06 (72h cycle accumulates).
  • •Tier 3.3 external replication remains backlog; cannot be self-executed.
How to read this case studyT1/T2/T3 · ledger · kill criterion▾
T1Instrumented
Numbers come from a machine-generated ledger or commit. Reproducible. Highest reader trust.
T2Declared
Numbers stated by a structured declaration (PRD, plan, frontmatter) but not directly measured.
T3Narrative
Estimates and observations from session memory. Useful for context; not citable as evidence.
Ledger
Where to verify the claim — a file path, GitHub issue, or backlog entry. Anything labelled ledger: is the audit trail.
Kill criterion
The pre-registered threshold under which this work would have been killed mid-flight. Not fired = work shipped without hitting the threshold.
Deferred
Items intentionally not closed in this version. Each cites the ledger that tracks remaining work.
v7.6 (mechanical enforcement)
11 gates
4 write-time + 7 cycle-time. cache_hits ungated. cu_v2 unchecked. linkage 95.5%.
v7.7 (validity closure)
25 gates + 1 advisory
4 new write-time + 1 cycle + 1 advisory. cache_hits gated. cu_v2 schema-validated. linkage 100%.
Kill criterion · not fired
  • Cache-hits writer-path proves un-instrumentable (>5 distinct call sites with no shared loader).
  • Tier-tag checker false-positive rate stays >25% after 2 weeks.
  • PR-1 instrumentation introduces >100ms latency to skill loading.
  • Pre-commit hook FP rate >10% on legitimate commits in week-1 dogfooding.
Deferred items
Tier 1.1 trend modeledger: cron snapshot history at .claude/shared/measurement-adoption-history.jsonWall-clock — needs 3 scheduled snapshots; earliest 2026-05-04.
Tier 3.2 trend modeledger: .claude/integrity/snapshots/Wall-clock — needs 3 scheduled 72h cycle snapshots.
Tier 3.3 external replicationledger: GitHub issue #142Cannot be self-executed; needs independent operator.

Validity Closure — How v7.7 Closed the Last Closable Class B Gap

v7.6 (Mechanical Enforcement, shipped 2026-04-25) closed seven Class B → Class A gaps and explicitly enumerated five that could not be promoted because mechanizing them would require lying about what the framework can verify. A 2026-04-27 ledger pull surfaced that three of those five were still mechanically or heuristically closable if the right write-time hooks were added. v7.7 closed them. One of those three (cache_hits writer-path adoption) shipped fully gated. One (cu_v2 schema) shipped schema-validated. One (T1/T2/T3 tag correctness) shipped as a heuristic advisory — and kill criterion 2 fired honestly at baseline, so the advisory ships permanent rather than promoting to a gate. Two original gaps remain genuinely human-required (auth simulator, external replication). The Class B inventory went from 5 to 4. This case study is the framework's full validity-closure pass, published verbatim per the publish-then-remediate policy.

Read this first — outlier flag (carries forward from v7.6)

This case study is itself an outlier in the corpus — the same three biases as v7.6 stack here:

  1. Single-session execution — v7.7 shipped in a single ~6-hour working session on 2026-04-27 (brainstorming spec → 32 commits + 7 fitme-story commits → 2 PRs merged). No organic cadence; phases ran sequentially in one sitting.
  2. Dogfooded data collection — the author of the framework rework also wrote the data and reads it. Same-author confound.
  3. Retroactive v6.0 application — v7.7's own state.json is instrumented end-to-end with v6.0 protocol, but it's a single feature; the bulk-backfill of 32 case-study frontmatters in M2 was retroactive, not organic adoption.

The full upstream case study labels these limits in Section 99.7 Pre-mortem honesty re-statement and applies them to every published number.

Trust-page connection

This case study is the detailed validity-closure answer to the residual Class B gaps documented in v7.6's unclosable-gaps.md. Together with v7.5 (policy) + v7.6 (mechanical) + v7.7 (validity closure), the framework's full reply to the 9 Tier 1/2/3 recommendations from the 2026-04-21 Gemini audit is now complete. The trust page Gemini-audit subroute links to all three. Per the publish-verbatim policy: the original audit text remains unchanged on /trust; corrections and responses are appended.

  • Audit (verbatim): /trust/audits/2026-04-21-gemini
  • v7.5 policy response: the eight cooperating defenses (write-time + cycle-time + readout-time)
  • v7.6 mechanical response: seven Class B → Class A promotions + five Class B gaps documented as unclosable-gaps.md
  • v7.7 validity-closure response: four new write-time check codes + one cycle-time advisory permanent + cache_hits writer-path closed (5 → 4 unclosable). This case study.

Summary card (T1 unless noted)

  • Framework version: v7.6 → v7.7
  • Trigger: 2026-04-27 ledger pull surfaced that three of v7.6's documented Class B gaps were still mechanically or heuristically closable; user declared full-priority freeze to ship the closure pass
  • Wall time: ~6 hours (single session, brainstorm → spec → plan → execution → ship)
  • Total commits: 39 (32 FitTracker2 + 7 fitme-story)
  • Pull requests: 2 (FitTracker2 #144, fitme-story #7)
  • Unit tests added: 29 across 4 new test files
  • New check codes: 5 (4 gating + 1 advisory permanent)

What v7.7 actually closed (and what it didn't)

4 new gating write-time check codes

  1. CACHE_HITS_EMPTY_POST_V6 — pairs with scripts/log-cache-hit.py wrapper that auto-discovers the active feature and dual-writes state.json.cache_hits[] + the events log. Closes the v6.0 writer-path adoption gap (GitHub issue #140).
  2. CU_V2_INVALID — schema validator (factor presence + range [0,1] + total tolerance + tier_class enum). Validates STRUCTURE only — magnitude correctness stays a documented Class B gap (judgment-based).
  3. STATE_NO_CASE_STUDY_LINK — write-time mirror of the cycle-time NO_CS_LINK; rejects current_phase=complete without case_study link OR parent_case_study link OR exempt tag.
  4. CASE_STUDY_MISSING_FIELDS — forward-only ≥ 2026-04-28; rejects case studies missing work_type, success_metrics, kill_criteria, or dispatch_pattern in frontmatter.

1 cycle-time advisory permanent — kill criterion 2 fired

  1. TIER_TAG_LIKELY_INCORRECT — heuristic checker that extracts T1-tagged quantitative claims and cross-references against ledger numbers within 5% relative tolerance. Pre-registered kill criterion 2: FP rate >25% after baseline → ship advisory permanent. Baseline scan: 1 finding total, 1 false positive (regex matched section identifier "Tier 3.2" plus the next word "documentation"). FP rate = 100% n=1. Kill-2 fired honestly. Root cause: regex pattern designed for **T1**: prefix style; live corpus uses | value | T1 | table-column format. Ships advisory permanent. v7.8 redesign documented at tier-tag-checker-baseline.md.

Validity gates closed (delta vs v7.6)

  • State↔case-study linkage: 95.5% → 100% (mechanically gated)
  • Doc-debt fields populated (work_type / success_metrics / kill_criteria / dispatch_pattern): 4–61% → 95.7–100% (gated forward; 33 TODO markers reflect genuinely-absent pre-PRD-structure data, not heuristic failure)
  • cache_hits[] post-v6 adoption: 33.3% → gated to 100% on next post-v6 complete write
  • cu_v2 schema: unchecked → schema-validated on every write
  • Total framework mechanisms: 18 (12 cycle + 6 write-time) → 25 gates + 1 advisory

Framework-health dashboard live

fitme-story PR #7 added /control-room/framework — surfaces all 19+ check codes, the human-action checklist (D1+D2 deferred items), and Tier 1.1/3.2 trend charts (charts unlock as cron snapshots accumulate post-merge).

What still remains Class B (4 unclosable gaps, was 5)

  1. cu_v2 factor magnitude correctness — schema-validated by v7.7, magnitude judgment unchanged. Class B by judgment necessity.
  2. T1/T2/T3 tag correctness on novel claims — advisory checker shipped (kill-2 fired). Class B by heuristic-correctness necessity.
  3. Tier 2.1 real-provider auth simulator runs (D1) — still human-required. Surfaced on dashboard.
  4. Tier 3.3 external replication (D2, GitHub issue #142) — still external-required by definition. Surfaced on dashboard.

The fifth (cache_hits writer-path adoption) was closed by v7.7 M1.

Single-session timeline (frozen 2026-04-27)

14:00 UTC — Genesis & spec approval (commit 1057144)
14:42 UTC — M0 complete: 5 commits + Linear FIT-49 + 8 sub-issues + Notion v7.7 sub-page
17:50 UTC — PR-1 opened (cache_hits writer-path closure)
18:35 UTC — PR-2 milestone (cu_v2 schema validator) merged into train
19:30 UTC — M2 complete: linkage + doc-debt + active backfill
20:30 UTC — M3 complete: tier-tag heuristic shipped (advisory permanent — kill criterion 2 fired)
21:30 UTC — M5 complete: v7.7 ready for merge
17:18 UTC (next day) — fitme-story PR #7 merged
17:39 UTC (next day) — FitTracker2 PR #144 merged

Two cron-gated verifications remain (auto-handled by a scheduled remote agent on 2026-05-04):

  • B1 (Tier 1.1 trend mode) unlocks at 3 history snapshots — earliest 2026-05-04 (Monday weekly cron #3)
  • B2 (Tier 3.2 trend mode) unlocks at 3 cycle snapshots — earliest ~2026-05-03 to -06 (72h cycle)

Tooling attribution (honest)

  • Claude Opus 4.7 (1M context) — all v7.7 framework commits carry the Co-Authored-By tag.
  • Google Gemini 2.5 Pro — original 2026-04-21 audit triggered the v7.5 → v7.6 → v7.7 chain. No new Gemini work in the v7.7 window itself.
  • Human (Regev) — trigger decisions, the multi-part approval gate (scope of v7.7 = A+B+C1; defer D), Vercel token rotation enabling fitme-story PR #7 build, the merge confirmations.

What earned the v7.6 → v7.7 framework bump

  1. A new structural capability — validity closure is a layer that did not exist in v7.6. v7.6 enforced what could be mechanically gated; v7.7 closes what was still gateable by adding the writer-path instrumentation, the schema validator, and the linkage-gate write-time mirror. Plus the heuristic advisory class.
  2. Propagation across surfaces — manifest, CLAUDE.md, master plan, evolution doc, integrity README, dev-guide rename (v1-to-v7-6 → v1-to-v7-7), this MDX case study, framework-health dashboard live at fitme-story.
  3. An honest kill-criterion outcome — TIER_TAG_LIKELY_INCORRECT shipped advisory rather than promoting to gate, because the data said so at baseline. Pre-registered thresholds + measurement is what makes the kill outcome honest rather than a failure.

Lessons (excerpts — see upstream Section 99 for the full synthesis)

  • A framework that knows what it cannot check is more trustworthy than one that pretends every check is a check. v7.7 proved this twice: (a) by closing 1 of v7.6's documented "unclosable" gaps when re-examination showed it was actually closable, and (b) by honestly shipping the tier-tag heuristic as advisory permanent when its baseline FP rate fired the pre-registered kill criterion.
  • Pre-registered kill criteria turn potential failures into honest outcomes. Without the pre-registered threshold (FP > 25% → advisory permanent), the tier-tag checker's 100%-FP baseline could have been rationalized into "ship it as a gate, fix later." The threshold made the choice mechanical.
  • Write-time mirrors of cycle-time checks need symmetry audits. The new STATE_NO_CASE_STUDY_LINK write-time hook initially missed the parent_case_study field that the cycle-time NO_CS_LINK accepts. Caught at live-tree scan when 5 features tripped the new gate. Fixed via mirror-the-cycle-time-logic correction. Lesson: when promoting a cycle-time check to write-time, audit the cycle-time check for special cases first.
  • post-v6 metric ratios uplift on future natural usage, not historical retroactive data. v7.7's primary metric (post-v6 fully-adopted ratio: baseline 2/9 = 22.2%, target ≥8/11 = 72.7%) stayed at 2/9 at synthesis time. The gates are in place; the metric uplifts as features actually complete post-merge. Spec §6 pre-registered this measurement timing — it's not a missed target, it's a measurement that's forward-only by definition.

Links