# E-039 — Close counts overstated analysis readiness

## Objective hierarchy

Parent objective `O-PASSIVE-VALIDATION`: determine whether observed trajectories justify extending Objective Guard beyond simple change-size baselines. Round `R-F4-DATA-AUDIT` contributes to `KR-DATA-READINESS` through reproducible provenance, grouping and labeling readiness.

## Observed trajectory and evidence

The initial handoff inspected the versioned 46-close report and four current closes. A subsequent read of registered source logs found 157 historical Objective Guard closes and 50 Schema Forge closes, including two explicit correct-outcome labels. The earlier finding of no labels applied only to the inspected snapshots, not all available logs.

The [audit](../product/09-passive-data-audit.md) preserves 211 closes, eight without identity and 44 identified groups. There are 159 additional closes within groups and seven groups span contract amendments. Only 21 latest observations have explicitly normal context and valid scope; none carries its own outcome label. Both existing labels refer to earlier observations and cannot automatically label later closes.

Among 20 eligible complete nonzero surfaces, churn supplies a median 93.61% of the heterogeneous sum. This diagnoses feature scaling in this sample, not whether churn or the sum predicts drift better.

## Expected and actual guard behavior

Preflight correctly reported the prior Awareness integration round as closed. `awc start` first rejected the successor with `task-objective-mismatch,round-terminal`. The published `oguard next-round` opened the declared successor and preflight returned ready. Retrying `awc start` then failed with `awareness-transition-failed` because its installed Awareness entry point was missing. The installed standalone `awareness focus` recorded the work instead. No AWC package was changed in this analysis round.

The declared analysis check tests identity grouping, missing measurements, label provenance/conflicts, duplicate events and censoring. It does not certify that historical outcomes are correct or that statistical features help agents. Guard closure evaluates this implementation contract, not scientific acceptance.

## Classification and correction

This is a material correction of data readiness and an integration limitation. It is not a measured false negative of the drift detector: sample independence, scientific adjudication and AWC dependency packaging were not declared detector capabilities.

Keep the old snapshot intact, freeze sanitized source measurements, group amendments with their original round and preserve missing metadata. Leave outcomes unassigned unless evidenced on the exact observation. Retain the AWC failure as a follow-up rather than substituting an integration repair for the requested data analysis.

## Impact on the thesis and follow-up

The audit contributes to the parent objective by exposing what cannot yet be inferred. It neither validates nor refutes useful early drift detection. Continue with independent adjudication and prospective observations; compare simple baselines before activating a statistical signal. Repair the installed AWC successor/dependency workflow in a separate integration round. Passing the declared guard only establishes that its checks found no in-scope violation.

## Append-only correction — explicit feature obligations

At the first close, all 163 tests and 1,060 documentation checks passed, and alignment reported zero outside-scope files. Nevertheless, both `oguard round-end` and `npm run round:end` returned `UNKNOWN`: the author had declared `AC-ANALYSIS` but omitted the feature's behavior and compatibility obligation references. Preflight had returned ready; it did not establish closure readiness. This was a useful closure signal about an incomplete contract, not failed analysis behavior or scientific validation.

Commit `80ff9ca` preserves that implementation and original contract as a checkpoint. The author then opened successor `R-F4-DATA-VALIDATION` with the same parent objective, explicit behavior/compatibility references and full-suite plus documentation checks. This is retrospective validation of existing work, not evidence that obligations were frozen before the original implementation. No thresholds or outcomes were changed to obtain closure, and the frozen data snapshot excludes this later dogfooding.