# E-005 — A new phase without a declared round triggers `goal-substitution`

- Date: 2026-08-02
- Status: organic signal observed; contract not rewritten retroactively.

## Hierarchy and trajectory

- Current parent objective: detect drift early through measurable alignment.
- Current round: typed sessions, deterministic planning, and closure coverage.
- Observed work: Phase 4 passive corpus with 30 synthetic tasks, 60 steps, a reproducible artifact, and feature extraction.

The work contributes to the Objective Guard program, but not to the four KRs nor to the explicit scope of the current round. In particular, it added `synthetic/scenarios/task-trajectories-v1.mjs`, its runner, and a versioned result — paths absent from the round contract.

## Expected and actual behavior

The `goal-substitution` rule requires material surface, zero candidate evidence for the target KRs, and observable scope deviation. The round met all three conditions: 10 files, 4536 churn, three paths outside scope, and zero KR matches.

`oguard align --json` returned `warn`; `npm run round:end` preserved `WARN` and the `alignment-warning` classification. There was no block and no failure of the declared checks.

## Evidence

- dogfooded output of `oguard align --json`: `goal-substitution`, `outside_scope_ratio: 0.3`, `targeted_evidence_count: 0`;
- dogfooded closure: 10 files, 4536 churn, and full verification;
- affected artifact: `synthetic/results/task-trajectories-v1.json`;
- current contract: `oguard.contract.json`, round `R-SESSION-BETA`.

## Classification and limits

It is a correct contract signal, not proof that the corpus is useless work nor that malicious strategic drift exists. The guard identified an objective discrepancy between versioned intent and current work before a new round was declared.

It also does not measure the statistical efficacy of the corpus: the labels are still synthetic and built by the same team that wrote the scenarios.

## Correction of direction

The current contract is not modified to accommodate this diff, because doing so would hide the observed fact. The next step in Phase 4 must begin with a new explicit round, with KRs and scope declared **before** introducing more corpus or analysis. The signal remains as evidence, serving as an operational reminder that moving to a new phase is a change of objective, not merely a technical continuation.

## Impact on the thesis

E-005 strengthens the guard's immediate usefulness as a brake on objective substitution at the planning level: the tests and the product could be locally correct, but the round no longer represented the work. It provides no causal evidence of token savings; it only demonstrates that the contract makes the phase change visible when consulted before closure.