[v0.8] Manifestation report for approved guidance

[Stahl-G]

Context

Natural-language guidance is a soft, cheap experimentation layer. Its usefulness should be observed, not mistaken for deterministic enforcement.

The first real loop test showed both sides:

• approved guidance can influence later output;
• it can also fail to manifest, partially manifest, or conflict with other instructions.

v0.8 should measure this behavior without pretending memory is a contract or gate.

Problem

Approved guidance currently has limited post-run observability:

• operators cannot easily see whether a guidance entry manifested;
• failures may be misread as bugs in the preference layer;
• useful regressions may be missed;
• there is no structured way to learn which kinds of guidance should be promoted into templates, contracts, or gates.

Scope

Add a manifestation report for approved guidance.

For each approved guidance entry in the run snapshot, record an observable status such as:

• explicitly_reflected
• partially_reflected
• contradicted
• not_observable

Optionally record supporting notes or artifact references.

Boundary

This report is diagnostic. It must not automatically:

• mutate guidance;
• approve new guidance;
• reject old guidance;
• rewrite the brief;
• turn preference memory into a quality gate.

Acceptance criteria

• Each run can report whether approved guidance from the snapshot appears to manifest.
• Report distinguishes full manifestation, partial manifestation, contradiction, and not-observable cases.
• Report is clearly documented as diagnostic, not enforcement.
• At least one evaluation or fixture demonstrates a guidance entry that is not observable.
• The report can inform future promotion of stable preferences into templates/contracts/gates, but does not do that automatically.

Non-goals

• No autonomous memory updates.
• No preference compiler.
• No hard guarantee that natural-language guidance will be followed.
• No current-run hot reload.
• No automatic quality score.

[Stahl-G]

Triage after v0.8.2: keep open and attach to v0.8.5 / 080.

080 schema validation exists, but the runtime-facing pieces are not implemented yet: register-run, score-run, guidance manifestation report import, summarize, scaffold-condition, and public-safe pilot fixture.

This issue maps directly to v0.8.5 and should remain open until guidance_manifestation_report exists as a diagnostic report, not a production gate.

[Stahl-G]

Closed by the merged Guidance Manifestation diagnostic work (#331, with #334 follow-up). BriefLoop now has a run-local guidance_manifestation_report.json surface with schema validation, status / Quality Panel projection, materialized-entry checks, and a public-safe not_observable eval fixture. Boundary: this is diagnostic-only. It does not mutate Improvement Memory, approve new guidance, rewrite briefs, become a gate, or claim output-quality improvement.