Rule Candidates

The staging area for reusable workflow rules before they become permanent.

Active Run · /home/jonas/social-carousel-codex/docs/rule-candidates.md

Last modified: 2026-03-29T12:28:12.882Z

# Rule Candidates

## Purpose

This file is the durable staging area for reusable workflow rules that were learned from a specific run or operator feedback but are not yet approved for promotion into:

- `AGENTS.md`
- `.codex/skills/`
- `docs/workflow-spec.md`
- `docs/qc-execution.md`
- other durable workflow guidance

It exists so the repo can capture generalizable feedback without silently turning one run's friction into global policy.

## Status Vocabulary

- `proposed`: captured durably, waiting for explicit approval
- `approved`: explicitly approved, waiting to be promoted into durable docs or skills
- `promoted`: already reflected in durable docs or skills
- `rejected`: intentionally not promoted

## Approval-Gated Flow

1. Check whether new human feedback is local to one run or generalizable across runs.
2. If it looks reusable, add a candidate entry here with:
   - candidate id
   - source run
   - summary
   - rationale
   - proposed promotion targets
   - current status
3. Present the candidate explicitly for approval.
4. Only after approval, update durable repo guidance and mark the candidate `promoted`.

## Active Proposed Candidates

There are no active proposed candidates right now.

## Recently Promoted Candidates

### RC-20260329-004

- Status: `promoted`
- Source run: `IDEA-20260329-postmeal-walk-glucose-realrun-02`
- Summary:
  Cover hooks may be stronger and more curiosity-driven than the full research framing as long as the later slides add the missing nuance inside the approved lock.
- Why it exists:
  The approved research boundary for the post-meal walking run was acceptable, but the first preview underperformed because the cover tried to carry too much nuance too early instead of behaving like a social-media hook.
- Proposed promotion targets:
  - `AGENTS.md`
  - `.codex/skills/carousel-workflow/SKILL.md`
  - `docs/workflow-spec.md`
  - `docs/main-skill-design.md`
- Implementation result:
  Durable guidance now explicitly allows curiosity-led covers when the later slides carry the nuance and the full sequence stays inside the approved lock.

### RC-20260329-005

- Status: `promoted`
- Source run: `IDEA-20260329-postmeal-walk-glucose-realrun-02`
- Summary:
  Each content slide should carry one distinct editorial role, not just one more restatement of the same research point.
- Why it exists:
  The first approved draft was evidence-safe but flat because the slides did not clearly separate roles like mechanism-level framing, realism, caveat handling, and practical takeaway.
- Proposed promotion targets:
  - `AGENTS.md`
  - `.codex/skills/carousel-workflow/SKILL.md`
  - `docs/workflow-spec.md`
  - `docs/main-skill-design.md`
  - `docs/qc-execution.md`
- Implementation result:
  Durable guidance now treats distinct editorial roles per content slide as the default expectation for future runs and copy review.

### RC-20260329-006

- Status: `promoted`
- Source run: `IDEA-20260329-postmeal-walk-glucose-realrun-02`
- Summary:
  The carousel should tell a logical story across the content slides rather than repeating the same claim in slightly different words.
- Why it exists:
  The first preview improved only after the copy was reorganized into a deliberate sequence:
  why it helps, why it is realistic, what not to overclaim, and how to use the takeaway.
- Proposed promotion targets:
  - `AGENTS.md`
  - `.codex/skills/carousel-workflow/SKILL.md`
  - `docs/workflow-spec.md`
  - `docs/main-skill-design.md`
  - `docs/qc-execution.md`
- Implementation result:
  Durable guidance now treats story progression, not repetition, as the default content-slide sequencing expectation.

### RC-20260329-007

- Status: `promoted`
- Source run: `IDEA-20260329-postmeal-walk-glucose-realrun-02`
- Summary:
  Evidence caveats should be translated into natural consumer language instead of sounding like academic disclaimers inside the body slides.
- Why it exists:
  The evidence boundary stayed the same, but the revision became more usable only after the caveats were rewritten from abstract safety language into direct consumer phrasing like `may-help habit` and `not a guaranteed fix`.
- Proposed promotion targets:
  - `AGENTS.md`
  - `.codex/skills/carousel-workflow/SKILL.md`
  - `docs/workflow-spec.md`
  - `docs/main-skill-design.md`
  - `docs/qc-execution.md`
  - `docs/helper-surface.md`
- Implementation result:
  Durable guidance now treats natural consumer-language caveat translation as the default within-lock copy path instead of leaving academic-sounding disclaimers in body-slide prose.

### RC-20260329-001

- Status: `promoted`
- Source run: `IDEA-20260329-protein-breakfast-satiety-realrun-01`
- Summary:
  Make preview artifact folders version-specific instead of reusing one folder per idea.
- Why it exists:
  `V2` preview generation overwrote the `V1` preview artifact bundle in place even though the ledger kept version lineage.
- Proposed promotion targets:
  - `scripts/runtime_state.py`
  - `docs/helper-surface.md`
  - `docs/preview-verification.md`
- Implementation result:
  New preview payloads now derive `folderName` from `<idea_id>-<draft_version>`. Older per-idea preview folders remain valid and are not rewritten in place.

### RC-20260329-002

- Status: `promoted`
- Source run: `IDEA-20260329-protein-breakfast-satiety-realrun-01`
- Summary:
  Prefer the public verified preview URL in `preview.json` and the ledger when public verification succeeds.
- Why it exists:
  The current workflow verifies the public route successfully but still records the local `127.0.0.1:3101` URL as the primary preview URL.
- Proposed promotion targets:
  - `scripts/runtime_state.py`
  - `docs/helper-surface.md`
  - `docs/preview-verification.md`
- Implementation result:
  `verify-preview` now promotes the public verified URL to the top-level `viewer_url` and `api_url` when the public HTTP checks pass, while preserving artifact-local URLs separately.

### RC-20260329-003

- Status: `promoted`
- Source run: `IDEA-20260329-protein-breakfast-satiety-realrun-01`
- Summary:
  Separate a pre-approval research-lock proposal artifact from the final approved lock so `approved_at` is not required before human approval.
- Why it exists:
  The current wrapper writes a helper-ready lock input before research approval, which creates semantic awkwardness around approval-only fields.
- Proposed promotion targets:
  - `scripts/literature_review_wrapper.py`
  - `docs/literature-review-wrapper.md`
  - `docs/helper-surface.md`
- Implementation result:
  The wrapper now writes `research-lock-proposal.json` for approval-gated runs and only writes `research-lock-approval-input.json` when it is actually auto-persisting an approved lock.

## Already Approved Durable Changes

These are now default workflow rules, not pending candidates:

- concise cover headlines by default
- curiosity-led covers are acceptable when later slides carry the nuance inside the approved lock
- at least 4 content slides by default unless shorter is explicitly justified
- standalone readability for each content slide
- one distinct editorial role per content slide
- logical story progression and non-repetition as explicit workflow expectations
- evidence caveats translated into natural consumer language by default
- default `copy_within_lock` handling for this family of revision feedback
- version-specific preview artifact folders
- public verified preview URLs as the preferred recorded review URL
- pre-approval research-lock proposal artifacts
- approval-gated promotion flow for future reusable rule changes

Ready