PE Donor Promotion Checkpoint

2026-06-15

PE Donor Promotion Checkpoint

Status: slice_pass Date: 2026-06-15 Branch: codex/pe-donor-promotion-2026-06-15

The Single Idea

ainative-academy/personalization-engine-aina is now treated as a first-class donor source candidate, not as bulk production truth. The engine room promotes only audit-backed or clearly useful contract/source-evidence surfaces; prompt/workflow/ontology bulk stays advisory until audited; known-bad market_v2 and context-blind K2 surfaces stay quarantined.

What Was Checked

Authenticated GitHub access confirmed the private repo is the older AINA personalization engine repo with ontology, vector recall, exercise composer, prompt assets, workflows, governance, and source-intelligence lineage. The public web page does not expose the private contents, so the check used gh plus local donor copies.

Key donor surfaces found:

Surface	Decision	Why
Verified data foundation status	Promote candidate, verify first	The handoff documents O*NET ingest, JD pool, title-to-canonical role, and tool registry evidence.
Evidence roadmap/internal-preview slice	Promote candidate, verify first	Useful source-evidence scoring and role/workflow roadmap lineage, but not public-production authority.
Ship-loop/operator contract	Promote candidate, verify first	Strong contract lineage for assessment to packet to sandbox to evaluator.
Prompt/workflow/ontology bulk	Advisory only	Large and valuable, but bulk counts are not audit proof.
Title taxonomy and bucket outputs	Promote candidate, verify first	High-value JD-aware cleanup lineage, but still needs the missing audit or replacement proof before row promotion.
Workflow grounding consensus batches	Promote candidate, verify first	Post-May-15 work should be reconciled before new workflow LLM review.
`market_v2` and context-blind K2	Blocked/quarantined	Explicitly known-bad lineage; preserve only as anti-loop evidence.

Claude Checkpoint

Claude Opus was invoked in interactive plan mode. The first pass tried to explore too deeply and was stopped. A second constrained pass answered from the verified facts only.

Verdict: proceed, but as gated allowlist promotion of the verified subset, not wholesale adoption. Claude agreed that using attested donor work before more title-by-title LLM repair is the right order, and warned that the large prompt/workflow/learning-object surfaces should remain candidates until each family has an audit receipt.

What Changed

Updated src/aina_data_engine/prior_work_source_authority_promotion.py to add seven explicit personalization-engine-aina donor asset rows:

pe_aina_verified_data_foundation_status
pe_aina_evidence_roadmap_internal_preview
pe_aina_ship_loop_operator_contract
pe_aina_prompt_workflow_ontology_bulk_candidates
pe_aina_title_taxonomy_bucket_outputs
pe_aina_workflow_grounding_consensus_batches
pe_aina_market_v2_and_context_blind_k2_quarantine

Updated tests/test_prior_work_source_authority_promotion.py so temp fixtures prove the policy without relying on /home/ali/... donor paths.

Regenerated:

The regenerated receipt now reports 26 total prior-work assets, 7 PE donor rows, 5 PE promote candidates, 1 PE advisory bulk candidate, and 1 PE blocked/quarantined surface.

Validation

Commands run:

uv run pytest tests/test_prior_work_source_authority_promotion.py -q
uv run ruff check src/aina_data_engine/prior_work_source_authority_promotion.py tests/test_prior_work_source_authority_promotion.py
uv run aina-data-engine --root /srv/aina/aina-data-engine-room ain-506-p0-gate
uv run aina-data-engine --root /srv/aina/aina-data-engine-room validate
uv run aina-data-engine --root /srv/aina/aina-data-engine-room ain-510-retrieval-promotion-gate
uv run pytest tests/test_prior_work_source_authority_promotion.py tests/test_donor_retirement_pack.py tests/test_source_authority_registry_v2.py -q
uv run aina-data-engine --root /srv/aina/aina-data-engine-room production-runtime-readiness
git diff --check

Results:

Focused prior-work tests: 5 passed
Focused registry/retirement/prior-work tests: 10 passed
Ruff: pass
AIN-506: pass
Validate: pass
AIN-510: promotion_ready, promotion_eligible: true
Production runtime readiness: ready_to_harden_headless_production_runtime
Diff check: pass

Next Slice

Promote actual PE-derived engine-room assets only after deterministic reconciliation:

Copy or derive a self-contained source-foundation package for the PE promote-candidate rows.
For each candidate, record input path, source hash, current engine-room consumer, and whether it supersedes an existing gap.
Start with the ship-loop/operator contract and evidence-roadmap lineage because they are smaller and contract-shaped.
Then reconcile title taxonomy/bucket outputs against current JD-aware role context and source authority receipts.
Keep prompt/workflow/ontology bulk as advisory until a family-level audit and clean derived artifact exist.

Resume Prompt

Resume the PE donor-promotion milestone in /srv/aina/aina-data-engine-room on branch codex/pe-donor-promotion-2026-06-15. Start from docs/handoff/2026-06-15-pe-donor-promotion-checkpoint.md. Treat personalization-engine-aina as a first-class donor source candidate but not bulk authority. Promote only audit-backed or contract-shaped PE assets into self-contained engine-room receipts, keep prompt/workflow/ontology bulk advisory, keep market_v2/K2 quarantined, and rerun prior-work, AIN-506, validate, AIN-510, and runtime-readiness gates before closeout.

Ali Mehdi Mukadam - co-authored with Codex - 2026-06-15

topics:
  - personalization-engine
  - source-authority
  - donor-promotion
subtopics:
  - personalization-engine-aina
  - clean-before-embed
  - no-redundant-llm-review