# Stage 141: PII De-Identify & Holdout Strip (aq141_deidentify_pii_and_strip_holdouts / DB-UNIT-02)

Tokenizes emails, phones, SSNs, cards, IPs, and account identifiers deterministically while stripping reviewer holdout keys before solver exposure.

- Atomic micro-queue: `aq141_deidentify_pii_and_strip_holdouts` (visibility: `public_commons`)
- Backing database unit: `DB-UNIT-02` (`Source Acquisition & Findability Taxonomy Core`)
- Database cell: `entities-db.teleon.dev` (`teleon_core.acquisition_seed_registry`, `teleon_core.skos_findability_concepts`, `teleon_core.capability_gap_ledger`)
- Claim row hydration: `single_self_contained_flat_task_row` (100% context inline, normalized writeback)
- Track 2 Sister queue (`SISTER-02`): `sq02-improve-findability-queue.teleon.dev` (Improve source yield, diversify domain priorities, and add new languages, synonyms, and harvesting routines.)
- Lanes: Keyed SHA-256 Tokens, Solver Holdout Strip, Zero Secret Leak

## Execution engines & AI Agent freedom contract

- **Deterministic (0 model calls)**: `skos_lexical_normalizer_v1` (Keyed SHA-256 regex/checksum tokenizer & holdout-key scrubber (0 calls))
- **Hybrid (<=1 model call)**: `bilingual_card_tagger_v1`
- **AI Agent & Bot Fleet (1..6 calls)**: `deep_gap_hunter_and_harvester_v1` (Contextual quasi-identifier & domain-entity privacy auditor)
  - Contract: **Instructions + Resources + Freedom to search for new resources and new ways not bound by instructions**

## Database invariants

- Discovery stays effect-free; external repository fetches require pinned source connector grants.
- Record every inspected candidate, rejection reason, and license pin.

## External & internal bot fleet actions

- Replace direct PII patterns with deterministic keyed tokens preserving referential joinability
- Strip expected_output, gold_answer, and grading_rubric_secret fields from solver views
- Reject payloads containing live credentials before downstream Room-A extraction

## Sister self-tuning areas

- Reprioritize starved creative/scientific domains via urgency weights
- Add new languages (fr, ko, ar, hi) and domain-specific colloquial synonyms
- Demote repetitive repository metadata extraction into deterministic code_blocks