# Stage 115: Text Similarity Queue (aq115_score_text_and_semantic_similarity / DB-UNIT-03)

Scores lexical, structural, and semantic similarity across documents, prompts, cards, and tasks in 0 model calls with LSH block emission.

- Atomic micro-queue: `aq115_score_text_and_semantic_similarity` (visibility: `public_commons`)
- Backing database unit: `DB-UNIT-03` (`Confusion Diagnostics & Card Disambiguation Core`)
- Database cell: `queues-db.teleon.dev` (`teleon_core.agent_confusion_incidents`, `teleon_core.disambiguation_patches`, `teleon_core.negative_constraint_rules`)
- Claim row hydration: `single_self_contained_flat_task_row` (100% context inline, normalized writeback)
- Track 2 Sister queue (`SISTER-03`): `sq03-improve-confusion-queue.teleon.dev` (Discover new confusion aspects, sharpen agent consumption formatting, and automate recurring repairs.)
- Lanes: Char 3/5-Gram Jaccard, 64-Perm MinHash, SimHash-256 Cosine

## Execution engines & AI Agent freedom contract

- **Deterministic (0 model calls)**: `schema_mismatch_rule_extractor_v1` (Shingle Jaccard, 64-perm MinHash & weighted SimHash-256 scorer (0 calls))
- **Hybrid (<=1 model call)**: `contrast_pair_card_rewriter_v1`
- **AI Agent & Bot Fleet (1..6 calls)**: `multi_trace_root_cause_investigator_v1` (Cross-lingual & paraphrase nuance arbitrator)
  - Contract: **Instructions + Resources + Freedom to search for new resources and new ways not bound by instructions**

## Database invariants

- Every confusion patch must include a deterministic regression probe reproducing the prior failure.

## External & internal bot fleet actions

- Compute sub-millisecond char-gram, token Jaccard, MinHash, and SimHash-256 similarity in 0 calls
- Deduplicate near-identical prompts, notes, and problem situations across surfaces
- Feed high-confidence semantic neighbors into Store S27 hyperedges

## Sister self-tuning areas

- Add new diagnostic aspects for cross-model tool-call quirks
- Tune negative-constraint formatting density for smaller local models (Gemma 4)
- Convert recurring confusion fixes into deterministic lint code_blocks
