# Stage 128: Dossier Learning Loop (aq128_loop_synthesize_and_apply_improvements / DB-UNIT-13)

Loops across ingested dossier units, deduplicates against already-shipped modules via SimHash-256, and synthesizes prioritized queue tasks and live admin.teleon.dev Todos.

- Atomic micro-queue: `aq128_loop_synthesize_and_apply_improvements` (visibility: `public_commons`)
- Backing database unit: `DB-UNIT-13` (`Polymorphic Solutioning, Harness Packages & 1-Up Efficiency Core`)
- Database cell: `entities-db.teleon.dev` (`teleon_core.solution_artifacts`, `teleon_core.competitive_one_up_challenges`, `teleon_core.faceted_worker_subscriptions`)
- Claim row hydration: `single_self_contained_flat_task_row` (100% context inline, normalized writeback)
- Track 2 Sister queue (`SISTER-13`): `sq13-improve-solutioning-queue.teleon.dev` (Improve working-directory package templates, reward 1-up token reductions, and automate agent-to-code demotion.)
- Lanes: SimHash-256 Novelty Dedup, 3-Pass Loop Synthesis, Admin Todo + Queue Emit

## Execution engines & AI Agent freedom contract

- **Deterministic (0 model calls)**: `openapi_schema_and_rule_factory_solver_v1` (Multi-pass SimHash-256 novelty deduplicator & Todo/queue synthesizer (0 calls))
- **Hybrid (<=1 model call)**: `guided_harness_package_synthesizer_v1`
- **AI Agent & Bot Fleet (1..6 calls)**: `polymorphic_solver_and_one_up_challenger_v1` (Cross-axis architecture upgrader & conformance verifier)
  - Contract: **Instructions + Resources + Freedom to search for new resources and new ways not bound by instructions**

## Database invariants

- Competitive 1-up challengers preserve the incumbent typed input/output contract while lowering token or compute cost.
- Solutions returning further_research_needed atomically enqueue targeted acquisition tasks back to Stage 01.

## External & internal bot fleet actions

- Separate novel architectural ideas from capabilities already shipped in main
- Emit concrete, test-gated engineering Todos and atomic queue tasks onto admin.teleon.dev
- Re-score 10-axis maturity after each implementation wave to close the learning loop

## Sister self-tuning areas

- Expand drop-in working-directory package templates across coding harnesses (DIM-03, DIM-07)
- Tune bounty multipliers for 1-upping high-token solutions into 0-call deterministic nodes (DIM-06, DIM-08)
- Add faceted routing labels for humanitarian goals, countries, ML competitions, math, and physics (DIM-01, DIM-04)