Commit Graph
118 Commits
Author SHA1 Message Date
Nico bb5d128079 Record completed Bug17 trial costs and false-positive diagnosis 2026-10-20 22:57:56 +02:00
Nico f2b13ae7cd Clarify investigation source contracts and recover nested recall 2026-10-20 22:42:48 +02:00
Nico 84c4f7f96c Record Bug17 source search failures and investigation progress 2026-10-20 22:27:22 +02:00
Nico cb61ac8cd6 Scope budget assessments and recover valid research after recall 2026-10-20 21:59:59 +02:00
Nico 3eb0972395 Record Bug17 assessment rejection after evidence recall 2026-10-20 21:40:27 +02:00
Nico 7b7779b769 Align checkpoint recall validation and resume lean investigation 2026-10-20 21:36:32 +02:00
Nico e92476517c Record Bug17 consolidation recall contract failure 2026-10-20 21:27:37 +02:00
Nico 167d8d5f5e Allow explicit operator resume after technical response repairs 2026-10-20 21:20:46 +02:00
Nico ecec96ce14 Return bounded source excerpts instead of rejecting oversized reads 2026-10-20 21:12:03 +02:00
Nico ab31924b79 Record Bug17 source range rejection and trial metrics 2026-10-20 21:02:55 +02:00
Nico a3e1ea0429 Prepare full bounded Bug17 evaluation and preserve cost baselines 2026-10-20 20:57:58 +02:00
Nico 86a02cb8c0 Record completed bounded contract v2 regression 2026-10-20 20:37:16 +02:00
Nico 424ad95018 Record contract v2 continuation through generation 97 2026-10-20 20:28:57 +02:00
Nico e2d092f1ce Record bounded contract v2 regression metrics 2026-10-20 20:22:20 +02:00
Nico 3a82a37ee6 Stabilize bounded investigation contracts and targeted recovery 2026-10-20 19:59:00 +02:00
Nico ed148b1a50 Record bounded investigation regression metrics and rejected outputs 2026-10-20 17:46:03 +02:00
Nico 682f275590 Bound investigation prompts and paginate research evidence 2026-10-20 17:04:49 +02:00
Nico 377644ffa0 Record startup batch trial metrics and successful scoped repair example 2026-10-20 13:42:12 +02:00
Nico b33208363c Add lean hypothesis startup round with joint evidence review 2026-10-20 12:53:55 +02:00
Nico 3beac3bd1c Archive standalone token usage reports and retention rules 2026-10-20 10:59:33 +02:00
Nico 3872735c0a Recover misplaced joint blocks and preserve unchanged entity citations 2026-10-20 09:50:56 +02:00
Nico c3a4690a55 Recover joint planning dependencies and preserve valid hypothesis siblings 2026-10-20 08:12:23 +02:00
Nico 83ef9e6d6c Add source-constrained joint planning and scoped dependency-aware recovery 2026-10-20 00:43:27 +02:00
Nico f9792586d1 Add joint case planning with independent interpretation and hypothesis recovery 2026-10-19 22:17:00 +02:00
Nico da1ea9c3a3 Capture and safely recover shared LLM responses with bounded field repairs 2026-10-19 20:26:22 +02:00
Nico c6f4ea9877 Ground handoff map values through explicit canonical fact bindings 2026-10-19 17:34:23 +02:00
Nico 061f82e6f1 Reduce verbose investigation and final phase prompt overhead 2026-10-19 16:19:24 +02:00
Nico 2fe1e478a5 Reject impossible format repairs and retain rejected proposals 2026-10-19 10:30:59 +02:00
Nico bfb353eca2 Normalize investigation formats and repair bounded response fragments 2026-10-19 08:21:40 +02:00
Nico f5a87dfca3 Constrain nested bootstrap findings with versioned native output shapes 2026-10-18 23:55:16 +02:00
Nico 67843454e4 Report JSON types and guide malformed investigation finding repair 2026-10-18 23:30:33 +02:00
Nico d957974498 Preserve bootstrap schema error paths for targeted correction 2026-10-18 22:37:56 +02:00
Nico af2ee6cb91 Rank source context using grounded typed anchors and scoped searches 2026-10-18 22:29:22 +02:00
Nico de3a37bfd0 Clarify coarse tool invocation and repair malformed database filters 2026-10-18 21:52:15 +02:00
Nico 20d191bad6 Add bounded source context analysis with lexical follow-up searches 2026-10-18 21:14:05 +02:00
Nico 4b9f263627 Add bounded read-only database relationship analysis profiles 2026-10-18 20:12:00 +02:00
Nico 43ff7555ce Ground coding database values and frame bootstrap JSON responses 2026-10-18 19:11:07 +02:00
Nico 43dca83e74 Use full bootstrap rounds and simplify lean investigation consolidation 2026-10-18 18:09:33 +02:00
Nico dd666026d3 Fix durable lean investigation repair feedback and context validation 2026-10-18 16:25:06 +02:00
Nico ca14271f20 Add lean investigation sessions with deferred consolidation 2026-10-18 16:10:17 +02:00
Nico d010f70f5e Carry investigation decision criteria and offer grounded query narrowing 2026-10-18 13:23:48 +02:00
Nico 29816dc45b Resolve nested tool schema failures into actionable repair feedback 2026-10-18 09:17:24 +02:00
Nico 6a9788114d Focus investigation turns on one question with deterministic result preflight 2026-10-18 00:27:42 +02:00
Nico 0307ebfac0 Configure lower reasoning effort for narrow early investigation reviews 2026-10-17 23:17:32 +02:00
Nico d44563650d Narrow early investigation review and improve safe response repair diagnostics 2026-10-17 20:22:29 +02:00
Nico 8e25bd106d Document compact investigation retry costs and live context growth 2026-10-17 19:38:29 +02:00
Nico d81f2f67f3 Use compact investigation working sets with durable local evidence recall 2026-10-17 18:54:14 +02:00
Nico e8d0602ed3 Allow isolated prompt probes with JSON mode and local schema validation 2026-10-17 13:01:03 +02:00
Nico f52af70646 Capture sanitized HTTP diagnostics for isolated prompt probes 2026-10-17 12:58:25 +02:00
Nico 2eaba70083 Adapt prompt probe schema to Gemini subset with full local validation 2026-10-17 12:55:34 +02:00