#15: bounded outcome dataset task-cost selection #55

Merged
lost-rob0t merged 7 commits from rage/15-outcome-dataset-task-cost-red into main 2026-09-01 05:20:23 +00:00
lost-rob0t commented 2026-09-01 05:13:16 +00:00 (Migrated from github.com)

Parent: #15

Document-first bounded Auto-ARADR/RAGE slice.

Canonical design: research/LLM-LOG-RESEARCH-052-outcome-dataset-task-cost.org.
RED evidence: evidence/LLM-LOG-EVIDENCE-046-outcome-dataset-task-cost-red.org.
GREEN evidence: evidence/LLM-LOG-EVIDENCE-047-outcome-dataset-task-cost-green.org.

Accepted RED on exact d8f0a86a58a0337ba8757a681a352c30fa6bdf18 / workflow 33472780086: existing contracts stayed green while the new Common Lisp/Tek9/SWI-Prolog contract failed because query_outcome_dataset ignored task-cost selectors.

Implementation reuses #13's bounded query-task-accounting projection. It adds exact task cost state/currency/min/max selectors, caches per-task accounting during one dataset query, rejects truncated accounting, and preserves immutable usage/cost assertion provenance. Numeric amount bounds only match complete known costs in the requested coherent currency. No export-time Prolog, no price recomputation, and task cost never grants outcome authority.

Implementation exact 662acc1ca9099a300a9bff01d26d124d50568b78 / workflow 33472989667 passed Common Lisp/Tek9/SWI-Prolog, Python policy/transport, and Home Manager. Final evidence-only exact head is being independently rerun before merge.

No Tek9 repo, Python capture/transport, dotfiles, provider-census, or configuration PR #5 mutation is included.

Parent: #15 Document-first bounded Auto-ARADR/RAGE slice. Canonical design: `research/LLM-LOG-RESEARCH-052-outcome-dataset-task-cost.org`. RED evidence: `evidence/LLM-LOG-EVIDENCE-046-outcome-dataset-task-cost-red.org`. GREEN evidence: `evidence/LLM-LOG-EVIDENCE-047-outcome-dataset-task-cost-green.org`. Accepted RED on exact `d8f0a86a58a0337ba8757a681a352c30fa6bdf18` / workflow `33472780086`: existing contracts stayed green while the new Common Lisp/Tek9/SWI-Prolog contract failed because `query_outcome_dataset` ignored task-cost selectors. Implementation reuses #13's bounded `query-task-accounting` projection. It adds exact task cost state/currency/min/max selectors, caches per-task accounting during one dataset query, rejects truncated accounting, and preserves immutable usage/cost assertion provenance. Numeric amount bounds only match complete `known` costs in the requested coherent currency. No export-time Prolog, no price recomputation, and task cost never grants outcome authority. Implementation exact `662acc1ca9099a300a9bff01d26d124d50568b78` / workflow `33472989667` passed Common Lisp/Tek9/SWI-Prolog, Python policy/transport, and Home Manager. Final evidence-only exact head is being independently rerun before merge. No Tek9 repo, Python capture/transport, dotfiles, provider-census, or configuration PR #5 mutation is included.
Sign in to join this conversation.
No description provided.