#13: bounded task rollup + restart accounting #39

Merged

lost-rob0t merged 11 commits from rage/13-task-rollup-red into main

2026-08-31 17:15:29 +00:00

lost-rob0t commented

2026-08-31 17:14:54 +00:00

(Migrated from github.com)

Replacement for closed draft #37 because the connected ready-for-review mutation still fails on the invalid Repository.fullDatabaseId selection. #37 had zero reviews and zero conversation comments; this replacement uses the identical branch and exact tested head.

Canonical design: research/LLM-LOG-RESEARCH-041-task-rollup-restart.org.
RED evidence: evidence/LLM-LOG-EVIDENCE-023-task-rollup-restart-red.org.
GREEN evidence: evidence/LLM-LOG-EVIDENCE-024-task-rollup-restart-green.org.

Executed RED: 9976d5239e80de79a92a66a8b4b6fe0d1704f211, run 33413493088.
Exact current head: 1c8a0947759e0941480da03a3966673d1e075205.
Exact-head GREEN: run 33418411885, all three expert-plane lanes successful.

This slice adds declared bounded query_task_accounting in the Common Lisp expert service with Tek9-indexed parent/child traversal, restart reconstruction, immutable usage/request deduplication, token rollups, provenance IDs, explicit truncation, coherent-currency handling, and known/partial/unknown cost state. Per-request price inference remains SWI-Prolog-owned; no Python rule engine and no Tek9 repository mutation.

This does not close #13: retry/attempt analysis and cross-dimensional cost/usage queries by model/provider/client/classification/outcome remain follow-on RED-first work. Dotfiles and ~/Documents/AI/proxy are untouched. PR #5 and provider census #6 remain separate transactions.

Replacement for closed draft #37 because the connected ready-for-review mutation still fails on the invalid `Repository.fullDatabaseId` selection. #37 had zero reviews and zero conversation comments; this replacement uses the identical branch and exact tested head. Canonical design: `research/LLM-LOG-RESEARCH-041-task-rollup-restart.org`. RED evidence: `evidence/LLM-LOG-EVIDENCE-023-task-rollup-restart-red.org`. GREEN evidence: `evidence/LLM-LOG-EVIDENCE-024-task-rollup-restart-green.org`. Executed RED: `9976d5239e80de79a92a66a8b4b6fe0d1704f211`, run `33413493088`. Exact current head: `1c8a0947759e0941480da03a3966673d1e075205`. Exact-head GREEN: run `33418411885`, all three expert-plane lanes successful. This slice adds declared bounded `query_task_accounting` in the Common Lisp expert service with Tek9-indexed parent/child traversal, restart reconstruction, immutable usage/request deduplication, token rollups, provenance IDs, explicit truncation, coherent-currency handling, and known/partial/unknown cost state. Per-request price inference remains SWI-Prolog-owned; no Python rule engine and no Tek9 repository mutation. This does not close #13: retry/attempt analysis and cross-dimensional cost/usage queries by model/provider/client/classification/outcome remain follow-on RED-first work. Dotfiles and `~/Documents/AI/proxy` are untouched. PR #5 and provider census #6 remain separate transactions.