nsaspy/llm-log
1
0
Fork 0
Code Issues 34 Pull requests 19 Projects Releases Packages Wiki Activity Actions

Expert: task graph + per-task token/cost accounting #13

New issue
Closed
opened 2026-08-29 12:49:40 +00:00 by lost-rob0t · 9 comments
lost-rob0t commented 2026-08-29 12:49:40 +00:00 (Migrated from github.com)
Copy link

Parent: #8
Depends on: #10

Goal

Build a task-tracking expert that turns raw LLM traffic into explicit task/subtask graphs and reproducible usage/cost accounting.

Task model

Persist stable Tek9 entities/edges for:

  • task
  • parent task / child task
  • originating user message
  • session
  • request / response / websocket turn
  • provider / model
  • usage observation
  • pricing snapshot
  • cost assertion

A task may span multiple provider requests, retries, rewrites, tools, models, and child tasks.

Identity rules

Use explicit task IDs from clients when available. Otherwise derive/assign stable local task IDs from observed message/session boundaries and expert rules. Never silently merge unrelated work solely because it shares a chat/session.

Store why a request was attached to a task.

Usage

Capture and normalize when available:

  • input tokens
  • output tokens
  • cached input/output tokens
  • reasoning tokens
  • provider-specific billable token classes
  • request count
  • wall-clock/provider latency
  • transport/client/provider/model

Missing usage remains unknown.

Pricing/cost

Costs must be reproducible:

  • persist the exact pricing snapshot/version/source used;
  • compute request cost from normalized usage + that snapshot;
  • aggregate per task and recursively across child tasks;
  • preserve original currency/unit;
  • unknown or ambiguous pricing => cost unknown, never guessed;
  • historical task totals must not silently change because today's provider price changed.

Analysis queries

Support at least:

  • cost/usage for task
  • task tree with child rollups
  • cost by model/provider/client
  • cost by request classification
  • cost by success/failure label
  • expensive failures/retries
  • token burn before successful completion

RED-first

Fixtures must prove baseline cannot correctly account for:

  • multiple requests under one task;
  • parent/child task rollup;
  • retry cost;
  • unknown pricing;
  • historical pricing snapshot stability.

Acceptance

  • one synthetic task with multiple requests aggregates exactly;
  • child task costs roll into parent without double-counting;
  • unknown price remains unknown while usage remains queryable;
  • every cost assertion links usage observation + pricing snapshot + rule/version;
  • task assignment carries provenance;
  • restart preserves totals through Tek9.
Parent: #8 Depends on: #10 ## Goal Build a task-tracking expert that turns raw LLM traffic into explicit task/subtask graphs and reproducible usage/cost accounting. ## Task model Persist stable Tek9 entities/edges for: - task - parent task / child task - originating user message - session - request / response / websocket turn - provider / model - usage observation - pricing snapshot - cost assertion A task may span multiple provider requests, retries, rewrites, tools, models, and child tasks. ## Identity rules Use explicit task IDs from clients when available. Otherwise derive/assign stable local task IDs from observed message/session boundaries and expert rules. Never silently merge unrelated work solely because it shares a chat/session. Store why a request was attached to a task. ## Usage Capture and normalize when available: - input tokens - output tokens - cached input/output tokens - reasoning tokens - provider-specific billable token classes - request count - wall-clock/provider latency - transport/client/provider/model Missing usage remains unknown. ## Pricing/cost Costs must be reproducible: - persist the exact pricing snapshot/version/source used; - compute request cost from normalized usage + that snapshot; - aggregate per task and recursively across child tasks; - preserve original currency/unit; - unknown or ambiguous pricing => cost `unknown`, never guessed; - historical task totals must not silently change because today's provider price changed. ## Analysis queries Support at least: - cost/usage for task - task tree with child rollups - cost by model/provider/client - cost by request classification - cost by success/failure label - expensive failures/retries - token burn before successful completion ## RED-first Fixtures must prove baseline cannot correctly account for: - multiple requests under one task; - parent/child task rollup; - retry cost; - unknown pricing; - historical pricing snapshot stability. ## Acceptance - one synthetic task with multiple requests aggregates exactly; - child task costs roll into parent without double-counting; - unknown price remains unknown while usage remains queryable; - every cost assertion links usage observation + pricing snapshot + rule/version; - task assignment carries provenance; - restart preserves totals through Tek9.
lost-rob0t commented 2026-08-31 15:32:15 +00:00 (Migrated from github.com)
Copy link

Bounded #13 substrate merged via #35 at bee9fa8dd3ad1d7ff375359104a48fcfb5ce99b0 after exact-head GREEN on 982f4dea839f459db04799919a55e0389b1b0ac9 (workflow 33408766713). Canonical design remains research/LLM-LOG-RESEARCH-040-task-graph-usage-cost.org; RED/GREEN evidence is in EVIDENCE-021/022.

This slice establishes typed task/usage/pricing/cost materialization in Common Lisp/Tek9, declared SWI-Prolog task_cost, explicit unknown pricing, immutable pricing provenance, and bounded request deduplication. No Tek9 repo or dotfiles mutation.

#13 stays open. Remaining acceptance is RED-first: recursive parent/child rollups without double counting, restart-preserved totals/query surfaces, retries, and analysis by model/provider/client/classification/outcome including expensive failures and token burn before success.

Bounded #13 substrate merged via #35 at `bee9fa8dd3ad1d7ff375359104a48fcfb5ce99b0` after exact-head GREEN on `982f4dea839f459db04799919a55e0389b1b0ac9` (workflow 33408766713). Canonical design remains `research/LLM-LOG-RESEARCH-040-task-graph-usage-cost.org`; RED/GREEN evidence is in EVIDENCE-021/022. This slice establishes typed task/usage/pricing/cost materialization in Common Lisp/Tek9, declared SWI-Prolog `task_cost`, explicit unknown pricing, immutable pricing provenance, and bounded request deduplication. No Tek9 repo or dotfiles mutation. #13 stays open. Remaining acceptance is RED-first: recursive parent/child rollups without double counting, restart-preserved totals/query surfaces, retries, and analysis by model/provider/client/classification/outcome including expensive failures and token burn before success.
lost-rob0t commented 2026-08-31 17:15:47 +00:00 (Migrated from github.com)
Copy link

Bounded #13 rollup/restart slice merged via #39 at e33ff07889b0095ed9a0f12084d098cbdc3fda44 from exact tested head 1c8a0947759e0941480da03a3966673d1e075205 (run 33418411885 GREEN across CL/Tek9/SWI, Python expert-policy/transport, and Home Manager). Canonical design: research/LLM-LOG-RESEARCH-041-task-rollup-restart.org; RED/GREEN evidence: evidence/LLM-LOG-EVIDENCE-023-task-rollup-restart-red.org and ...024-task-rollup-restart-green.org.

Keep #13 OPEN. Remaining bounded acceptance: retry/attempt accounting and analysis/query dimensions by model/provider/client/classification/outcome, including expensive failures and token burn before success. No Tek9 or dotfiles mutation was needed for this slice.

Bounded #13 rollup/restart slice merged via #39 at `e33ff07889b0095ed9a0f12084d098cbdc3fda44` from exact tested head `1c8a0947759e0941480da03a3966673d1e075205` (run `33418411885` GREEN across CL/Tek9/SWI, Python expert-policy/transport, and Home Manager). Canonical design: `research/LLM-LOG-RESEARCH-041-task-rollup-restart.org`; RED/GREEN evidence: `evidence/LLM-LOG-EVIDENCE-023-task-rollup-restart-red.org` and `...024-task-rollup-restart-green.org`. Keep #13 OPEN. Remaining bounded acceptance: retry/attempt accounting and analysis/query dimensions by model/provider/client/classification/outcome, including expensive failures and token burn before success. No Tek9 or dotfiles mutation was needed for this slice.
lost-rob0t commented 2026-08-31 18:14:42 +00:00 (Migrated from github.com)
Copy link

Bounded #13 retry/attempt slice reached an accepted RED gate.

Canonical design: research/LLM-LOG-RESEARCH-042-task-retry-attempt-accounting.org
Evidence: evidence/LLM-LOG-EVIDENCE-025-task-retry-attempt-red.org
Draft PR: #41

Executed RED head: fdc351217621efa6efafc4a7131b1753b2bb0828, workflow 33423647092. Existing #10/#12/#13 integration behavior, Python expert-policy/transport, and Home Manager remained green. The new restart fixture preserved 3 requests, 240 input tokens, 120 output tokens, and 0.48 USD, but attempt_count, retry_request_count, attempt_ids, and retry_request_ids were absent. This verifies that current production bills distinct retry requests but drops explicit attempt/retry identity from the durable usage projection.

No production mutation was made in this bounded cycle. Next slice is limited to Common Lisp validation/persistence of explicit attempt/retry metadata plus bounded Tek9 restart reconstruction through the existing task/usage index path. No Python rule engine, outcome dependency, corpus scan, Tek9 repo change, or dotfiles change.

Bounded #13 retry/attempt slice reached an accepted RED gate. Canonical design: `research/LLM-LOG-RESEARCH-042-task-retry-attempt-accounting.org` Evidence: `evidence/LLM-LOG-EVIDENCE-025-task-retry-attempt-red.org` Draft PR: #41 Executed RED head: `fdc351217621efa6efafc4a7131b1753b2bb0828`, workflow 33423647092. Existing #10/#12/#13 integration behavior, Python expert-policy/transport, and Home Manager remained green. The new restart fixture preserved 3 requests, 240 input tokens, 120 output tokens, and 0.48 USD, but `attempt_count`, `retry_request_count`, `attempt_ids`, and `retry_request_ids` were absent. This verifies that current production bills distinct retry requests but drops explicit attempt/retry identity from the durable usage projection. No production mutation was made in this bounded cycle. Next slice is limited to Common Lisp validation/persistence of explicit attempt/retry metadata plus bounded Tek9 restart reconstruction through the existing task/usage index path. No Python rule engine, outcome dependency, corpus scan, Tek9 repo change, or dotfiles change.
lost-rob0t commented 2026-08-31 19:22:26 +00:00 (Migrated from github.com)
Copy link

Bounded #13 retry/attempt slice merged via PR #42.

Canonical design: research/LLM-LOG-RESEARCH-042-task-retry-attempt-accounting.org.
Evidence: evidence/LLM-LOG-EVIDENCE-025-task-retry-attempt-red.org and evidence/LLM-LOG-EVIDENCE-026-task-retry-attempt-green.org.

RED/GREEN chain:

  • primary accepted RED: fdc351217621efa6efafc4a7131b1753b2bb0828, run 33423647092 — durable task accounting preserved 3 requests / 240 input / 120 output / 0.48 USD but attempt/retry metadata was absent after restart;
  • initial implementation: 184c5963b921775d783d105bcea6094522751fa1, run 33429223078 GREEN;
  • adversarial migration RED: 01954012c761524bea5cabcfffb7e6222e235da5, run 33429724537 — legacy immutable replay failed specifically with usage_conflict: usage:usage-retry-legacy because absent optional metadata keys changed projection identity;
  • corrected implementation: 1d8e50dca6a885b601681d3ac65130311eb4e71c, run 33429869506 GREEN across Common Lisp/Tek9/SWI, Python policy/transport, and Home Manager;
  • final evidence-only exact head: beaab8aa2806d6bb2a7081b6b579a77e781d551c, run 33429992265 GREEN across all three lanes;
  • merged PR #42 with expected-head protection; merge commit 3e6ba6775623f9d59e8f67f9f984402e1f38cbbc.

Verified behavior: Common Lisp validates and immutably persists explicit attempt identity / ordinal / retry predecessor when supplied; bounded query_task_accounting reconstructs attempt/retry IDs and counts after restart using the existing Tek9 task/usage indexes; missing attempt metadata stays explicit and does not suppress usage/cost; retries are never guessed from timestamps/session adjacency; old usage projections replay idempotently after upgrade. SWI-Prolog remains pricing-inference owner. No Python retry rule engine, Tek9 repo mutation, or dotfiles mutation was introduced.

Keep #13 OPEN. Remaining acceptance is the cross-task analysis/query slice by model/provider/client/classification/outcome, including expensive failures/retries and token burn before eventual success. Outcome-dependent analysis should respect #15's evidence-vs-label boundary rather than smuggling transport success into task outcome.

Bounded #13 retry/attempt slice merged via PR #42. Canonical design: `research/LLM-LOG-RESEARCH-042-task-retry-attempt-accounting.org`. Evidence: `evidence/LLM-LOG-EVIDENCE-025-task-retry-attempt-red.org` and `evidence/LLM-LOG-EVIDENCE-026-task-retry-attempt-green.org`. RED/GREEN chain: - primary accepted RED: `fdc351217621efa6efafc4a7131b1753b2bb0828`, run `33423647092` — durable task accounting preserved 3 requests / 240 input / 120 output / 0.48 USD but attempt/retry metadata was absent after restart; - initial implementation: `184c5963b921775d783d105bcea6094522751fa1`, run `33429223078` GREEN; - adversarial migration RED: `01954012c761524bea5cabcfffb7e6222e235da5`, run `33429724537` — legacy immutable replay failed specifically with `usage_conflict: usage:usage-retry-legacy` because absent optional metadata keys changed projection identity; - corrected implementation: `1d8e50dca6a885b601681d3ac65130311eb4e71c`, run `33429869506` GREEN across Common Lisp/Tek9/SWI, Python policy/transport, and Home Manager; - final evidence-only exact head: `beaab8aa2806d6bb2a7081b6b579a77e781d551c`, run `33429992265` GREEN across all three lanes; - merged PR #42 with expected-head protection; merge commit `3e6ba6775623f9d59e8f67f9f984402e1f38cbbc`. Verified behavior: Common Lisp validates and immutably persists explicit attempt identity / ordinal / retry predecessor when supplied; bounded `query_task_accounting` reconstructs attempt/retry IDs and counts after restart using the existing Tek9 task/usage indexes; missing attempt metadata stays explicit and does not suppress usage/cost; retries are never guessed from timestamps/session adjacency; old usage projections replay idempotently after upgrade. SWI-Prolog remains pricing-inference owner. No Python retry rule engine, Tek9 repo mutation, or dotfiles mutation was introduced. Keep #13 OPEN. Remaining acceptance is the cross-task analysis/query slice by model/provider/client/classification/outcome, including expensive failures/retries and token burn before eventual success. Outcome-dependent analysis should respect #15's evidence-vs-label boundary rather than smuggling transport success into task outcome.
lost-rob0t commented 2026-08-31 20:27:06 +00:00 (Migrated from github.com)
Copy link

Bounded #13 provider/model/client analytics slice completed and merged via #43.

Canonical artifacts:

  • research/LLM-LOG-RESEARCH-043-task-usage-breakdowns.org
  • evidence/LLM-LOG-EVIDENCE-027-task-usage-breakdown-red.org
  • evidence/LLM-LOG-EVIDENCE-028-task-usage-breakdown-green.org

Accepted RED: 9f3e54909fa3d9527bc7fc658c9598e9ac7d193f, run 33435473703. Existing aggregate/retry contracts were green after restart; the new fixture failed specifically because query_task_accounting had no breakdowns projection.

Exact evidence-only GREEN head: 4eaabb3ddc60fc288819d4f4bffb44b89ec708f4, run 33435931708, all Common Lisp/Tek9/SWI-Prolog, Python expert-policy/transport, and Home Manager lanes green. Merged as 2c8a07fb5031cb7175925c41bd7d06db6a14070e.

Verified restart-durable breakdowns now cover provider/model/client with bounded task-rooted retrieval, token/request counts, known/partial/unknown pricing state, and usage/cost provenance. No Tek9 repo or dotfiles mutation was required.

Keep #13 open: classification breakdowns still need a provenance-preserving join to #12 assertions, and outcome/expensive-failure/token-burn-before-success analytics must wait for #15's explicit outcome evidence/label semantics rather than treating transport success as task success.

Bounded #13 provider/model/client analytics slice completed and merged via #43. Canonical artifacts: - `research/LLM-LOG-RESEARCH-043-task-usage-breakdowns.org` - `evidence/LLM-LOG-EVIDENCE-027-task-usage-breakdown-red.org` - `evidence/LLM-LOG-EVIDENCE-028-task-usage-breakdown-green.org` Accepted RED: `9f3e54909fa3d9527bc7fc658c9598e9ac7d193f`, run 33435473703. Existing aggregate/retry contracts were green after restart; the new fixture failed specifically because `query_task_accounting` had no `breakdowns` projection. Exact evidence-only GREEN head: `4eaabb3ddc60fc288819d4f4bffb44b89ec708f4`, run 33435931708, all Common Lisp/Tek9/SWI-Prolog, Python expert-policy/transport, and Home Manager lanes green. Merged as `2c8a07fb5031cb7175925c41bd7d06db6a14070e`. Verified restart-durable breakdowns now cover provider/model/client with bounded task-rooted retrieval, token/request counts, known/partial/unknown pricing state, and usage/cost provenance. No Tek9 repo or dotfiles mutation was required. Keep #13 open: classification breakdowns still need a provenance-preserving join to #12 assertions, and outcome/expensive-failure/token-burn-before-success analytics must wait for #15's explicit outcome evidence/label semantics rather than treating transport success as task success.
lost-rob0t commented 2026-08-31 21:29:27 +00:00 (Migrated from github.com)
Copy link

Bounded classification-analytics slice merged via #44.

Canonical design: research/LLM-LOG-RESEARCH-044-task-classification-breakdowns.org
RED evidence: evidence/LLM-LOG-EVIDENCE-029-task-classification-breakdown-red.org
GREEN evidence: evidence/LLM-LOG-EVIDENCE-030-task-classification-breakdown-green.org

Exact evidence head c30920ef00153e247deb7a1e454ce2e6bfe9035b passed Common Lisp/Tek9/SWI-Prolog, Python expert policy/transport, and Home Manager in workflow 33441459372, then #44 merged with expected-head protection as merge commit 79cac2906f2a70764b021d6c3c1fe2107601cc71.

Now verified after expert-host restart: task-rooted classification breakdowns are joined through bounded Tek9 request/source/assertion indexes; multi-label and ambiguous states remain explicit; usage/cost/classifier assertion/rule/evidence provenance is preserved. No Tek9 repo mutation, Python rule engine, outcome inference, or dotfiles change.

#13 remains open. Provider/model/client/classification analytics are now covered. Outcome/failure/partial breakdowns, expensive-failure analysis, and token burn before eventual success must consume #15's explicit outcome evidence/label semantics rather than infer success from transport status.

Bounded classification-analytics slice merged via #44. Canonical design: `research/LLM-LOG-RESEARCH-044-task-classification-breakdowns.org` RED evidence: `evidence/LLM-LOG-EVIDENCE-029-task-classification-breakdown-red.org` GREEN evidence: `evidence/LLM-LOG-EVIDENCE-030-task-classification-breakdown-green.org` Exact evidence head `c30920ef00153e247deb7a1e454ce2e6bfe9035b` passed Common Lisp/Tek9/SWI-Prolog, Python expert policy/transport, and Home Manager in workflow 33441459372, then #44 merged with expected-head protection as merge commit `79cac2906f2a70764b021d6c3c1fe2107601cc71`. Now verified after expert-host restart: task-rooted classification breakdowns are joined through bounded Tek9 request/source/assertion indexes; multi-label and ambiguous states remain explicit; usage/cost/classifier assertion/rule/evidence provenance is preserved. No Tek9 repo mutation, Python rule engine, outcome inference, or dotfiles change. #13 remains open. Provider/model/client/classification analytics are now covered. Outcome/failure/partial breakdowns, expensive-failure analysis, and token burn before eventual success must consume #15's explicit outcome evidence/label semantics rather than infer success from transport status.
lost-rob0t commented 2026-09-01 01:14:28 +00:00 (Migrated from github.com)
Copy link

Bounded #13 outcome-breakdown slice merged via PR #50.

Canonical design: research/LLM-LOG-RESEARCH-048-task-outcome-breakdowns.org.
Accepted RED: evidence/LLM-LOG-EVIDENCE-038-task-outcome-breakdown-red.org at 54dd2d97263a6b59c340c8bd47ce9160ae4ad255 / run 33456066504.
Implementation GREEN: 8f29d0c153efb3ddbb0fac21c31de00a4de124d0 / run 33457650079.
Evidence-only exact head a1e7ed2b951d50e1d04ced24b5d49915b63c5222 / run 33457811497 passed Common Lisp/Tek9/SWI-Prolog, Python expert policy/transport, and Home Manager before expected-head merge.

Verified: bounded request-local join to current stored #15 outcome assertions; explicit unknown stays distinct from unlabeled; superseded assertions are excluded; multiple current assertions fail as integrity errors; usage/cost plus outcome assertion/rule/evidence provenance survives restart. No transport-status outcome inference, no Python rule engine, no Tek9 repo mutation, no dotfiles mutation.

Keep #13 open. Next bounded analytics boundary is expensive-failure/retry and token-burn-before-eventual-success analysis using explicit #15 labels plus existing retry/cost projections.

Bounded #13 outcome-breakdown slice merged via PR #50. Canonical design: `research/LLM-LOG-RESEARCH-048-task-outcome-breakdowns.org`. Accepted RED: `evidence/LLM-LOG-EVIDENCE-038-task-outcome-breakdown-red.org` at `54dd2d97263a6b59c340c8bd47ce9160ae4ad255` / run 33456066504. Implementation GREEN: `8f29d0c153efb3ddbb0fac21c31de00a4de124d0` / run 33457650079. Evidence-only exact head `a1e7ed2b951d50e1d04ced24b5d49915b63c5222` / run 33457811497 passed Common Lisp/Tek9/SWI-Prolog, Python expert policy/transport, and Home Manager before expected-head merge. Verified: bounded request-local join to current stored #15 outcome assertions; explicit `unknown` stays distinct from `unlabeled`; superseded assertions are excluded; multiple current assertions fail as integrity errors; usage/cost plus outcome assertion/rule/evidence provenance survives restart. No transport-status outcome inference, no Python rule engine, no Tek9 repo mutation, no dotfiles mutation. Keep #13 open. Next bounded analytics boundary is expensive-failure/retry and token-burn-before-eventual-success analysis using explicit #15 labels plus existing retry/cost projections.
lost-rob0t commented 2026-09-01 02:22:00 +00:00 (Migrated from github.com)
Copy link

#13 acceptance is now complete after merged PR #51 (d46bfee8a9cc66095cb102a5ceeb75e1f46f6bcd).

Canonical final slice: research/LLM-LOG-RESEARCH-049-outcome-retry-economics.org; RED/GREEN evidence: evidence/LLM-LOG-EVIDENCE-040-task-outcome-retry-economics-red.org and ...041-task-outcome-retry-economics-green.org.

Final RED: da000c95c0f466158fce22e615d00278c9a9604b, run 33461795855: existing task/retry/outcome behavior, Python expert policy/transport, and Home Manager were green; the new contract failed only because query_task_accounting.result.analysis was absent.

Tightened GREEN: f2cb7ca9f182767c0ced05c6daddfd13e71bf4ea, run 33462054052; final evidence-only exact head acb52ff6993b90bc90bc20423d79d3de53e04435, run 33462157182, green across Common Lisp/Tek9/SWI-Prolog, Python expert policy/transport, and Home Manager.

The completed task-accounting surface now covers bounded task/child rollups, restart durability, retries/attempts, immutable pricing provenance and unknown-price semantics, provider/model/client/classification/outcome breakdowns, explicit expensive-unsuccessful/retry projections, and token/cost burn before successful completion. “Before success” is strict explicit retry_of_request_id ancestry only; HTTP status/timestamps/session adjacency never grant outcome authority. No Tek9 repo mutation or Python rule engine was required.

#13 acceptance is now complete after merged PR #51 (`d46bfee8a9cc66095cb102a5ceeb75e1f46f6bcd`). Canonical final slice: `research/LLM-LOG-RESEARCH-049-outcome-retry-economics.org`; RED/GREEN evidence: `evidence/LLM-LOG-EVIDENCE-040-task-outcome-retry-economics-red.org` and `...041-task-outcome-retry-economics-green.org`. Final RED: `da000c95c0f466158fce22e615d00278c9a9604b`, run `33461795855`: existing task/retry/outcome behavior, Python expert policy/transport, and Home Manager were green; the new contract failed only because `query_task_accounting.result.analysis` was absent. Tightened GREEN: `f2cb7ca9f182767c0ced05c6daddfd13e71bf4ea`, run `33462054052`; final evidence-only exact head `acb52ff6993b90bc90bc20423d79d3de53e04435`, run `33462157182`, green across Common Lisp/Tek9/SWI-Prolog, Python expert policy/transport, and Home Manager. The completed task-accounting surface now covers bounded task/child rollups, restart durability, retries/attempts, immutable pricing provenance and unknown-price semantics, provider/model/client/classification/outcome breakdowns, explicit expensive-unsuccessful/retry projections, and token/cost burn before successful completion. “Before success” is strict explicit `retry_of_request_id` ancestry only; HTTP status/timestamps/session adjacency never grant outcome authority. No Tek9 repo mutation or Python rule engine was required.
nsaspy commented 2026-09-20 01:20:03 +00:00
Owner
Copy link

Extension requirement from #98: add stable session identity and session-level cost/token rollups to the existing task accounting substrate. Session analytics must remain restart/backfill durable, preserve immutable pricing provenance, and use bounded indexed queries rather than whole-corpus scans.

Extension requirement from #98: add stable session identity and session-level cost/token rollups to the existing task accounting substrate. Session analytics must remain restart/backfill durable, preserve immutable pricing provenance, and use bounded indexed queries rather than whole-corpus scans.
Sign in to join this conversation.
No Branch/Tag specified
Branches Tags
main
feat/108-full-kb-mcp-20260920
feat/mcp-expert-kb-20260920
feat/105-opencode-profile
feat/104-python-admission
feat/starworld-rsi
feat/58-structured-output-anomaly
feat/58-tool-schema-anomaly
feat/58-tool-call-anomaly
fix/61-alert-policy
fix/61-retry-routing-correlation
fix/61-transport-error-evidence
fix/68-expert-response-ingest
fix/67-stream-integrity
feature/cl-native-bulk-infill
feature/expert-backfill-query-export
feat/79-quota-telemetry
feat/79-provider-quota-telemetry
fix/74-relay-socket-ownership
feat/stats-analytics-api
feat/71-rate-profile-foundation
feat/75-rate-admission
feat/70-request-scheduler
rage/59-init-files-r3
fix/websocket-expert-plane
rage/58-60-quant-anomaly-alerts-r4
rage/58-60-quant-anomaly-alerts-r3
rage/57-openrouter-quant-observability-r2
rage/58-60-quant-anomaly-alerts-r2
rage/59-init-files-r2
rage/58-60-quant-anomaly-alerts
rage/59-init-files
rage/57-openrouter-quant-observability
rage/openrouter-quantization-presets
noop-unused
rage/15-outcome-dataset-task-cost-red
rage/15-outcome-dataset-classification-red
rage/15-outcome-dataset-metadata-red
rage/13-outcome-retry-economics-red
rage/13-outcome-breakdown-red
rage/15-outcome-dataset-red
rage/15-outcome-history-red
rage/15-outcome-evidence-red
rage/13-classification-breakdown-red
rage/13-usage-breakdown-red
rage/13-retry-attempt-red
rage/13-task-rollup-red
rage/13-task-accounting-red
rage/12-classifier-task-id-red
rage/12-classifier-time-range-red
rage/12-classifier-query-red
rage/12-request-classifier-red
rage/10-expert-runtime-red
rage/10-recorder-deps-red
rage/4-unified-config
rage/home-manager-service
rage/1-capture-proxy-prolog-kb
No results found.
Labels
Clear labels   
accessibility
Barrier affecting people with disabilities

  
bug
Something isn't working

  
documentation
Improvements or additions to documentation

  
duplicate
This issue or pull request already exists

  
enhancement
New feature or request

  
good first issue
Good for newcomers

  
help wanted
Extra attention is needed

  
invalid
This doesn't seem right

  
question
Further information is requested

  
wontfix
This will not be worked on

No labels
accessibility
bug
documentation
duplicate
enhancement
good first issue
help wanted
invalid
question
wontfix
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
2 participants
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
nsaspy/llm-log#13
No description provided.