Repository navigation
APR-OBS-001: dogfood observability (OBS-00 + row checklist) #4487
Copy link
Copy link
Open
Labels
P1High priorityHigh priorityowner:aprender-cdOwned by session aprender-cdOwned by session aprender-cdsize:MK̂ 3-4 hK̂ 3-4 h
Milestone
Description
Row OBS-00 of APR-OBS-001 (spec: docs/specifications/APR-OBS-001-dogfood-observability.md, committed on branch la-71/4487-obs-00). Work, contract and done-when: spec §7 row OBS-00; falsifiers §3; hard rules §6; STOPs §8. Owner: aprender-cd (reassigned by the cop 2026-10-05; was aprender-60). Minted by the cop (single-minter). Report per §9.
APR-OBS-001 rows (family folded here per ISSUE-TREE-001, 2026-10-04)
Each row was its own issue; it is closed into this checklist and its body is kept in the closed issue.
- APR-OBS-001 OBS-01: lane rows (apr-lane-row-v1) per review-lane call #4488 — OBS-01: lane rows (apr-lane-row-v1) per review-lane call
- APR-OBS-001 OBS-02: nightly liveness: five-whys on cancelled cuda-nightly/qwen-story-daily; NIGHTLY-RED counts cancels #4489 — OBS-02: nightly liveness: five-whys on cancelled cuda-nightly/qwen-story-daily; NIGHTLY-RED counts cancels
- APR-OBS-001 OBS-03: serve timings on every backend (SRV-TIM-001) #4490 — OBS-03: serve timings on every backend (SRV-TIM-001)
- APR-OBS-001 OBS-04: self-report agreement (join lane + serve rows by request_id) #4491 — OBS-04: self-report agreement (join lane + serve rows by request_id)
- APR-OBS-001 OBS-05: nightly apr-vs-llama perf ledger (closes infra#1057) #4492 — OBS-05: nightly apr-vs-llama perf ledger (closes infra#1057)
- PERF-HIST-001 = build APR-OBS OBS-05 (apr-perf-ledger-v1): schema + checks; no second table #4551 — PERF-HIST-001 = build APR-OBS OBS-05 (apr-perf-ledger-v1): schema + checks; no second table
- APR-OBS-001 OBS-06: RED rules over the ledger (report-only through 0.70) #4493 — OBS-06: RED rules over the ledger (report-only through 0.70)
- APR-OBS-001 OBS-07: release-tag baseline (single writer: release train) #4494 — OBS-07: release-tag baseline (single writer: release train)
- APR-OBS-001 OBS-08: one view rendered from ledgers only #4495 — OBS-08: one view rendered from ledgers only
- APR-OBS-001 OBS-09: serve tracing: X-Trace-Level runs inference_trace (first row of 0.71 serve epic) #4496 — OBS-09: serve tracing: X-Trace-Level runs inference_trace (first row of 0.71 serve epic)
- APR-OBS-001 OBS-10: loop admission gate for speed-signal consumers #4497 — OBS-10: loop admission gate for speed-signal consumers
- APR-OBS-001 OBS-11: 0.71 perf PRs carry before/after apr-trace-v1 diff #4498 — OBS-11: 0.71 perf PRs carry before/after apr-trace-v1 diff
- APR-OBS-001 OBS-12: hourly probe decision report after 14 nights #4499 — OBS-12: hourly probe decision report after 14 nights
- APR-OBS-001 OBS-13: serve tier #4572 — OBS-13: serve tier
- APR-OBS-001 OBS-14: epochs + backfill #4573 — OBS-14: epochs + backfill
- APR-OBS-001 OBS-15: kernel path #4574 — OBS-15: kernel path
- APR-OBS-001 OBS-18: 0.73 hosts #4575 — OBS-18: 0.73 hosts
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
P1High priorityHigh priorityowner:aprender-cdOwned by session aprender-cdOwned by session aprender-cdsize:MK̂ 3-4 hK̂ 3-4 h
Activity
docs(lookahead-0.73): align P1/BPM/R3 with unmerged OBS identity + ke…
lookahead(0.71): DoR rows 1/5/7/10 restaged on la-71/4487-obs-00 (apr…
fix(contracts): OBS-00 formals carry Σ symbols, so pv lint's sigma ga…
lookahead(0.71): OBS-00 baseline = contracts/lint-baseline.json sigma…
docs(lookahead): 0.71 DoR - OBS-09 cites #4487's apr-trace-v1, owner …
docs(lookahead): 0.71 DoR - cite the PP-LLAMA head with the probe rul…
docs(lookahead): 0.71 epic hygiene audit - section 6 epic bullet not …
docs(lookahead): epic hygiene - ground H-1 in rules 2 and 6, H-7 in t…
docs(lookahead): DoR first-5 PRs follow the like-kind batching rule
Merge origin/main 316dee2 into la-71/4487-obs-00 for the OBS-00 regen
chore(contracts): regen for OBS-00 — contracts.nt, ONT witness, READM…
[-]APR-OBS-001 OBS-00: schemas + identity block as pv contracts; file infra fleet_perf_dir issue[/-][+]APR-OBS-001: dogfood observability (OBS-00 + row checklist)[/+]9 remaining items
chore(contracts): regen for OBS-00 after the main merge — contracts.n…
noahgift commentedon Oct 9, 2026
OBS-03 (SRV-TIM-001), a fact found while measuring #4483. No action is taken here; the row stays with its owner.
main(acd599e, release 0.70.3),crates/aprender-serve/src/api/qwen35_chat_backend.rssends notimingsblock.git log -S timingson that file is empty. A PRM-S1 v2 run against a main-builtaprrecordsprompt_ms: nullon every Qwen3.5 row. Measured 2026-10-09 on an RTX 4090.caf6dfa9c8(Qwen3.5 chat, stream and non-stream),fb88f00e34(every chat backend) andec5c4dba25(/v1/completions). None of them is an ancestor ofmainor of anyv0.70.xtag.[qwen35] batched prefilllog lines and says so.Agent: aprender-wprodmodels
feat(serve): X-Trace-Level step|layer runs the tracer, and says measu…
chore(contracts): regen for the OBS-00 fold on main — contracts.nt, O…
noahgift commentedon Oct 12, 2026
OBS-00 fold: branch
fold/4487-obs-00@ 0f1cc98 carries 7 commits picked fromla-71/4487-obs-00onto main 03d4af1, plus one fresh regen. The contracts gate is 8/8 green on it.Still owed: OBS-TRACE-003, the trace ledger row rule. The branch's
contracts/apr-trace-v1.yamlwas an older schema stub. Main already has a liveapr-trace-v1.yamlfor OBS-09, cited bycrates/aprender-serve/src/api/serve_trace.rs, which discharges FALSIFY-OBS-TRACE-001/002 with real tests. The fold keeps main's file. The one part of the stub that main does not have is still owed and must come with the trace-ledger writer PR:row_admissible: atrace.jsonlrow carries the §2.1 identity block,schema = "apr-trace-v1", the keys{trace_level, provenance, wall_ms, steps, payload_path, payload_sha256}, andsha256(read(payload_path)) = payload_sha256;payload_sha256does not match its side file is inadmissible;The stub text is at
la-71/4487-obs-00:contracts/apr-trace-v1.yaml.🤖 Generated with Claude Code
noahgift commentedon Oct 14, 2026
merged into #5305 (master ticket for epic #3598; triage rule 7b, 2026-10-14). It is a checklist row there; reopen this issue when work on it starts or its PR opens.
noahgift commentedon Oct 14, 2026
Moved: this sub-ticket is now a row in #5322 (Speed measurement harness: resource receipts, observability and the ollama/llama.cpp ratio), split out of #5305 by theme.