ODF docs
GitHub ↗
Documentation menu

ODF Fast-Lane Baseline — 2026-09-13

2 min read

#Cohort and window

  • Window: 2026-08-31 through 2026-09-13 UTC, using metrics --days 14.
  • Cohort: 95 completed, non-span task calls from the configured daily JSONL telemetry.
  • Raw telemetry lines: 320. This backward-compatible count includes lifecycle markers and spans; it is not the baseline sample size.
  • Duration meaning: recorded phase/task duration_ms, not entry-to-final-gate time. Percentiles use the sorted completed-call durations with linear interpolation.
  • Odoo version/project: not available in the current sanitized telemetry.

#Overall baseline

Sample p50 p95 Average
95 148,073 ms 900,134 ms 287,339 ms

#Outcomes

Outcome Count Rate
ok 26 27.4%
blocked 39 41.1%
error 9 9.5%
timeout 21 22.1%

#Phase latency

Phase Samples p50 p95
ASSESS 12 315,640 ms 735,366 ms
DESIGN 12 542,997 ms 735,064 ms
EXPLORE 1 13,819 ms 13,819 ms
FIX 23 10,289 ms 1,466,558 ms
IMPLEMENT 35 242,673 ms 690,789 ms
PROPOSE 5 83,371 ms 175,124 ms
QA-PLAN 7 240,284 ms 600,373 ms

FIX is a canonical phase label, not proof of an end-to-end /odf-fix run.

#Dimension coverage

  • Work type: 32/95 reported (33.7%): feature 30 and small-change 2; 63 calls are unknown.
  • Model availability: the availability flag is present for 95/95 calls; 0 report an available model and 95 report unavailable. Model identity and profile are not available.
  • Validation: 0/95 completed task calls report validation evidence. There are no scheduler-join records in this cohort, so no validation ratio is available there either.
  • Receipts: 0/95 completed task calls report a receipt reference. Receipt-producing failures cannot be measured from this cohort.
  • Escalation: the current telemetry schema has no escalation dimension. The baseline reports this as unavailable; blocked, error, and timeout outcomes are not treated as escalations.

#End-to-end measurement gap

The current telemetry cannot prove /odf-new or /odf-fix entry-to-final-verified-gate latency. It has phase/task durations and sanitized change/run identifiers, but no bounded entry event, command-entry discriminator, or final verified-gate completion marker that can establish those cohorts. Therefore both end-to-end p50 and p95 are N/A, and this report makes no /odf-new or /odf-fix latency claim.

The baseline command remains:

node scripts/odf-toolkit.js metrics --days 14 --json

The JSON output preserves the existing records and by_phase fields and adds a baseline object. The baseline excludes started markers, spans, and scheduler joins from completed task-call samples. No raw prompts, paths, database names, secrets, usernames, or PII are included.