How this is built

This repository is built mostly by automated agent runs, merged without human review once CI passes. These are the numbers we steer by: how fast work lands, how often CI catches it, what each run costs, and what is left lying around. Built 2026-10-31 19:23 UTC from 541f0b1cde.

The same numbers as words, every day: the engineering brief, about fifteen minutes to read or listen to, and The Merge Report, a few lighter minutes for everyone else.

Ventures
1,325
Research projects
1,059
Merges, last 24h
304
Branch to merge (p50)
0.7 h
oldest commit to merge, last 7 days
PR open to merge (p50)
0 h
CI green on main
73%
completed push runs
PRs green first try
100%
Open branches
641
CI runs passed, 14 days
105,101
CI runs failed, 14 days
13,762
failure, timeout or startup failure
CI pass rate
88%
of runs that reached a verdict
Workflows red on main
2
latest run on main failed
Merge to live (p50)
87.7 min
last 3 days
Deploys
136
last 3 days

Merges to main per day

2026-10-02: 49 merges2026-10-03: 56 merges2026-10-04: 60 merges2026-10-05: 51 merges2026-10-06: 59 merges2026-10-07: 57 merges2026-10-08: 58 merges2026-10-09: 66 merges2026-10-10: 55 merges2026-10-11: 49 merges2026-10-12: 51 merges2026-10-13: 68 merges2026-10-14: 60 merges2026-10-15: 222 merges2026-10-16: 305 merges2026-10-17: 277 merges2026-10-18: 330 merges2026-10-19: 341 merges2026-10-20: 379 merges2026-10-21: 325 merges2026-10-22: 299 merges2026-10-23: 299 merges2026-10-24: 322 merges2026-10-25: 243 merges2026-10-26: 248 merges2026-10-27: 224 merges2026-10-28: 246 merges2026-10-29: 250 merges2026-10-30: 400 merges2026-10-31: 198 merges 400
Table
DayMerges
2026-10-0249
2026-10-0356
2026-10-0460
2026-10-0551
2026-10-0659
2026-10-0757
2026-10-0858
2026-10-0966
2026-10-1055
2026-10-1149
2026-10-1251
2026-10-1368
2026-10-1460
2026-10-15222
2026-10-16305
2026-10-17277
2026-10-18330
2026-10-19341
2026-10-20379
2026-10-21325
2026-10-22299
2026-10-23299
2026-10-24322
2026-10-25243
2026-10-26248
2026-10-27224
2026-10-28246
2026-10-29250
2026-10-30400
2026-10-31198

Last 30 days: 5657 merges, 0 reverts, 40 fix-forwards of a red main, 748 ventures added. Branch-to-merge time: p50 0.7 h, p90 53.3 h over 7 merge commits (merge commits of the last 7 days: oldest commit on the branch to the merge; squash merges are not visible here).

Pull requests, reviews and CI

PRs merged, 7 days282
PR open to merge, hoursp50 0 · p90 0.3 (n=282)
PRs closed unmerged, 7 days18
First review, hoursp50 not measured · p90 not measured; 0 of 40 sampled PRs had any review
Review verdicts (sampled PRs)none
Open PRs38 (16 drafts); 20 older than 7 days; oldest 32.8 days
CI on main73% green over 200 runs; failing most: Portfolio validate (21), Prompts (15), ventures tests (11)
CI on PRs80% of runs pass; 100% of branches green on their first commit; mean 0 red commits before green (17 branches)
Branches641

Delivery: deploys, merge to live, red time on main

Gate on mainRuns decidedFailedRed spellsTime to restore p50 / p90Hours redShare of window red
Portfolio Pages334198 (59%)220.2 h / 8.85 h46.264%
Portfolio validate (red now)755494 (65%)290.35 h / 2.8 h43.961%

Last 3 days (2026-10-28 19:23 to 2026-10-31 19:23 UTC). 136 deploys (passing portfolio-pages.yml runs on main; 2026-10-28: 0, 2026-10-29: 21, 2026-10-30: 29, 2026-10-31: 86). Merge to live, from a merge commit to the end of the first passing deploy created after it: p50 87.7 min, p75 260.4 min, p90 429.2 min, longest 666.3 min; 492 of 890 merges took over an hour, and 2 are not live yet. A red spell runs from a failing run on main to the next passing one; one still open at the window end counts up to now. A spell that began before the window is counted from its first failing run inside it.

CI runs: passed, failed, cancelled

2026-10-18: 7434 passed2026-10-18: 603 failed2026-10-18: 1824 cancelled2026-10-18: 113 skipped2026-10-19: 9976 passed2026-10-19: 762 failed2026-10-19: 2120 cancelled2026-10-19: 179 skipped2026-10-20: 13841 passed2026-10-20: 1023 failed2026-10-20: 3411 cancelled2026-10-20: 678 skipped2026-10-21: 9638 passed2026-10-21: 628 failed2026-10-21: 1781 cancelled2026-10-21: 289 skipped2026-10-22: 7098 passed2026-10-22: 239 failed2026-10-22: 1100 cancelled2026-10-22: 186 skipped2026-10-23: 7572 passed2026-10-23: 573 failed2026-10-23: 1558 cancelled2026-10-23: 154 skipped2026-10-24: 10383 passed2026-10-24: 1541 failed2026-10-24: 4256 cancelled2026-10-24: 319 skipped2026-10-25: 5982 passed2026-10-25: 1175 failed2026-10-25: 1298 cancelled2026-10-25: 113 skipped2026-10-26: 4972 passed2026-10-26: 1117 failed2026-10-26: 817 cancelled2026-10-26: 101 skipped2026-10-27: 4529 passed2026-10-27: 1119 failed2026-10-27: 964 cancelled2026-10-27: 100 skipped2026-10-28: 5358 passed2026-10-28: 1090 failed2026-10-28: 1233 cancelled2026-10-28: 111 skipped2026-10-29: 6411 passed2026-10-29: 1227 failed2026-10-29: 1012 cancelled2026-10-29: 207 skipped2026-10-30: 8383 passed2026-10-30: 2022 failed2026-10-30: 2790 cancelled2026-10-30: 427 skipped2026-10-31: 3524 passed2026-10-31: 643 failed2026-10-31: 1352 cancelled2026-10-31: 912 skipped 18953

PassedFailedCancelledSkipped

Table
DayPassedFailedCancelledSkipped
2026-10-187,4346031,824113
2026-10-199,9767622,120179
2026-10-2013,8411,0233,411678
2026-10-219,6386281,781289
2026-10-227,0982391,100186
2026-10-237,5725731,558154
2026-10-2410,3831,5414,256319
2026-10-255,9821,1751,298113
2026-10-264,9721,117817101
2026-10-274,5291,119964100
2026-10-285,3581,0901,233111
2026-10-296,4111,2271,012207
2026-10-308,3832,0222,790427
2026-10-313,5246431,352912

148,268 completed workflow runs in the last 14 days: 105,101 passed, 13,762 failed (failure, timeout or startup failure), 25,516 cancelled, 3,889 skipped. 88% of runs that reached a verdict passed. More than 2,500 runs of one outcome in a single hour on the days listed; their counts are a floor: 2026-10-24.

The tables below read the newest 1000 runs (2026-10-31 16:12 to 19:23 UTC, 3.2 h), across 23 workflows. In them 0% were re-runs; a run took p50 0.5 min, p90 2.6 min (run start to last update: wall-clock, including time jobs waited for a runner), and CI used 6.2 hours of run time per hour of clock.

Red on main right now

From the newest 500 push runs on main (2026-10-31 13:09 to 19:20 UTC, 6.2 h). Time for a workflow to pass again on main after it failed there: p50 0.2 h, p90 0.8 h over 15 red spells.

By trigger

TriggerPassedFailedCancelledSkippedPass rate
pull_request308881616678%
push to main1765735076%
pull_request_target29012100%
schedule70010100%
workflow_run00060not measured

Workflows that fail most

WorkflowRunsPassedFailedCancelledSkippedPass rateMinutes p50 / p90
Portfolio validate760552010%0.6 / 0.8
ventures tests64153316031%1.4 / 2.3
Prompts63183312035%0.5 / 0.6
README live links32023900%0.2 / 0.3
Portfolio Pages7454118198%9 / 17.5
Conflict markers89670211100%0.6 / 0.7
Control-plane guard8979091100%0.2 / 0.2
Portfolio Pages site build64440200100%0.8 / 1
Agent cross-review6100061not measurednot measured / not measured
Agent PR auto-merge6000060not measurednot measured / not measured
Doc references53360161100%0.3 / 0.3
Sibling links47340130100%0.5 / 0.6
Link contrast46360100100%0.2 / 0.3
Page integrity46310150100%2.1 / 2.6
Checklist coverage4031090100%0.2 / 0.3
Branch cleanup3330012100%0.5 / 0.8
agent-permission-harness CI2518070100%3.4 / 4.3
Research1211010100%0.5 / 0.6
Duplicate symbols1111000100%0.3 / 0.3
Claude routine trigger1000010not measurednot measured / not measured

The 20 workflows with the most failed runs in the sample, of 23. The same numbers are in stats.json.

Agent runs by track

4251 run records in the last 30 days; 24 integrity issues. Cost, time and leverage come from prompts/runcost.py and exist only for runs it measured; leverage is the agent's own estimate of a human's cost over the API-equivalent cost, so treat it as low-certainty.

TrackRunsSuccessCI fails / runMeasuredAPI $ (median)Active min (median)Leverage (median)
research146896%0.057610.854.5510.4×
ventures138387%5.44100.663.5327.5×
process43177%2.861750.482.1145.7×
redteam33295%not measured1450.422.1213.1×
creative25194%not measured990.653.2572.8×
explore20395%1650.592.7201×
review12394%not measured530.411.7171×
polish4787%not measured80.51.4not measured
perf1377%not measured20.459.3not measured

Agent runs by model

ModelRunsSuccessCI fails / runMeasuredAPI $ (median)Active min (median)Leverage (median)
claude-sonnet-5-5395090%0.0116620.693.3345.5×
unknown18492%4.50not measurednot measurednot measured
claude-opus-5-511498%2.71553.4615184.1×
null250%not measured0not measurednot measurednot measured
redacted10%not measured19.5827.4167.1×

Agent runs by prompt version

Track/versionRunsSuccessCI fails / runMeasuredAPI $ (median)Active min (median)Leverage (median)
research/0128495%0.046190.844.4511×
ventures/1109587%63270.663.5327.8×
process/138578%3.171680.482.1145.4×
redteam/033295%not measured1450.422.1213.1×
ventures/225686%not measured820.633.5313.7×
creative/025194%not measured990.653.2572.8×
explore/020395%1650.592.7201×
research/118497%21420.884.7509.7×
review/19994%not measured440.41.6171×
polish/04787%not measured80.51.4not measured
process/24068%not measured60.441.8not measured
ventures/03291%412.4213.8330.6×
review/223100%not measured90.512.1142.1×
perf/01377%not measured20.459.3not measured
process/06100%111.46.6285.2×
review/010%not measured0not measurednot measurednot measured

Documentation that cost runs time

Machine-readable: stats.json. Metrics marked "not measured" had no source in this build; none is shown as zero.