Useful code.
Measured delivery.
The factory is the portfolio. An openly automated pipeline turns real problems into reviewed contributions. This readout shows its output, acceptance and engineering limits.
No live requests. No tracking. No private records.
Delivery over time
Weekly output
PRs by submission date and merge date. These are separate event cohorts.
View weekly values
| Week (UTC) | Opened | Merged |
|---|---|---|
| 2026-10-19 | 22 | 3 |
| 2026-10-26 | 68 | 45 |
| 2026-11-02 | 35 | 39 |
| 2026-11-09 | 4 | 12 |
| 2026-11-16 | 0 | 1 |
First-round quality
Clean approved or merged PRs / reviewed PRs, by submission week. Incomplete histories are excluded; recent cohorts are unfinished.
View weekly values
| Week (UTC) | Clean first round |
|---|---|
| 2026-10-19 | 60.0% |
| 2026-10-26 | 58.5% |
| 2026-11-02 | 33.3% |
| 2026-11-09 | 50.0% |
| 2026-11-16 | Not tracked |
Candidate-to-PR lead time
Median hours from candidate creation to submission, by submission week. Missing timestamp pairs leave gaps, not zeroes.
View weekly values
| Week (UTC) | Median hours |
|---|---|
| 2026-10-19 | 1.1 h |
| 2026-10-26 | 1.4 h |
| 2026-11-02 | 5.7 h |
| 2026-11-09 | 5.4 h |
| 2026-11-16 | Not tracked |
Date coverage — submission dates: 129/129; merge dates: 100/100. Line-chart vertical axes show percent and hours respectively; gaps mean no eligible observations.
All weekly output, acceptance and speed values
| Week (UTC) | Opened | Merged | Currently open¹ | Concluded cohort² | Cohort merge rate | Closed share | First-round approval | Clean / reviewed | History coverage | Median lead | Lead samples | Request to fix | Owner interventions |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2026-10-19 | 22 | 3 | 7 | 15 | 86.7% | 13.3% | 60.0% | 9/15 | 22/22 | 1.1 h | 22 | 10.1 h | Not tracked |
| 2026-10-26 | 68 | 45 | 7 | 61 | 91.8% | 8.2% | 58.5% | 38/65 | 68/68 | 1.4 h | 68 | 0.9 h | Not tracked |
| 2026-11-02 | 35 | 39 | 5 | 30 | 90.0% | 10.0% | 33.3% | 10/30 | 35/35 | 5.7 h | 35 | 6.8 h | Not tracked |
| 2026-11-09 | 4 | 12 | 0 | 4 | 100.0% | 0.0% | 50.0% | 2/4 | 4/4 | 5.4 h | 4 | Not tracked | Not tracked |
| 2026-11-16 | 0 | 1 | 0 | 0 | Not tracked | Not tracked | Not tracked | 0/0 | 0/0 | Not tracked | 0 | Not tracked | Not tracked |
¹ Currently open PRs by submission week, not historical inventory. ² Merged or closed without merge, by submission week.
What the numbers can support
11 metrics are not tracked. A missing measurement is a visible engineering gap, never a zero or a guessed result.
- Read-only snapshot of retained local evidence. No GitHub refresh is performed.
- Coverage is eligible records with needed fields / the stated population, not proof of complete historical capture.
- Partial means observed records only. Not tracked means the evidence cannot support the metric; missing values are never zero.
- All weeks start Monday in UTC. Acceptance uses submission cohorts at their current saved outcome, not week-of-decision cohorts.
- Weekly open counts are currently open PRs grouped by submission week, not historical week-end inventory.
- Factory speed measures candidate-to-submission lead time and request-to-fix work, not waiting for maintainers. Response/merge latency and silent-thread share are intentionally excluded.
- Zero event counts mean no events in retained records for that week. Input read errors and omitted dates can undercount.
Input diagnostics — missing files: 188, invalid pr rows: 0, duplicate pr rows: 0.
Output
| Metric | Recorded value | Definition & limits | Coverage & source |
|---|---|---|---|
| PRs opened | 129tracked | Distinct pull request URLs in the ledger, including all outcomes. | 129/129PR ledger: saved outcomes |
| PRs merged | 100tracked | Distinct PRs with a saved merged outcome. Saved state only; this command does not refresh GitHub. | 129/129PR ledger: saved outcomes |
| PRs open | 19tracked | Distinct PRs with a saved open outcome. Saved state only; this command does not refresh GitHub. | 129/129PR ledger: saved outcomes |
Acceptance
| Metric | Recorded value | Definition & limits | Coverage & source |
|---|---|---|---|
| Merge rate | 90.9%partial | Merged / (merged + closed without merge). Open and unclassified outcomes are excluded. Coverage counts concluded PRs against all classified PRs; open PRs are not failures. | 110/129PR ledger: saved outcomes |
| Closed without merge | 9.1%partial | Closed without merge / (merged + closed without merge). | 110/129PR ledger: saved outcomes |
First-round quality
| Metric | Recorded value | Definition & limits | Coverage & source |
|---|---|---|---|
| First-round approval | 51.8%tracked | Approved or merged with no non-self change request, no self non-merge commit/force-push after submission, and (if merged) no non-self corrective update before merge / reviewed PRs. Reviewed means merged or at least one non-self APPROVED, CHANGES_REQUESTED or COMMENTED review; self reviews do not count. Coverage is complete cached histories / ledger PRs; unavailable histories are excluded, not clean. Tracked requires at least 90% coverage. Committer dates proxy ordinary pushes, not actual push times or batches; force-push times are exact. Merge commits are excluded. GitHub cannot recover every removed commit or delayed push. | 129/129GitHub PR metadata, paginated reviews, issue timeline and commits: saved role-only history cache |
| Upstream review rounds per PR | 1.1 roundstracked | Mean non-self CHANGES_REQUESTED review count + 1 over reviewed PRs. Distinct from pre-publication independent review. Coverage is complete cached histories / ledger PRs; unavailable histories are excluded, not clean. Tracked requires at least 90% coverage. Committer dates proxy ordinary pushes, not actual push times or batches; force-push times are exact. Merge commits are excluded. GitHub cannot recover every removed commit or delayed push. | 129/129GitHub PR metadata, paginated reviews, issue timeline and commits: saved role-only history cache |
| Pre-publication independent review rounds per PR | 1.4 roundspartial | Mean preserved independent review verdicts per PR with review records; current and numbered archives are deduplicated by head and verdict. Pre-publication independent reviews, not upstream maintainer rounds. Missing or overwritten archives undercount rounds. | 129/129Item review files and numbered review archives |
Defects
| Metric | Recorded value | Definition & limits | Coverage & source |
|---|---|---|---|
| PRs with fixes pushed after review | 25tracked | PRs with a self non-merge commit or force-push after submission and after the first non-self review. Total updates counts distinct commits and force-push events, not network push batches or proven code defects. 29 commit/force-push updates in total. Coverage is complete cached histories / ledger PRs; unavailable histories are excluded, not clean. Tracked requires at least 90% coverage. Committer dates proxy ordinary pushes, not actual push times or batches; force-push times are exact. Merge commits are excluded. GitHub cannot recover every removed commit or delayed push. | 129/129GitHub PR metadata, paginated reviews, issue timeline and commits: saved role-only history cache |
| Pre-review self-corrections (defect signal) | 7tracked | PRs with a self non-merge commit or force-push after submission before any non-self review, including PRs not yet reviewed. Total updates counts distinct commits and force-push events, not network push batches or proven code defects. 9 commit/force-push updates in total. Coverage is complete cached histories / ledger PRs; unavailable histories are excluded, not clean. Tracked requires at least 90% coverage. Committer dates proxy ordinary pushes, not actual push times or batches; force-push times are exact. Merge commits are excluded. GitHub cannot recover every removed commit or delayed push. | 129/129GitHub PR metadata, paginated reviews, issue timeline and commits: saved role-only history cache |
| Candidate-selection misses | Not trackednot tracked | PRs closed without merge because they were duplicate, already fixed or unwanted; each PR counts once. Closed outcomes have no structured closure-reason classification. Private notes and generic closed events cannot reliably distinguish selection misses from other closures. | 0/10PR ledger: closed outcomes; follow-up closure events |
| PRs with observed CI failures | 4partial | Distinct submitted PRs with a failing CI snapshot or a ci-failing follow-up event. Repeated observations count once; this is not a code-caused bug rate. Observed minimum only. Successful latest checks do not prove that earlier runs passed; failures on the base or infrastructure are not attributed to the PR. | 119/129PR ledger: checks; follow-up events: ci-failing |
| Post-merge reverts | Not trackednot tracked | Merged PRs with a confirmed subsequent revert. Follow-up stops observing merged PRs; no post-merge incident registry exists. Red/green test reverts are not production reverts. | 0/100Follow-up events; PR ledger |
| Post-merge regressions | Not trackednot tracked | Merged PRs with a confirmed subsequent regression. Follow-up stops observing merged PRs; no post-merge incident registry exists. Red/green test reverts are not production reverts. | 0/100Follow-up events; PR ledger |
Filtering
| Metric | Recorded value | Definition & limits | Coverage & source |
|---|---|---|---|
| Recorded candidates | 192partial | OSS item records persisted under the runs directory; pre-item scouting ideas are not counted. This is the persisted candidate population, not all opportunities inspected. | 192/192Item metadata |
| Rejected at qualification | 6partial | Candidates with a saved skip verdict. Earlier rejections overwritten by a later check are not recoverable. | 190/192Item qualify records |
| Dropped at value review | 0partial | Candidates with a complete structured value review whose verdict is drop. Legacy prose is not classified. | 6/192Item value-review records |
| Sent back by independent review | 44partial | Candidates with at least one preserved fix-needed independent review verdict; each candidate counts once. | 160/192Item review files and numbered review archives |
Factory speed
| Metric | Recorded value | Definition & limits | Coverage & source |
|---|---|---|---|
| Review request to pushed fix | 1.4 htracked | Median hours from each non-self CHANGES_REQUESTED review to the next self non-merge commit or force-push. Comment-only reviews are too ambiguous to establish a fix request; requests without a subsequent fix are excluded. 9/9 change requests have a subsequent fix. Coverage is complete cached histories / ledger PRs; unavailable histories are excluded, not clean. Tracked requires at least 90% coverage. Committer dates proxy ordinary pushes, not actual push times or batches; force-push times are exact. Merge commits are excluded. GitHub cannot recover every removed commit or delayed push. | 129/129GitHub PR metadata, paginated reviews, issue timeline and commits: saved role-only history cache |
| Median candidate-to-PR lead time | 2.5 htracked | Median hours from the persisted candidate item's creation to PR submission; scouting before item creation is not recorded. Date-only, naive and reversed timestamps are excluded. | 129/129Item metadata: created_at; PR ledger: submitted_at |
Efficiency
| Metric | Recorded value | Definition & limits | Coverage & source |
|---|---|---|---|
| Agent minutes per PR | Not trackednot tracked | Total attributable agent minutes / PRs in the same work cohort. Shift telemetry excludes nested workers and has no PR attribution. Command runtime is not agent time; dividing partial recent shifts by lifetime PR outcomes would be misleading. | 0/129Shift summaries; item command receipts |
| Tool calls per PR | Not trackednot tracked | Total attributable tool calls / PRs in the same work cohort. Shift telemetry excludes nested workers and has no PR attribution. Command runtime is not agent time; dividing partial recent shifts by lifetime PR outcomes would be misleading. | 0/129Shift summaries; item command receipts |
| Agent minutes per merged PR | Not trackednot tracked | Total attributable agent minutes / merged PRs in the same work cohort. Shift telemetry excludes nested workers and has no PR attribution. Command runtime is not agent time; dividing partial recent shifts by lifetime PR outcomes would be misleading. | 0/100Shift summaries; item command receipts |
| Tool calls per merged PR | Not trackednot tracked | Total attributable tool calls / merged PRs in the same work cohort. Shift telemetry excludes nested workers and has no PR attribution. Command runtime is not agent time; dividing partial recent shifts by lifetime PR outcomes would be misleading. | 0/100Shift summaries; item command receipts |
| Shifts run unattended | Not trackednot tracked | Completed shifts with no manual intervention from start to finish. Automated launch and exit status do not establish absence of human help. Shifts have no intervention flag or linked intervention log. | 0/11Shift summaries; owner actions |
| Owner interventions per week | Not trackednot tracked | Manual interventions actually performed, grouped by UTC week. Owner action notes mix requests, resolutions and delegated work; account-operation logs record agent actions, not manual interventions. They cannot be counted as human actions. | 0/70Owner action notes; delegated account-operation records |
| Recorded coordinator minutes | 334.1 minutespartial | Sum of active_minutes in saved automated-shift summaries; not full-factory cost. Coordinator only. Worker sessions are excluded by the shift summarizer; telemetry covers only retained shift logs. | 11/11Shift summaries |
| Recorded coordinator tool calls | 880 callspartial | Sum of tool_calls in saved automated-shift summaries; not full-factory cost. Coordinator only. Worker sessions are excluded by the shift summarizer; telemetry covers only retained shift logs. | 11/11Shift summaries |
Factory health
| Metric | Recorded value | Definition & limits | Coverage & source |
|---|---|---|---|
| Shift completion rate | 81.8%tracked | Automated shift summaries with status exactly exited 0 / summaries with a recorded terminal status. Failed starts count against completion. | 11/11Shift summaries: status |
| Factory test count | 1,566tracked | Test count or wall-clock runtime from the explicitly supplied full-suite runner report; never inferred from source or unrelated item tests. Recorded suite passed. | 1/1Full-suite runner JSON report supplied with --suite-report |
| Factory test runtime | 165.8 secondstracked | Test count or wall-clock runtime from the explicitly supplied full-suite runner report; never inferred from source or unrelated item tests. Recorded suite passed. | 1/1Full-suite runner JSON report supplied with --suite-report |
| Factory defects found | Not trackednot tracked | Distinct confirmed factory defects found. There is no stable defect identifier linking discovery and resolution. A repair commit may fix several defects or revisit one; commit counts are shown separately. | 0/0Factory git history; no defect registry |
| Factory defects fixed | Not trackednot tracked | Distinct confirmed factory defects fixed. There is no stable defect identifier linking discovery and resolution. A repair commit may fix several defects or revisit one; commit counts are shown separately. | 0/0Factory git history; no defect registry |
| Factory repair commits | 19tracked | Non-merge commits touching factory, bin, schedule, hooks or tests whose subject begins fix, fixes or fixed. Counts commits, not distinct defects. Only explicitly labeled repair commits; differently worded repairs are excluded. | 158/158Local git history |
How this factory works
Find demand
Look for unclaimed problems in projects that welcome automated contributions.
Prove the change
Qualify the value, make the smallest useful repair and capture executable evidence.
Review independently
Separate implementation, independent review and coordinator release checks.
Ship and learn
Disclose automation, follow feedback and use acceptance and quality to improve the factory.
Read the architecture and workflow · Read the portfolio direction and metric contract