What the factory produced

Claims are cheap; counts are not. Every number on this page is read when the site is built, from the repository itself or from a committed snapshot that records the command that made it.

Delivery

commits: 2,187; 24 from the 2014 desktop app
merged pull requests: 834; 954 including Dependabot
PRs about the process itself: 135; the factory improving the factory
issues: 985; 169 of them labelled type:bug

The first merged pull request in the snapshot is from 2025-12. The latest month, 2026-10, is also the busiest: 435 merged PRs by 2026-10-28.

merged PRs (no Dependabot) PRs about the process itself bug issues opened per merged PR (right axis)

Show the data as a table

Monthly delivery data
Month	Commits	Merged PRs	Process PRs	Bug issues	Bugs / PR
2026-03	87	15	1	0	0.00
2026-04	133	38	6	6	0.16
2026-05	50	33	0	3	0.09
2026-06	191	94	2	17	0.18
2026-07	149	69	5	54	0.78
2026-08	135	52	18	17	0.33
2026-09	694	97	34	32	0.33
2026-10	670	435	69	40	0.09

Snapshot generated 2026-10-28 by python3 workspace/ci/generate-ai-sdlc-history.py. Correlation, not proof of causation: bug issues are counted by the month they were opened.

The factory itself

The process is code too. These are counted from the folders that hold it.

464

OpenSpec changes archived

one per shipped change

326

living capability specs

docs/openspec/specs

skills

.claude/skills

agent definitions

.claude/agents

CI/CD workflows

.github/workflows

process scripts

workspace/sdlc

102

Playwright E2E suites

testing/e2e

architecture decisions

ADRs

use cases

docs/100-business

security case families

security/cases

Quality held while speed grew

Backend coverage (JaCoCo)

0%instructions

0% of branches

From public/metrics.json, regenerated on every deploy (as of unknown). Full report.

Ratchets: quality only moves one way

mvn verify enforces the JaCoCo ratchet floor; Checkstyle and SpotBugs counts may only shrink; architecture baselines may only shrink. The case study tells how the hexagonal baseline went to zero.

↗ enforce-quality-ratchet.py ↗ quality-ratchet-baseline.json ↗ hexagon-baseline.txt

Each mechanism, when it arrived

2025-12-22
First agent entry file
AGENTS.md
2026-07-25
Local preflight mirrors CI
workspace/sdlc/preflight.sh · #720
2026-08-08
Engineering Constitution
CONSTITUTION.md · #781
2026-08-08
Constitution wired into OpenSpec (Gate 1)
workspace/sdlc/validate-sdlc-plan.sh · #783
2026-08-28
Dashboarded pre-PR pipeline
workspace/sdlc/run_pipeline.sh · #856
2026-09-22
Harness hooks enforce the workflow
.claude/hooks/block-push-to-main.sh · #1027
2026-09-28
Process checks in CI
.github/workflows/sdlc-process.yml
2026-09-28
Local-model foreman/worker harness
local-ai · #1074
2026-10-02
Cloud agent fleet
docs/300-development/304-ai-sdlc-cloud · #1114
2026-10-03
Heavy-CI merge gate
workspace/sdlc/check-heavy-ci.sh · #1133
2026-10-10
Context budget: 16k → 8k tokens
CONSTITUTION-AGENT-CARD.md · #1259
2026-10-11
Security case catalog before every PR
security/cases/catalog.yml · #1500
2026-10-11
Archive-in-the-PR hygiene
workspace/sdlc/check-openspec-archive-hygiene.py · #1252
2026-10-13
Atomic issue claim and lanes
workspace/sdlc/claim-issue.py · #1623
2026-10-14
Continuous publication
workspace/sdlc/publish-progress.sh · #1681
2026-10-14
Shrink-only ArchUnit baseline
backend-api/src/test/resources/architecture/hexagon-baseline.txt · #1695
2026-10-14
Shrink-only frontend boundary baseline
frontend/eslint-suppressions.json · #1696
2026-10-28
Stack-agnostic framework extracted
framework/ai-sdlc · #1958

Reading these numbers honestly

Throughput is not value. Many October PRs are small, and a share of them are about the process itself.
Bug issues are counted by the month they were opened. A falling ratio is a correlation, not proof that the gates caused it.
The history snapshot was generated 2026-10-28 by python3 workspace/ci/generate-ai-sdlc-history.py; re-run it to refresh.

Next: apply it to your project →