# What I Worked On

Two weeks (2 April through 15 April) produced 25 merged MRs and roughly 40 commits, overwhelmingly CI and testing infrastructure with a handful of features and a security upgrade layered in. The work forms one continuous arc: wire mutation testing into CI → make it work on MR pipelines → write tests that actually pass the quality bar → shard the Supabase stack so CI runs three jobs in parallel. Every MR was mergeable on its own and every walk-back is visible in the commit log, not hidden with force-pushes.

# Week 1 (2-8 Apr): Mutation Testing Infrastructure

Week 1 was dominated by getting mutmut and Stryker to run cleanly in CI. 19 MRs merged. Mutation testing does not configure itself on the first try — each pipeline run revealed a new failure mode, and each fix became its own MR:

Each of these is standalone. No stacked drafts, no “WIP” commits sitting on a branch for days. Partly discipline, partly pragmatism: when the CI runner takes 8-12 minutes to surface a failure, batching changes across pipelines wastes more time than it saves. Merge the fix, start the next one, get feedback.

Three application features also shipped in week 1 (invoice number generation, auth on GET client endpoints, Sentry instrumentation for Celery) and three carry-over CI MRs from the prior week closed on April 2.

# Week 2 (9-15 Apr): Per-MR Mutation Feedback + Parallel Slots

Week 2 shifted focus from “does it run?” to “does it give useful feedback on MRs?” and “does it scale to concurrent pipelines?” Five MRs merged plus SIRA-274 as MR !197 (merged on the final day). The full list:

Shortest gap between merges was 40 minutes (!184 at 14:22 → follow-up in !185). Week 2 also included three direct commits that did not warrant their own MRs: the axios GHSA-3p68-rc4w-qgx5 SSRF upgrade, two rounds of mutation-killing tests (b8930639 + fe69e036), and Sentry performance spans for risk scoring.

# The Walking-Back Pattern (!186 → !187)

!186 made Stryker auto-run on MRs. A day later, !187 walked that decision back and made it manual again. Both are in the week’s commit log. This pattern is worth showing, not hiding:

1. Stryker was manual because it took 38 minutes on the CI runner
2. I removed the typescript-checker plugin (mutants with type errors already die at runtime under vitest, so the checker was redundant)
3. With the checker gone I expected the job to drop below 20 minutes and auto-run safely
4. It dropped to about 28 minutes — still too slow to block MR merges

Walking back a decision by writing a new commit (not a force-push, not a revert with no context) is the discipline piece. The MR title says exactly what changed and why. A reviewer looking at !186 and !187 in sequence can reconstruct the reasoning without asking. Git history stays honest.

# SIRA-274: Four Commits, One Merged MR

The SIRA-274 branch had four commits before squash-merging to main:

dba63d41 chore(ci): shard mutation:python Supabase stacks across 3 slots
8c2f9970 chore(ci): shard api:integration-test Supabase stacks across 3 slots
00bfa6ea chore(ci): fix ci-supabase-slot.sh subshell FD bug
1b5a697d chore(ci): add ci-supabase-slot.sh for parallel local Supabase stacks

Read bottom to top: add the script, fix a bug in it, apply it to each job that needed a Supabase instance. A disposable test MR (!198) was opened alongside !197 to verify concurrent slot allocation on a real pipeline, then closed after the all-green run.

The fix commit (00bfa6ea) sits between the introduction and the two shard commits. The bug was specific: calling acquire_slot inside command substitution meant the file descriptor for the flock got opened in a subshell and released immediately. The fix changed the calling convention. Full design notes are in this week’s b2 programming blog.

Pipeline 15028 (post-merge of !197) ran all green: api:integration-test 1m59s, mutation:python 4m36s, SonarQube 85% coverage gate passed, all other jobs green.

# Commit Quality Across Both Weeks

Conventional commit prefixes stayed consistent end to end:

fix(ci): fix mutation test commands for Python and TypeScript
fix(ci): use pnpm exec for Stryker mutation testing
chore(ci): add disk cleanup step before every pipeline
fix(ci): fix Stryker plugin discovery for pnpm + mutmut
feat(api): integration test infrastructure + all domain tests
test(api): strengthen integration tests for mutation testing
chore(ci): mutation testing on MR pipelines + report comments
fix(ci): mutmut 0% score + linear false tagging
fix(web): upgrade axios 1.13.5 → 1.15.0 (GHSA-3p68-rc4w-qgx5 SSRF)
chore(ci): shard api:integration-test Supabase stacks across 3 slots

The prefix vocabulary stayed narrow and meaningful:

• fix(ci): something was broken in the pipeline
• chore(ci): not broken, but incomplete
• feat(api) / feat(web): new user-facing capability
• test(api): test-only changes, no behavior change
• fix(web): bug fix or dependency upgrade

No vague “update”, “tweak”, or “wip” messages. Every message is a one-line summary of what a reviewer will see in the diff.

# Results

# Evidence

• Week 1 MRs (19 total): !138, !140, !141, !144, !145, !146, !147, !148, !149, !150, !152, !153, !156, !158, !159, !130 (SIRA-123), !154 (SIRA-82), !105 (SIRA-135), CLAUDE.md docs MR
• Week 2 MRs (6 total): !183, !184, !185, !186, !187, !197 (SIRA-274)
• Week 2 disposable: !198 — concurrency test MR (opened, verified, closed)
• Week 2 direct commits: 2e5bc2fd (axios SSRF), b8930639 (round 2 tests), fe69e036 (round 3 tests), 1d957968 (Sentry spans), 7aceb817 (CSV fix)
• Post-merge pipeline: 15028, all green, integration-test 1m59s, mutation-python 4m36s