Work log: March 12, 2026

What shipped today

Today was a high-throughput pipeline day — 25 commits landed on main across three major themes: engine test infrastructure, production reliability, and UI improvements.

Engine test coverage went from 35 to 115 tests. The parent issue #1051 (“engine has 1 test file across 42 handlers”) is now fully closed. We built shared pytest fixtures (conftest with mock_db, mock_pool, make_command_row), then wrote comprehensive unit tests for the four most critical engine modules: the poller (payload decode, stuck command recovery, fail/retry, handler dispatch), the scheduler (idempotency keys, briefing hour matching, 5-minute window rounding, timezone-aware nudges), social_publish (credential validation across all 5 platforms, publish dispatch for both legacy and social queue paths), and briefing_generate (lifecycle classification with boundary tests and edge cases). All tests use consistent patterns — parametrized inputs, datetime mocking, and SQL assertion via call_args inspection.

Production bug fix: KeyError in content.stage_start handler. A user’s content creation via the AI assistant tool was failing because assistant_tools.py sent "stage" instead of "stage_name" in the command payload. Every other caller in the codebase (15+ places) used the correct key — this was a simple typo that only triggered through the Maya chat path. Fixed the caller and hardened the handler to accept both keys for in-flight commands.

Infrastructure hardening. The asyncpg connection pool was running with minimal config — no connection acquisition timeout, no TCP keepalive. After a cluster of Sentry TimeoutErrors and a transient Railway deploy failure, we added timeout=30 and keepalive settings to detect stale connections through pgbouncer. Also triaged and closed 7 duplicate Sentry/deploy issues that all stemmed from the same transient database hiccup.

Ideas page sort toggle. Added strength/newest/oldest sort buttons to the ideas list page, persisted via URL params across tab switches and search queries.

Completed

• #1080 — Add pytest conftest with db mocking fixtures for engine tests
• #1081 — Add poller unit tests for payload decode, stuck commands, and dispatch
• #1082 — Add scheduler unit tests for idempotency keys and briefing hour matching
• #1083 — Add social_publish unit tests for credential validation and publish dispatch
• #1084 — Add briefing_generate unit tests for lifecycle classification
• #1051 — Engine test coverage parent (all 5 children complete)
• #1089 — [Sentry] TimeoutError (pool hardening)
• #1091 — Deploy failure (transient, resolved)
• #1094 — Deploy failure (duplicate of #1091)
• #1096 — [Sentry] KeyError: ‘stage_name’ (bug fix)
• #1095 — Sort ideas by date
• #1087, #1088, #1097, #1098, #1099 — Duplicate Sentry TimeoutError issues (closed)

Release progress

• v1.5: 1 open / 4 closed
• v2: 0 open / 20 closed (complete)
• v1-outbound: 0 open / 19 closed (complete)

Carry-over

• #1100 — Add words target on content initiation (ready-for-prep, not yet started)
• #1036 — In-memory rate limiter (blocked on Upstash setup)

Risks

• The content that failed due to the stage_name bug should auto-retry now that the fix is deployed. If it doesn’t, the command may need manual reset in the commands table.

Flags and watch-outs

• Engine tests run from within engine/ directory — import paths are engine.poller not engine.engine.poller
• SQL parameter assertions: idempotency key is at call_args.args[5] (args[0] is SQL, args[1-5] are $1-$5)
• The deploy alert workflow labels issues as “Failed Vercel deploy” even for Railway failures — minor but confusing

Next session

• Prep and exec #1100 (words target on content initiation)
• Check that the stage_name fix actually unblocked the user’s content (verify in commands table)
• Consider prepping #1036 if Paul has set up Upstash
• Run /scout to find more improvement opportunities now that engine test infra is solid