[scosman]

Adds two company fixtures in test mode, small_co and big_co, and the account table they need.

Summary

Fixture	What it is	Size
empty	The schema plus the sandbox account, in live mode. Its rows are unchanged; only its description is updated.	0.6 MB
small_co	Wrenfield Labs: a seed-stage SaaS company with 45 hand-written customers, from its 2025-08 beta to NOW.	1.8 MB
big_co	Ostrander Systems: field-service software with 379 simulated HVAC, plumbing and electrical businesses, over 18 months, plus scripted loose ends.	23.7 MB

All three are frozen at 2026-09-01T14:00:00.000Z. Each company fixture is built by replaying its history through the world's own dispatch and the four engine functions billing_engine.md §7 provides for fixture generators. The build never writes SQL straight to the tables. python fixtures_src/generate.py rebuilds every fixture byte-for-byte.

The only world change: an account table

• What it is: schema/006_account.sql is a one-row account table. It replaces the hardcoded account in startup.py, so a fixture carries its own account identity and mode.
• The mode rule: a row with no mode is stamped with the mode it is first opened in. A stored mode that differs from an explicit livemode raises WorldBug.
• Version and tests: the world version goes from 0.1.0 to 0.2.0. tests/test_account_table.py covers the table.
• Nothing else changes: no other file under src/ is touched. Running git diff main -- src/ shows only schema/006_account.sql, startup.py and the version string in world.py.

Fixture generator (fixtures_src/)

• replay/: the clock swap, the timeline, the dispatch client (one savepoint per call), the subscription and payout drivers, the finishing writes, and check_fixture. check_fixture checks the consistency invariants and every expectation, and each check has its own planted-violation test.
• small_co/: the catalog, the 45-member cast and the story.
• big_co/: the shape, the population draw, the simulation and the scripted loose ends.
• guide.py checks the sql landmark= blocks in docs/fixtures.md against the committed fixtures. There are 27 landmarks, and none differ.

Deviations from the spec, decided along the way

• big_co is about 380 customers, not 3,000. The benchmark projected 135–250 MB for 3,000, and the budget is 25 MB.
• Disputes come only from one-off payments. Each is a payment confirmed through POST /v1/payment_intents on a dispute test card. The world opens no dispute from a renewal charge, and the generator calls no engine function to force one.
• World bugs are filed, not fixed here: Subscription renewal invoices bill the period that just ended, not the new one #12, Repeating coupons never end, and coupons.times_redeemed is never incremented #13, Pending invoice items are never swept into subscription invoices #14 and canceled_at stays null on a cancel_at_period_end request #15. The fixtures avoid them where possible: they use only once/forever coupons and bill setup fees on their own invoice. The rest are declared under "Known gaps" in docs/fixtures.md.
• The Wrenfield name stays. wrenfield.app is a live, unrelated games site. No company called "Wrenfield Labs" was found.
• Further deviations are listed in specs/projects/company_fixtures/phase_plans/phase_4.md and phase_plans/catchup.md.

Test plan

• uv run ruff format --check && uv run ruff check
• uv run ty check
• uv run seahaven check
• uv run pytest. The full suite passes, including:
  • the byte-identical rebuild of every fixture;
  • the expectations and loose ends for small_co and big_co;
  • the planted violations;
  • the fork cost;
  • the guide.
• python -m fixtures_src.guide reports 27 landmarks and 0 differences.
• CI is green on this PR.

🤖 Generated with Claude Code

https://claude.ai/code/session_01T4ry4bFLC4jJqmCF5aowCN

---

Generated by Claude Code