Motir

Found while running MOTIR-2196 (the createTestWorkItem fixture fix). An out-of-scope finding logged rather than absorbed (notes.html #27): MOTIR-2196's blast-radius grep turned up three MORE sites writing a position no shipped code path can emit. They are deliberately NOT in that card's diff — see "Why this is separate" below.

The three sites

Verified, not reasoned:

$ node -e "const{generateKeyBetween}=require('fractional-indexing');
  for (const k of ['000001','p0000001','a0'])
    { try{generateKeyBetween(k,null);console.log(k,'VALID')}catch(e){console.log(k,'INVALID:',e.message)} }"
000001   INVALID: invalid order key head: 0
p0000001 INVALID: invalid order key: p0000001
a0       VALID

What each one costs

• scripts/seed-large.ts is the worse of the two shapes, because a HUMAN uses its output. pnpm db:seed:large fills a dev database, and the first board drag that lands next to one of those cards calls keyForAppend/keyBetween on a head-'0' key → the move API 500s and the board renders "Move not allowed". That is exactly the failure scripts/plan-seed/seed.ts:396-412 was fixed for and wrote five lines of comment about; the sibling large-seeder never got the message. (scripts/seedLargeBoard.ts:155 DID — it carries the warning.)
• The two board E2E specs are latent rather than live: keyBetweenSafe (lib/workItems/positioning.ts) treats an invalid bound as an open end, so a drag in those specs currently degrades to an append instead of throwing. Nothing is red today. What is wrong is that the rows are not the system under test — a board-at-scale spec is asserting reorder behaviour over a key space the product cannot produce, so the tolerance path is what is under test rather than the ordinary one.

Why this is separate from MOTIR-2196

MOTIR-2196 fixed the shared createTestWorkItem fixture and the four vitest seeders that copied it, all verified locally against the real Postgres. These three are a different risk shape:

• scripts/seed-large.ts is not a test fixture at all — it is a dev-data script, the other half of the pair plan-seed/seed.ts already fixed.
• The E2E seeds change behaviour, not just data: with valid bounds keyBetweenSafe mints a key strictly BETWEEN the neighbours instead of falling back to an append, so a spec asserting post-drag card order can legitimately move. That has to be validated by running the board E2E specs, which MOTIR-2196's run could not do locally (the sandbox's /dev/shm and the shared inngest port make heavy board E2E unreliable — prodect-e2e-run-harness-oom).

Absorbing them would have put an unverifiable E2E behaviour change inside a card whose whole point was a verified fixture fix.

Acceptance criteria

• scripts/seed-large.ts mints its position via lib/workItems/positioning.ts, chained globally in creation order, and carries the same warning scripts/plan-seed/seed.ts:396-412 does.
• Both board E2E specs seed real fractional-index keys, chained so the seeded card order is unchanged.
• The board E2E specs pass — including any drag/reorder assertion, which now exercises keyBetween rather than keyBetweenSafe's invalid-bound fallback. If an assertion moves, the PR body says which and why.
• A grep for a position: written as a padded or prefixed number returns nothing across tests/, scripts/ and lib/ — the whole class is gone, not three named files.

Context refs

• lib/workItems/positioning.ts — keyForAppend / keyBetween / keyBetweenSafe (the tolerance path these currently lean on) / isValidOrderKey.
• scripts/plan-seed/seed.ts:396-412 — the warning, and the global-chain argument these should mirror.
• scripts/seedLargeBoard.ts:155 — the sibling script that already learned it.
• tests/fixtures/workItemFixtures.ts — MOTIR-2196's nextTestPosition, the shared helper to reuse where a test can reach it.