Fix plugin tests: add missing drivers field to mock contexts - #133

… reason) beside FILTER_TEXT_CASES (#18438)

Fixes #18113

`@objectstack/spec/data` now publishes the case-insensitive-contains
**text-comparand door** — the predicate for the two REJECTION rows
`FILTER_TEXT_CASES` has declared since #5701, and the reason text they
are refused with — so every face reads one implementation instead of
writing its own copy of both halves. Executes objectui#9048 batch #133
item 1, ruling **D**.

## What landed

| path | what |
|---|---|
| `packages/spec/src/data/filter-text-comparand.ts` | new —
`isRefusedTextComparand`, `textComparandRefusalReason`, and the
module-internal `describeComparand` |
| `packages/spec/src/data/index.ts` | re-export (the entry-reachability
hop the ruling names) |
| `packages/spec/src/data/filter-text-comparand.test.ts` | new — both
acceptance pins, driven from the table |
| `packages/spec/api-surface/data.json`,
`packages/spec/export-origins/data.json` | regenerated: **2 added, 0
removed** |
| `.changeset/18113-icontains-text-comparand-refusal.md` | `minor` on
`@objectstack/spec` |

Measured file face: **6 paths**, all under `packages/spec` plus the
changeset — disjoint from the sibling cards named at dispatch.

## The port is byte-exact, and that is measured rather than asserted

The reference is objectui `packages/core/src/utils/text-comparand.ts` at
objectui `origin/main` @ `55f39ee9`. That blob is
`c29782139426fc1af51dac86f0a4b8d0b5d2827e` at `55f39ee9`, at objectui
`HEAD` (`ff1d5ea8`) and in the read-only checkout — one file, not three
readings.

- **Source text**: the three function bodies extracted from both files
and compared — `describeComparand` 159 bytes, `isRefusedTextComparand`
115, `textComparandRefusalReason` 928, **all three identical** (only the
`export` modifier normalised, which is the publication decision, not the
behaviour).
- **Behaviour**: both implementations driven over 17 comparand shapes x
2 arriving spellings x 2 field names — **85 readings, 0 mismatches**,
including the two shapes `describeComparand` exists for (a BigInt and a
cyclic object, on which `JSON.stringify` throws).

Both are one-off scripts, not committed test files.

## The acceptance pins

`pnpm --filter @objectstack/spec exec vitest run
src/data/filter-text-comparand.test.ts` — **46 passed**.

1. **Every** `FILTER_TEXT_CASES` case is driven through
`isRefusedTextComparand` (24 cases, one `it` each), plus a whole-table
set assertion that its TRUE set equals the rows the table declares
refused for this operator, plus every `FILTER_TEXT_ROWS` stored `name`.
2. The reason for `''` and for a non-string comparand carries each row's
`mustMention` tokens, **per arriving spelling**, plus a transcription
pin on the exact bytes — `mustMention` only requires `$icontains`, so it
cannot catch a reword, and a reword is what the fence forbids.

**Ablation** (mutate, prove it hit disk, run, restore, prove the restore
— both legs restored with `git hash-object` matching the HEAD blob and
`git diff HEAD` empty):

| leg | on-disk proof | result |
|---|---|---|
| reword one word of the reason | removed-text count 1 to 0, injected 1
| **1 failed** / 45 passed, exit 1 |
| drop the `=== ''` arm from the predicate | removed-text count 1 to 0,
injected 1 | **2 failed** / 44 passed, exit 1 |

## Two readings the card asked for

**⚠️ The second acceptance clause is falsified for one spelling, and the
fence outranks it.** The card asks that the reason contain each case's
`mustMention` tokens "for BOTH arriving spellings". `mustMention` is
`['$icontains']`; for an arriving `icontains` the reason names what
arrived and therefore does **not** contain the `$`-dialect token.
Rewording it to would break the byte-for-byte fence and diverge from two
shipped faces. This is also what objectui#9152 already does about it —
it names the `$` twin in its own tail, with a comment saying why the
substitution would misdirect a view author. So the pin asserts the
measured truth in both directions: the token for the `$` spelling, the
token minus its sigil for the infix one, and `not.toContain` on the `$`
form there. Flagged for the seat rather than settled by me.

**The fold measurement — measured, not folded.** The note names
`@objectstack/formula` and the engine. Neither is a literal copy of this
predicate:

- `packages/formula/src/matches-filter.ts:249` fuses the De Morgan
complement into its positive arm (`typeof v === 'string' && v !== ''`)
and **answers `false`** rather than refusing — its docblock says the
totality is deliberate.
- `packages/objectql/src/having-filter.ts:388` **refuses**, so it is not
one of the faces that answer 0 of 9 rows.

So: report the reading and leave it. What the sweep did turn up is in
the acceptance notes below.

## Verification

- `pnpm --filter @objectstack/spec test` — **483 files, 13734 tests, all
passed**.
- `pnpm --filter @objectstack/spec typecheck` — exit 0. Coverage proven
rather than assumed: the main `tsconfig.json` excludes `**/*.test.ts`,
and `--listFiles` on `tsconfig.test.json` (what `check:test-typecheck`
runs) shows **both** new files in the program.
- `pnpm --filter @objectstack/spec check:generated` — proved exactly
`api-surface/` and `export-origins/` stale, `--fix` regenerated only
those two. The gate's own diff line is the published-face evidence: `+
isRefusedTextComparand (function)`, `+ textComparandRefusalReason
(function)` on `./data`, nothing else, and no `describeComparand`.
- **Derived gate families**: `scripts/pm/dispatch-gates.mjs` derived
**85** from the change set; **83 run, 2 NOT MEASURED, 0 unrun**,
reconciled through `--ran` with a recorded exit code per family.
- **Lint, narrowed and declared** (at `812a5bf565`): eslint's own config
lints `**/*.{ts,tsx,mts,cts,js,jsx,mjs,cjs}` repo-wide; `--format json`
reports **3 files linted, 0 errors, 0 warnings**; and type-aware linting
is **not** enabled for these files (`parserOptions.project`,
`projectService` and `EXPERIMENTAL_useProjectService` all null under
`--print-config`), so this diff cannot move the verdict on any file it
does not touch. The repo-wide sweep is CI's.
- **Changeset publishes** — measured, with both controls.
`@objectstack/spec` is public at `17.4.0` and its `files[]` is `dist`,
`json-schema`, `liveness`, `prompts`, `llms.txt`, `README.md`,
`src/**/*.zod.ts`, `CHANGELOG.md`, `api-surface`, `spec-changes.json`.
**Positive control**: `isRefusedTextComparand` appears in **6** shipped
files under `dist/`, `dist/data/index.js` / `.mjs` / `.d.mts` and the
browser build among them. **Negative control**: `ARRIVING_SPELLINGS`,
which exists only in the new test file (2 occurrences in `src/`),
appears in **0** shipped files — so the grep discriminates and the test
layer does not ship. Note `src/**/*.zod.ts` IS published, but the new
module is not a `.zod.ts`, so it reaches consumers only as built output.
A published symbol moved, so this is `minor` — ⛔ not `skip-changeset`.

**NOT MEASURED (2 of 85), both `exit 3 PREREQUISITE NOT MET`** —
`check:dual-build-cjs-loads` and `check:type-check-debt` both refuse
until the repo-wide build exists, and 54 of 69 buildable packages have
no `dist/` in this worktree. A repo-wide build is CI's run (`Build
Core`, `TypeScript Type Check`), not this card's; declaring the
narrowing rather than reporting a green.

## Acceptance notes

**A gate whose verdict depends on whether you built —
`check:cross-package-test-inputs`, measured red here and green on
`main`.** It names
`packages/cli/test/init-created-files-summary.e2e.test.ts` descending
into `packages/spec/dist/`; it names none of my files. Ablated to be
sure: same tree, same sources, `dist/` parked aside then restored —
**exit 1 with `packages/spec/dist/` present, exit 0 without it** (115
entries, restored). A pristine `origin/main` worktree also exits 0,
because nothing there is built. So the finding is real and CI is
structurally blind to it: the lint job does not build spec, and
following AGENTS.md's mandatory "regenerate spec artifacts before
pushing" is exactly what makes it visible. Filed as a finding for the
seat, not fixed here.

**Five faces carry the same discrimination inline, and none is in the
note's named set.** `driver-sql:4047`, `driver-mongodb:296`,
`driver-memory/filter-refusal.ts:911`, `objectql/having-filter.ts:388`
and `driver-turso/remote-transport.ts:2989` each spell `typeof X !==
'string' || X === ''` and each throws its own `icontainsComparandError`.
Folding them is a five-package diff with a message-text question inside
it, well outside this card's measured face and outside what the note
authorises. Noted, not filed; the natural carrier is whichever card next
consolidates that refusal family.

**`Clause-②: yes`** — two new exported symbols on a published entry. The
carrier is the seat's on both the card and this PR; not hung, not
stripped, not waited on here.

---
_Generated by [Claude
Code](https://claude.ai/code/session_01KB5PFtxuy1x3dcR5gxudx6)_

---------

Co-authored-by: Claude <noreply@anthropic.com>