Defect: NV-002: Text schema `unknown()` is accepted as generated data instead of rejected

[eviltester]

Parent testing issue: #1

Source defect report: https://github.com/eviltester/anywaydata-ai-testing/blob/main/docs/testing/20260625/issue-230-001/defects/NV-002-text-schema-unknown-call-accepted.md
Testing artifact folder: https://github.com/eviltester/anywaydata-ai-testing/tree/main/docs/testing/20260625/issue-230-001
Original grid-table-editor story: eviltester/grid-table-editor#230
Original grid-table-editor PR: eviltester/grid-table-editor#247

The detailed defect report from the testing artifact is copied below.

---

NV-002: Text schema unknown() is accepted as generated data instead of rejected

Severity: Low/Medium

Status: confirmed in negative-validation lane.

Environment

• URL: https://eviltester.github.io/grid-table-editor/app.html
• Test environment branch/commit: codex/230-method-picker-mvc / 04570e0e428d
• Evidence: ../logs/negative-validation-test-log.md, ../support/negative-validation-observations.json, ../screenshots/negative-14-text-unknown-busy.png

Steps to Reproduce

1. Open the deployed app and expand Test Data.
2. Switch schema editing to text mode.
3. Enter a schema with column Mystery and rule unknown().
4. Click Generate.

Expected Result

A method-like unknown function call should be rejected with a clear validation message, or the docs should clearly state that unknown function-like text is intentionally treated as literal data.

Actual Result

The schema generated a row with value unknown, making the malformed function-like call look accepted rather than invalid.

Triage Decision

This is due to the autotyping of a value - and in edit as text mode, when you click preview or generate, you don't see the interpreted schema which has moved to regex(mustry()) - and when we switch from a regex in ui schema to text edit schema we render it as mustry() not regex(mustry()) - in the same way that we re-render the raw csv for an enum to be enum(1,2,3,4)

Let's start rendering regex in the edit as text schema as regex(mustry()) and when the Edit as Text view is open, we should refresh it to show the interpreted version of the schema, not the raw schema - this way, when an action is taken, the text schema shows the actual schema interpreted and it is obvious that the function has been interpreted as a regex and not as an invalid function.

Docs should make clearer how the schema text interpretation is done so that this is obvious.

Notes

This may be a parser ambiguity rather than a method-picker regression, but it matters because issue 230 surfaces richer method names and examples; malformed method-like text should not silently pass as generated data without clear intent.

[eviltester]

Fixed in eviltester/grid-table-editor commit 9632b6adc4f75faad36f238168e6c69b649134ae on branch codex/230-method-picker-mvc.

What changed:

• Regex schema rows now serialize back to Edit as Text as explicit regex(...) rules, so a row interpreted as regex is visible as regex(unknown()) rather than unknown().
• Text-mode preview/generate paths now refresh the open schema text from the interpreted rows after successful parsing/validation, making the interpretation visible before data is generated.
• Added regression coverage for the mapper, shared schema editor controller, generator runtime service, and text-mode generate flow.
• Updated the schema definition docs to explain that unknown method-like text is interpreted as regex and refreshed in explicit regex(...) form.

Verification:

• Focused Jest regression set: passed (6 suites, 62 tests).
• pnpm run verify:ui: passed.
• pnpm run verify:local: passed.

Published to the test environment:

• Test environment: https://eviltester.github.io/grid-table-editor/
• Generator page checked: https://eviltester.github.io/grid-table-editor/generator.html returned HTTP 200.
• Storybook checked: https://eviltester.github.io/grid-table-editor/storybook/index.html returned HTTP 200.
• Publish workflow run: https://github.com/eviltester/grid-table-editor/actions/runs/28223011085