fix(ts-sdk)!: bound text at 1-10,000 chars measured as the caller sent it

[adnanrhussain]

Brings §4.1 bounds to their spec defaults and removes input normalization.

Limits

min 10 → 1, max 100,000 → 10,000. The SDK defaults carry no product opinion: the minimum excludes empty input and nothing else. Meaningful minimums belong to each evaluator's input schema — which math-question-alignment already does, reading maxLength straight from its input_schema.json.

One length, measured on the caller's text

The SDK previously validated the trimmed length while sending the original text to the model, so the bound enforced described a string the model never saw. Telemetry reported a third thing, agreeing with neither.

	Before	After
Validation bounds	trimmed	as sent
text_length_chars	as sent	as sent
Sent to model	as sent	as sent

trim survives only as a validity test — whitespace-only input is rejected outright, never repaired. Padding is the caller's to remove; absorbing it silently means never telling them their data is malformed, and billing them for characters we pretended weren't there.

Breaking

• Text over 10,000 characters now raises InputValidationError, down from 100,000.
• Text that only fits once trimmed is rejected, with the real length reported (received 10,004 characters). A trailing newline on an otherwise-maximal text is enough — deliberately, since that text is what reaches the model.
• Short text is no longer rejected. min: 1 means anything non-empty passes; evaluators that need a real floor must declare it.

Nothing in the repo is affected: the largest fixture input is 2,910 characters.

Known dead branch

With min: 1 the minimum check is unreachable — text.length < 1 implies the empty string, which the whitespace-only check rejects first. Mutation testing confirms it: if (false) survives and the block reports as uncovered.

Left in place because it goes live the moment an evaluator declares a higher minimum, and because deleting it would mean re-deriving the bound later. Flagged here so the permanent coverage gap is not mistaken for a missing test — it should not be papered over with one.

Spec follow-up

§4.1 in #162 says "Trim … then validate", "Trimmed length < min", "Trimmed length > max", and "text_length_chars … is the trimmed length". This PR contradicts all four. §4.1 also never says whether the trimmed text is the input or only the measurement — the ambiguity that let the SDK validate one string and send another. Both need fixing before #162 merges.

Tests

12 in a new file, plus 6 existing suites updated where they asserted the old min: 10.

Verified load-bearing by mutation, not just by passing: reintroducing the trim in validateText fails 1 test, and in an evaluator's telemetry call fails 3 — so the validation and reporting paths are pinned independently. One test asserts the reported length equals the text the prompt actually carried, which is the invariant that was broken.

668/668 pass · tsc --noEmit clean · lint 0 errors.

[Copilot]

Pull request overview

This PR brings the TypeScript SDK’s text-length handling into §4.1 conformance by (1) capping validated text at the spec default of 10,000 characters and (2) ensuring telemetry/log-reported text_length_chars reflects the same trimmed text length that validation actually bounds.

Changes:

• Reduced VALIDATION_LIMITS.MAX_TEXT_LENGTH from 100,000 to 10,000 and kept validation based on trim()ed text.
• Introduced textLengthChars(text) and migrated evaluators to use it so logging/telemetry consistently report trimmed length.
• Added unit tests covering trimmed-length reporting (success and error paths) and the 10,000-char cap behavior.

Reviewed changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
sdks/typescript/src/evaluators/base.ts	Lowers max text length to 10,000, adds textLengthChars(), and uses it in validation logging; validation continues to use trimmed length.
sdks/typescript/src/evaluators/conventionality.ts	Uses textLengthChars() for start/success/error telemetry/log length reporting.
sdks/typescript/src/evaluators/vocabulary.ts	Uses textLengthChars() for consistent telemetry/log length reporting across paths.
sdks/typescript/src/evaluators/text-complexity.ts	Uses textLengthChars() for consistent telemetry/log length reporting.
sdks/typescript/src/evaluators/smk.ts	Uses textLengthChars() for consistent telemetry/log length reporting.
sdks/typescript/src/evaluators/sentence-structure.ts	Uses textLengthChars() for consistent telemetry/log length reporting.
sdks/typescript/src/evaluators/purpose.ts	Uses textLengthChars() for consistent telemetry/log length reporting.
sdks/typescript/src/evaluators/organizational-structure.ts	Uses textLengthChars() for consistent telemetry/log length reporting.
sdks/typescript/src/evaluators/intertextuality.ts	Uses textLengthChars() for consistent telemetry/log length reporting.
sdks/typescript/src/evaluators/grade-level-appropriateness.ts	Uses textLengthChars() for consistent telemetry/log length reporting.
sdks/typescript/src/evaluators/math/standards-alignment.ts	Uses textLengthChars() for consistent telemetry/log length reporting of question.
sdks/typescript/tests/unit/evaluators/text-length-reporting.test.ts	Adds tests verifying trimmed-length telemetry reporting and enforcing the new 10,000-char max constraint.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!