[aknight-ld]

Normalizing both python and js SDKs to use a neutral stringify for consistency in content hashing, message history, and judge reasoning. This helps us avoid shoving a bunch of python based conventions into js (and other future builds) and keep our standards language agnostic.

BEGIN_COMMIT_OVERRIDE
feat(evaluations): hash eventId and contentHash as RFC 8785 canonical JSON, the same in every SDK
fix(client): show a judge an output of 0, false, {} or []
END_COMMIT_OVERRIDE

---

Note

Overview
Replaces Python-specific text formatting with language-neutral JSON rules so the JS SDK matches other SDKs on evaluations hashing and judge-facing text.

contentHash / eventId: canonicalJson is now RFC 8785 (JCS)—sorted keys, no whitespace, JSON.stringify number/string spelling (no pythonJsonNumber or ASCII-escaping non-ASCII). contentHashFor hashes that canonical form directly; reference hashes in tests change to align with the shared Python canonical_json tests.

Judges: Removes python-format.ts. buildMessageHistory and parseJudgeResponse use toolResultString (JSON for non-strings). Empty sections use isEmptyPart instead of Python truthiness, so 0 and false outputs are included in message_history. Offline scorer/judge validation errors use describeValue (JSON) instead of pythonRepr, so messages look like got true / got "0.5" rather than Python literals.

Docs and tests are updated (judge-scoring.test.ts added, python-format.test.ts removed).

^{Reviewed by Cursor Bugbot for commit e4eff49. Bugbot is set up for automated code reviews on this repo. Configure here.}

[jeffdupont]

Nice cleanup: dropping python-format.ts is the right direction. canonicalJson matches py#209's canonical_json byte for byte on all 26 edge cases I tried (floats, integers past 2^53, lone surrogates, UTF-16 key order). The client's 1782 tests and tsc --noEmit pass locally.

The red "Type check" isn't from this PR. main has failed "Check API reports" since #135 merged without updating claude-agents' report. #160 fixes it. Once #160 merges, rebase and this goes green.

One thing before this merges:

feat(evaluations) hides a core judges change

buildMessageHistory is also used by the online judges in judges.ts (lines 191 and 413), so online judges now see an output of 0, false, {} or [] that they used to skip. (parseJudgeResponse is used only by evaluations, so the reasoning change stays experimental.) Under an evaluations scope, release-please lists the whole PR under Experimental, so that core fix never reaches the core changelog. Could you add this to the description?

BEGIN_COMMIT_OVERRIDE
feat(evaluations): hash eventId and contentHash as RFC 8785 canonical JSON, the same in every SDK
fix(client): show a judge an output of 0, false, {} or []
END_COMMIT_OVERRIDE

On the "rubric reads the same in every SDK" goal: JS is already the reference form here, so it needs no change. The gap is on the Python side (compact separators and number spelling), and I've left the details on py#209.