Add JSON edge-case coverage; clear parse-error messages on both engines

[ImMortaL0P]

What

Closes #169. assets/profiler-engine.js's parseJSON() and faircode/loaders_extra.py's read_table() had no test coverage for malformed/unusual JSON input, so it wasn't clear what a user actually sees when they drop a bad file.

Findings & fixes

Truncated/invalid JSON syntax used to leak a raw, engine-specific parser error — a browser SyntaxError in JS, an internal pandas ValueError in Python (which, worse, retried under orient="split" only to fail again with an equally confusing second error). Both now fail fast with the same clear "Unsupported JSON format" wording the existing tabular-shape checks already use.

Found while writing the tests: the JS engine's "columns orientation" check only verified each top-level value was a plain object, not that its entries were scalars — so a deeply-nested, non-tabular structure like {"a": {"b": {"c": 1}}} was silently misread as one column "a" with a row "b" whose cell was the object {"c": 1}, instead of raising an error. It now throws the same clear message as the other unsupported-shape cases.

Scope note: Python's pandas-backed reader is intentionally more lenient than the JS engine for two of the four cases named in the issue — a bare array of primitives ([1,2,3]) becomes a one-column DataFrame, and {} becomes an empty DataFrame — rather than erroring. I left that behavior alone and the new tests just pin what it currently does; changing pandas' own JSON-orientation handling felt like a bigger, riskier change than what a parse-error-message fix calls for. Happy to revisit if you'd rather tighten it for parity with the JS side.

Changes

• assets/profiler-engine.js: wrap JSON.parse syntax errors; tighten the columns-orientation check to reject nested (non-scalar) values.
• faircode/loaders_extra.py: pre-validate JSON syntax with json.loads before handing off to pd.read_json, so a genuine syntax error gets one clear message instead of two confusing pandas ones.
• scripts/parse-json-js.js (new): mirrors the existing scripts/profile-json-js.js pattern — reports parseJSON()'s result or error message as JSON on stdout so tests can assert on it via subprocess.
• tests/test_json_edge_cases.py (new): covers all four cases named in the issue (truncated JSON, array of primitives, empty object, deeply-nested structure) on both engines.

Testing

• pytest tests/ → 80 passed, 9 skipped (skips are pre-existing, for the optional excel/parquet extras I didn't install locally).
• tests/test_js_parity.py (the Python/JS profiler-output parity suite) passes unchanged — confirms the columns-orientation tightening doesn't affect any valid input already covered there.
• Manually re-verified all edge cases from the issue against the fixed parseJSON() in Node directly.

---

Disclosure: I used an AI coding assistant (Claude) to investigate this issue and write the fix/tests. I read and verified CLAUDE.md's paper-freeze scope first — this only touches the actively-developed profiler tool (profiler.py/profiler-engine.js), not the frozen audit/benchmark pipeline. Happy to iterate on review feedback.

[vercel]

@ImMortaL0P is attempting to deploy a commit to the yashkewlani2020-gmailcom's projects Team on Vercel.

A member of the Team first needs to authorize it.

[github-actions]

Hey @ImMortaL0P, your first PR on Fair Code, that's awesome, thank you.

This project is about making AI more accountable, and contributions like yours are what keep that work going. We'll review your changes shortly.

If you haven't already, give the contributing guide a quick read: it covers how audits are structured and what we look for in a review.

[yakew7]

Valid edits will be merging the pr