What this tests

parsing

Point your client at this specimen and find out — before prod does — whether it actually validates what arrives, or just hopes.

Anatomy

A Python pipeline serialized a dataframe and shipped {"balance": NaN, "score": Infinity} — legal in exactly one ecosystem and a SyntaxError in every other. JSON does not have NaN. JSON does not have Infinity. Python's json module emits them anyway, out of what can only be described as optimism, and reads them back the same way, so the bug is invisible to the team that produced it and fatal to everyone else. JavaScript throws. Go throws. Java throws. Python, the sole member of the "that's fine actually" party, reads it back and sees nothing wrong — which is precisely how this poison reached production, like a disease that only shows symptoms in other countries. Your client's behaviour here divides neatly into "rejects it" (correct), "parses it as float64 infinity" (defensible), and "silently stores it" (a future incident report with your name in the contributing factors).

Replay it

curl -i https://faultydata.com/r/FD-2026-037

// response appears here

Signature

{Status:200 Headers:map[] Body:{"balance": NaN, "score": Infinity, "floor": -Infinity, "name": "ada"} Corrupt: Delay:<nil> Flaky:<nil> Sequence:<nil> Drip:<nil> StallAfter:0 Hang:false Reset:false Raw: Phases:[]}

Keep digging

FD-2026-011 — Truncated JSON body FD-2026-012 — Malformed JSON (trailing comma) FD-2026-013 — Wrong types in a valid envelope FD-2026-014 — Encoding mismatch (claims UTF-8, sends Latin-1) FD-2026-015 — UTF-8 BOM prefix FD-2026-028 — Malformed channel file update (CrowdStrike, Jul 2024)