Email tim.almond@thehouseinc.xyz with the details. You will get an acknowledgement within 72 hours and a status update at least every 14 days. Please give us 90 days of coordinated disclosure before publishing; we will credit you in the advisory unless you ask otherwise.

Do not open public issues for suspected vulnerabilities. A vulnerability that also exists in upstream serde_json is reported to upstream as well, with your consent, so both trees are fixed together.

Only the latest published release receives security fixes.

Assets. The correctness of every byte this library reads and writes: a document that parses must parse to the same value upstream serde_json would produce, a value that serialises must serialise to the same bytes, and an invalid document must be rejected at the same place with the same error. Consumers (mID token bodies, rusty_time reports, SpaceDB values, Deputy receipts, every public API the house ships) stake their wire compatibility on that contract.

Adversaries and untrusted input. The whole input is hostile by definition: JSON arrives from the network, from files, from other tenants. The library opens no sockets, files or processes itself, holds no secrets and logs nothing; its attack surface is the bytes handed to from_* and the values handed to to_*.

STRIDE pass.

• Spoofing / Repudiation / Information disclosure: out of scope -- no identity, no logging, no secret state. Authenticity of a document is the consumer's job.
• Tampering: a divergence from upstream is the failure mode that matters. It is gated by the in-process differential oracle (bytes, Value, error text with line and column, float bits) over the corpus, several hundred edge documents and every number token, on every commit and under every feature flag.
• Denial of service: hostile input must never panic, overflow or recurse without bound. The recursion limit (128) is on by default; every public entry point returns a typed Error; the fuzz target from upstream is kept and M1 adds one per public entry point. Release builds set overflow-checks per the measured decision recorded in the ledger.
• Elevation of privilege: the only unsafe is upstream's twelve sites, inherited unchanged at M0 and inventoried in docs/UPSTREAM-CHANGES.md. M2 confines them to one audited island with a SAFETY: comment per block and a scalar twin gated byte-identical on every architecture.

Residual risks are listed in docs/plans/use-protection-please.md once the hardening audit runs (mission plan M8).