Function check_decode_truncation_series 

pub fn check_decode_truncation_series<D>(dec: D, valid: &[u8]) -> Conformance
where
    D: DecoderConfig + 'static,

A truncated (incomplete) input is categorized as incomplete client data — never mis-attributed as an internal bug, an OOM, an I/O error, or a caller fault — and never panics or silently decodes.

This enforces the one part of the ErrorCategory taxonomy that IS broadly distinguishable: whatever a codec’s exact error, cutting a known-good image short must land in the incomplete-input set (UnexpectedEof is ideal, but the rest of the image-bytes-origin Image arm (Malformed / Unsupported) is tolerated because a truncated prefix can genuinely look malformed). It catches the real bug class where a codec reads a length field out of truncated data and OOMs, or funnels a truncation into Internal — a 5xx for a 4xx-class client error.

valid is a KNOWN-GOOD encoded image the codec supplies (at least a few bytes). The check builds a deterministic series of prefixes and, for each, runs a full decode through the dyn-erased boundary — DynDecodeJob::push_decode into a throwaway sink. That is the dyn full-decode path, so the pixel body is actually read and body truncation is exercised, not just the header a probe would parse. Each decode is wrapped in catch_unwind so a codec panic on truncated input becomes a named failure, not a process abort. Every offending offset is collected and reported together.

Per offset n of len:

• panic → failure (truncated input must never panic).
• Ok (decode succeeded) when n <= len/4 (header/early region) → failure: a truncated valid image must not silently decode. For n > len/4 a success is tolerated (trailing-marker / already-complete-payload leniency).
• Err → the erased ErrorCategory is read back: an incomplete-input category passes; any other category fails, naming the offset. If the category did not survive erasure at all, the codec is Pattern A (native-enum type Error, not At<CodecError>) — reported once, exactly as check_decode_error_envelope reports it.

Like the envelope checks, this is opt-in and NOT part of check_all: the testkit’s own reference codec is a Pattern-A foil that fails it on the erased path, while the internal minimal envelope exemplar passes (its truncations classify as MalformedImage).