Separate common row constraints from established storage sizes. Preserve Q8_1 and K-layout uncertainty, plus numeric IDs and ambiguous textual identities for historically shifted IDs16–25. Keep zero-byte facts and identity limitations independently, and propagate producer facts through totals, cancellation and retained views. Pin released source evidence and document the compatibility cost.
Fixes#6848
Preserve format and Evidence semantics at their owning boundaries, with focused regression coverage and documentation where applicable.
Fixes#6772Fixes#6787
Share partial observation ownership across GGUF and Safetensors. Preserve completed header facts through cancellation and MCP error delivery without claiming finalized digests or file identity.
Fixes#6684
Keep unsigned integer declarations exact until their owning field is interpreted. Bound actual offsets at conversion and validate shape products in declaration order, preserving large empty tensor shapes without hiding an earlier overflow.
Fixes#6627
Publish alignment from the parser when explicit metadata validates it, or when metadata completion establishes the default. Preserve that fact through later parsing and hashing failures.
Fixes#6579
A safetensors length prefix such as 640 encodes as 80 02 00 ..., which the
provider read as a pickle PROTO marker and refused before the safetensors
parser ran. The parser that owns the length prefix now makes the call: a
pickle-like prefix is safetensors only when byte 8 is '{' and the declared
length is within the safetensors header ceiling. Real pickles stay refused
and are never deserialized.
Fixes#5895
* fix(engines): keep format ceilings fixed and recover the EVMole deadline
Promote the owned EVMole worker wall clock to timeout_ms and name it in
recovery. Fixed CPU, byte, and address-space refusals in this theme say
the adjustable deadline does not raise them.
Refs #5910
Co-authored-by: morluto <morluto@users.noreply.github.com>
* fix(engines): recover the allowance that refused the request
Model capacity recovery follows a structured limit instead of message text,
and the shared header ceiling no longer names Safetensors. Firmware signal
errors keep captured tool output, oversized reports do not suggest an
intermittent rerun, and NativeAOT hydration budget exhaustion explains the
fixed parser allowance on the partial report.
Co-authored-by: morluto <morluto@users.noreply.github.com>
---------
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: morluto <morluto@users.noreply.github.com>
Preserve unsupported GGUF version bytes and both UINT32 interpretations without inventing byte order at the shared parser. Keep existing layout support, typed errors, evidence identity, scoped reads and resource limits.
Validation: 78 focused tests, full static/architecture and documentation checks, real CLI/MCP version controls and an independently checked public model. Latest-head Codex found no issues; Greptile is 5/5 and all review threads are resolved.
Some CI remained pending at merge qualification. Merge under the human owner's explicit standing bypass authorization does not claim the full CI suite passed.
Fixes#5749.
Validate omitted GGUF strings without materializing discarded text while preserving full UTF-8 checks and retained Evidence. Record real-model costs and workflow coverage; preserve failed experiment receipts through cleanup.
Fixes#5774.
Keep parser-owned completed metadata and tensor declarations as typed partial Evidence on cancellation or physical read failure. Preserve unknown final identity and digests, exact caller inputs, primary failures, and retained cleanup owners; support inline and retained MCP error delivery.
Closes#5625
morluto