CacheInvariant v0.3.0

CacheInvariant v0.3.0 adds a registered direct-token prefix-divergence lane
for the exact pinned llama.cpp b10107 CPU runtime.

Evidence in this release:

• main and lightweight v0.3.0 both resolve to
  203f84a42bb7fb841abd3c1f5d221b0bed6de289;
• main CI run 30293701488 and tag CI run 30293833809 each passed 9/9
  jobs, including fresh Windows and Ubuntu exact-runtime integrations;
• the source-bound Windows schema-v3 reference verifies 77/77 registered
  checks, while the preserved schema-v2 and schema-v1 references still verify
  57/57 and 29/29 checks; and
• exact, shared-prefix, and first-token-divergence rows are compared with
  matched cache-off requests using bounded cache/prompt counters.

The direct-token lane explicitly sends n_predict=1, requires one reported
predicted token, then discards generated content and generated token values.
Every registered direct-token request also pins n_cache_reuse=0, disabling
that separate optional KV-shifting chunk-reuse path; the paired
cache_prompt=true and cache_prompt=false requests remain the tested prompt
reuse switch. This is not an output-quality or performance benchmark.

This release contains only the wheel and source distribution. Pinned runtime
binaries, fixture inputs, and the converted model are fetched and checked by
the reproducible integration path; they are not redistributed. There is no
PyPI publication or paid-service dependency.

Development and validation were AI-assisted and owner-operated. These results
do not establish independent reproduction, review, users, adoption,
production use, security, cross-runtime equivalence, model quality, latency,
throughput, memory savings, or recruiting signal.

CacheInvariant v0.2.0

CacheInvariant v0.2.0 adds a source-bound dual-stream cancellation and reuse
protocol for the exact llama.cpp b10107 CPU adapter.

What is registered:

• two fixed streams start behind one barrier in both launch/disconnect orders;
• both clients must receive a nonterminal event before either disconnects;
• the first cancelled slot must return to idle, then both slots must return to
  idle;
• both reused-slot results and prompt-work counts must match their isolated
  baselines; and
• the offline verifier accepts both the new schema-v2 evidence and the bundled
  schema-v1 v0.1.0 reference.

The bundled Windows v0.2.0 reference records 57/57 registered invariants. Free
GitHub Actions exercise the exact runtime lane on Windows and Ubuntu without
publishing generated CI evidence.

The record retains timing-sensitive processing samples as non-gating raw
observations. This release does not claim scheduling fairness, simultaneous
token generation, survivor liveness, throughput, production isolation,
cross-runtime equivalence, external reproduction, adoption, or users.

The attached wheel and source distribution are release artifacts only. This
release is not published to PyPI.

CacheInvariant v0.1.0

Pre-release evidence lab for version-pinned llama.cpp b10107 cache correctness, slot isolation, cancellation/reuse, state save/restore, and a full local process restart.

The source and sdist include one normalized Windows reference observation bound to its exact source revision. Tag CI generated and verified fresh Windows and Ubuntu observations without publishing those ephemeral bundles.

Boundaries: one tiny synthetic fixture, local CPU execution, and registered engineering invariants. This is not a model-quality or throughput benchmark, security certification, production validation, independent reproduction, external adoption, or a recruiting outcome. The release contains only the wheel and source distribution; it does not redistribute llama.cpp, Node, runtime binaries, fixtures, or converted models.