CURRENT LOCAL CONTRACT
Fictional CedarLab text only. Explicit setup retrieves public packages and pinned model files;
it does not send documents/queries to an inference provider. Actual encoding then uses local CPU.
Core lab.py only replays packaged vectors and does deterministic parsing/review metrics.
No API key, paid call, live tool action, public service or website DB.
Artifacts:source JSON, saved vectors and optional model/cache/venv in learner-controlled folder.
Do not add real user/employer data. Protect/delete own local copies according to their purpose.
Bounds:short text<=2000 characters and<=256 wordpieces; no silent model truncation; small corpus.
First recorded encoder run:about 1.81 seconds for 15 short texts including initialization/hash checks,
macOSarm 64 CPU, ONNXRuntime 1.23.2/numpy 2.5.4. One run, not p 95 or a promise.
No paid API bill; download about 91 MB plus package bytes and local storage/compute.
Unsupported questions need evidence review/abstention; no automatic entailment judge.

HYPOTHETICAL HOSTED ADDITION — NOT IMPLEMENTED/AUTHORIZED
Must identify endpoint/provider/model, data categories, retention, keys, cost caps, timeouts,
retry policy, owner authorization, applicable obligations and separately tested quality.
No price, privacy promise, operating latency or active adapter is supplied by this worksheet.
PYTHON CALLBACK NOTE
A supplied callback is ordinary Python code and can have side effects; the example is a
known local fixture function, not a security sandbox for arbitrary third-party callbacks.
