ECAI Implementation and Review Prompt

Use the following prompt when asking an assistant to explain or extend ECAI. It supplies an evaluation approach, not a model-training procedure or evidence that generated code works.

Explain ECAI using the implementation revision and documentation supplied for this task. Identify the relevant modules, schemas, dependencies and evidence. Separate implemented behaviour, observed test results and research hypotheses.

For knowledge retrieval, preserve source records, permissions, provenance and corpus version. Distinguish deterministic identity from semantic accuracy. Describe the actual retrieval algorithm and any model used to compose an answer.

For relation processing, state the input types and rules. Evaluate against known answers and a baseline before claiming a broader reasoning capability.

For cryptography, name the construction, parameters, trust anchor and threat model. Use reviewed libraries and specified test vectors. Do not describe hashes as reversible, curve membership as truth, ordinary ECC as post-quantum, or a lookup as a zero-knowledge proof.

For actions and payments, separate permission to retrieve, sign, execute and spend. A candidate repair requires review and validation; a passing test is not a settlement receipt.

When proposing code, list what was actually run and what remains untested. Do not invent a benchmark, deployment, API, library or acceptance result.

Start with the implementation overview, encoding and proof boundaries and capability matrix.