Quesarby MLAI

Security & trust

Evidence before promises.

Keep documented behavior, reported testing, measured results and targets separate.

Two independent axes

Implementation status and evidence provenance answer different questions. Current/Partial/Proposed describe the scope of an implementation. Measured/reported/target describe how a public numerical figure should be interpreted in the existing Quesar content contract.

Neither axis can replace the other. A reported test suite is not an independently reproduced benchmark. A proposed feature is not a delivered capability. An attractive graph is not evidence that its numbers were measured.

A minimum evidence record

A useful performance statement identifies the source revision, date, environment, configuration, workload, methodology and artifact containing the results. It also states limitations and keeps comparisons consistent.

This page publishes no runtime performance figures. ABI’s README labels its project-site dashboard data as synthetic samples. Those samples must not become latency, accuracy, energy-efficiency or throughput claims elsewhere.

What source review can establish

Source review establishes what a particular document says. Testing a website can establish whether its own links, search, controls and layouts work in the tested browser. Neither activity independently validates the linked software’s runtime, security, model quality or production deployment.

Links to further specifications are reading pointers, not an assertion that every linked document, implementation path or external service was audited.

Checked against these sources on .

All docsPrevious: Fail closed. Tag the figure.Next: A companion, not a capability claim.
Architecture

Next

Developers

The READMEs these articles cite.