Evidence Chain / Proof-of-Work

These examples and illustrative results are independently authored teaching materials, not measured model results.

Use case

Explain teaching request r17 returning 503 at 12:04. Materials contain db_disconnect for that request and current handler.ts line 42 mapping the error to 503, but no database monitoring. An observed mapping does not confirm the whole incident’s root cause.

Mechanism

List claims with scope and link each to inspected artifacts, time, version and location. Separate log observations, code paths and inference; reconcile request identity and deployed version. Record missing evidence and alternatives. Reproduce the same trigger when feasible or retain provisional status. Preserve correction evidence and avoid expanding one request to all failures.

Bad example

The database probably failed, so every 503 came from it. Mark root cause confirmed without reading logs or code.

Good example

Investigate r17 using its 12:04 log and the deployed error mapping. Report observed db_disconnect and the path to 503. Database cause and other 503 paths remain unverified. List missing monitoring and next checks; do not turn one request into a whole-incident cause.

Why the change matters

A chain decomposes conclusions into checkable links. Request and version identity give logs and code a common object. Explicit inference and gaps prevent a plausible explanation from becoming an independent fact.

Observable expectation

The teaching report separately locates log, mapping and provisional cause. A version mismatch weakens the conclusion and requires checking. Finding timeout→503 nearby invalidates any claim that all 503s came from disconnection. A reviewer can retrace citations.

Limits

Records can be incomplete or wrong; temporal proximity does not establish causation. Evidence covers inspected objects and conditions, and subjective confidence is not calibrated probability. Saved records and digests do not prove current state; important claims need current evidence.