FIELDROOKBY ROOKWELD

OPERATION PROVE IT / PUBLIC EVIDENCE LAB

A diagnosis
is not a verdict.

Real FieldRook investigation logic. Synthetic Windows scenarios. Every conclusion tied to evidence — including the ones that cannot be confirmed.

Offline fixtures · Simulated inference · No customer or production telemetry

F//EVIDENCE LAB● REPRODUCIBLE
—FIXTURE & TOOL TESTS
3PUBLIC TRIALS
0LIVE-MODEL CASES
Loading verified run…FIELDROOK
01 / EVIDENCE TRIALS

Three cases.
Three honest outcomes.

Pick an investigation, follow the chain of observations, and inspect the underlying fixture data.

Read-only demonstration.
No controls operate a real computer.

02 / METHODOLOGY

Proof requires
boundaries.

This is an auditable demonstration of specific code paths. It is not a claim that FieldRook can solve any Windows problem or that a language model is always correct.

01

Synthetic by design

Every displayed measurement comes from deterministic FieldRook fixtures, not someone else's workstation. Inference is explicitly simulated.

02

Evidence before inference

Facts and hypotheses cite exported evidence IDs. The UI lets you inspect the underlying bounded tool readings and known gaps.

03

Claims are not fixes

Recovery cases record a manual intervention claim. Fresh re-measurement determines VERIFIED_RESOLVED or NOT_RESOLVED. No automated recovery runs here.

04

Reproduce the test

Check out the referenced commit, run the offline gauntlet, and use the published export script to regenerate the evidence. No cloud API key needed.

REPRODUCTIONcd backend && python -m eval.run

Audit the code, fixtures, and methodology in the public repository. The three curated cases and source-backed results are intentionally small enough for independent inspection.

View reproducibility scripts ↗