Evaluate
Whetstone
Compare your current and candidate AI system on the same checks. Keep every gain and regression visible.
Public workbench↗
CYBERELF LABS / EST. IN CURIOSITY
Strange ideas.
Serious systems.
We build tools that make AI changes measurable, actions accountable, and complex work easier to trust.
INDEPENDENT BY DESIGNOPEN TO INSPECTION
START HERE / WHETSTONE
Better on average.
Worse where it matters?
An AI upgrade can gain six capabilities and lose three you already rely on. Whetstone makes those tradeoffs visible before the release decision.
Compare two agents ↗PAIRED RELEASE CHECKBLOCK
Baseline7/20
→Candidate10/20
+6Gains
3Regressions
11Ties
The score improved. Three regressions still block promotion.
Illustration of the gate. Source & boundaries →THE TOOLKIT
A tool for each
hard question.
Start with the problem you have. Whetstone stands on its own; the release tools can also form a governed chain.
Prove
SalienceLean
Turn supported, typed obligations into Lean-checked proof receipts with explicit assumptions.
Working prototype↗
Adopt
Foundry
Bind formal, empirical, and runtime receipts to the same candidate. Fail closed when an authority is missing or vetoes the change.
Working prototype↗
Authorize
ScopeWrit
Sign the permitted scope of an agent run. Check each consequential action and consume its authorization once.
Public alpha↗
Execute
Backplane
Package and execute signed, scoped runtime capabilities, retaining the decision and execution record.
Working runtime↗
Reconcile
Coherence
Compare records across independent systems, preserve source evidence, resolve exceptions, and resume a handoff exactly where it stopped.
Browser-local alpha↗
Six tools. Explicit responsibilities.Compare the tools →
THE OPERATING PRINCIPLE
Make a claim.
Build the test.
Keep the result.
We like ambitious ideas. We also like the part where they encounter a real constraint.
Proof, measurement, permission, and execution answer different questions. Keeping them separate makes the whole system easier to inspect.
How the pieces fit →AFTER HOURS / STILL IN THE LAB
Leave room
for the strange.
Visual instruments. Digital organisms. New ways to organize memory. The lab is where the next useful question starts.
Explore the experiments →
MorphoDigital biology / concept illustration ↗WORK WITH THE LAB
Bring a problem
worth getting right.
Have an AI change to evaluate, a system that needs a clearer boundary, or an experiment worth building? Talk to Justin.
justin@cyberelf.link ↗