Way to win 05 / Explainability
Solve the AI decision mystery.
See the exact rule, evidence, confidence, risk distance, and correction behind a Guardian call. Replay it and check the signed proof.
Not a made-up paragraph. A trail your product team, customer, auditor, and board can follow.
The level today
“The AI said so” is not an answer.
A smooth explanation can still be wrong. Teams need the facts that shaped the call and a way to check the same decision again.
Stories can drift
A model can invent a neat reason after the decision is already made.
Scores hide the line
A green badge does not show how close the answer came to known danger.
Screenshots do not replay
A saved output cannot prove the model, policy, and runtime path behind it.
Play the path
Watch Emu turn the hard part into a clear win.
Pick a step or let the level run. The Guardian stays in the loop from the first move to the proof.
PROOF DETECTIVE
Emu is on the move.
Choose a stage
Three easy moves
Start small. Get the win. Keep moving.
Collect the clues
Keep the input, sources, exact clause, factors, tool calls, correction, and outcome.
Map the decision
Show where the answer sat in meaning space and which correction moved it toward safety.
Replay and verify
Pin the decision, run it again, and check the result against its signed receipt.
The prize you can hold
See the product, not just the promise.
Every path ends in a useful record your team can open, share, and act on.
GUARDIAN PROOF RECEIPT
claims-review / DR-10482
- Exact rule
- claims-policy §7.4
- Risk distance
- 0.18
- Correction match
- 0.92 cosine
- Replay result
- Exact match
›original answer kept
›safe correction mapped
›receipt signature verified
What your team gets
Less friction. More AI moving forward.
Why
Made visible
See the rule, evidence, geometry, and fix.
Again
Replay the call
Check the pinned decision for yourself.
Next
Harden the Guard
Turn close calls into tests and training.
Why Trinitite
The Guardian makes this more than a nice dashboard.
The rule, not a story
The exact clause and policy version come from the recorded decision path.
The geometry, not a guess
See distance to known safe and risky regions plus the direction of the correction.
The receipt, not a screenshot
Signed lineage ties the model, policy, input, output, correction, and replay together.
Keep exploring
Use the full platform. Go deeper where you need it.
Focused playbooks
Straight answers
Know Why AI Decided, explained.
Does this expose hidden chain of thought?
No. It gives stronger evidence: recorded inputs, sources, rules, factors, geometry, actions, corrections, and signed lineage.
What is meaning-space distance?
It measures how close an answer is to patterns your team has marked safe or risky. It also shows how far a correction moved the answer.
Can we run the same decision again?
Yes. Supported hosted profiles pin the model, policy, adapter, input, seed, and runtime so the decision can be replayed and checked.
Your turn
Bring one decision nobody can explain.
We will connect the clues, map the risk distance, replay the call, and verify the receipt.