The context goes in
The redacted operating context the founder gave Eve — the messy, real input, not a demo script.
Proof, not promises
The Evidence Room
This room will replay one real workflow as receipts — the context that goes in, what Eve drafts, what the engine scores, what breaks, what changes, and what the evidence says. Nothing is on record yet: the first record is being earned right now, the journey below is its published shape, and the honesty rules are already binding. Every artifact carries the same evidence label the product itself uses — estimated, simulated, sandbox, pilot, or production — and nothing is presented above the stage it has earned.
Deveras and CommitGate are both owned by the founder. They are dogfood pilots, not customers, and we say so.
being earned
Eve drafts the architecture
Eve's actual first draft of the workflow — every step, paradigm prescription, and assumption, unedited.
The deterministic score receipt
The structural score with all 12 criterion results, the engine version, and the input snapshot — recomputable by anyone with the same engine. Zero model calls.
Something real breaks
The actual failure — what the sandbox or a real run caught, verbatim. If nothing has broken yet, this stage says so instead of inventing drama.
The approved revision
The operator-approved change and its re-score — the before/after delta with both receipts.
Evidence accumulates
Where the workflow sits on the readiness ladder today — estimated, simulated, sandbox, pilot, or production — with each claim labeled at its earned stage and nothing above it.
Our own companies run on Everform, and the first public evidence record is being earned right now — on the real Deveras workspace, through the same engine every customer workspace uses. It appears here the moment it clears redaction review, with its export digest and engine version attached. Nothing on this page will ever be a mock-up dressed as evidence.
Why you can trust the numbers that will appear here
The structural score is deterministic.
777 lines of our own code, 12 weighted criteria, the same input always scores the same. Zero model calls. The published receipt includes the engine version and the exact input snapshot, so the score can be recomputed — not believed.
Everything model-derived is labeled an estimate.
Eve's judgment — suitability dimensions, paradigm fit, readiness projections — comes from a model and is labeled that way. We never blend an estimate into a measurement and call the result deterministic.