Bind outcomes to the actual agent, version and principal.
Know what actually happened.
MachineOutcome is building independent infrastructure that verifies real AI-agent outcomes — then turns those verified outcomes into durable history, task-specific reliability and safer machine decisions.
Outcome truth
evidence-bound
01Agent actsIdentity · task · authority
02Evidence arrivesProvider · execution · result
03Outcome verifiedVERIFIED · FAILED · UNKNOWN
04History compoundsReceipts · cohorts · reliability
Payment systems can prove money moved. Observability can show what software executed. Identity can say who the agent is.
MachineOutcome answers a different question:
What was the actual outcome?
Know what the agent was permitted and expected to do.
Retain objective execution and provider evidence with provenance.
Return deterministic truth without converting missing evidence into confidence.
Build append-oriented, version-aware longitudinal outcome records.
Measure only comparable task-specific evidence with explicit uncertainty.
Current operational family
AI-agent repository changes.
The first permanent operational path is software.repository_change.v1: verify what happened after a coding agent changed a repository.
MachineOutcome is designed so additional outcome families can use the same core truth model without creating a parallel source of truth.
Truth contract
Designed to resist fake certainty.
- 01No universal trust score.
Reliability is task-specific, version-aware and evidence-bounded.
- 02UNKNOWN is valid.
Insufficient or conflicting evidence does not become a positive verdict.
- 03History is not silently rewritten.
Corrections, disputes and reversals preserve lineage.
- 04Outcome before reputation.
Higher-order risk and routing systems must be downstream of verified evidence.
Outcome→History→Reliability→Risk→Routing
The system grows only when the underlying evidence supports the next layer.