StableMindMachine Authority
Evidence · Independent consequence verification
StableMind Evidence preserves the attributable lineage of a consequential machine action and supports independent Proof of Consequence after execution. It keeps a hard boundary between what an executor reported, what an independent verifier observed, and what the evidence can actually support.
target: users/482
expected: account.enabled=false
collateral: no privilege expansion
Expected state observed independently; prohibited collateral effects not observed.
Synthetic architecture illustration. This public page does not inspect a customer system or establish external Production Proof of Consequence.
Product definition
Evidence is the StableMind system for attributable machine-action lineage and independent postcondition verification. It joins the semantic request, delegated authority, deterministic decision, approvals, Action Permit, execution receipt, expected-consequence contract, independent observation, and resulting proof without allowing any one source to silently upgrade itself into a stronger truth claim. An executor can report that an API returned success. That report matters, but it does not prove that the intended business state exists, that prohibited collateral effects did not occur, or that the consequence remains true after the call completed.
The evidence distinction
01 / FOUR TRUTHS
Evidence binds the semantic action to its delegation root, policy decision, approval state, Action Permit, target, and request digest so the later outcome cannot drift away from the authorization that preceded it.
A certified connector may report success, failure, partial completion, or an error. That receipt is attributable execution evidence, not independent confirmation that the intended consequence exists.
A distinct read-only verifier observes the target state after execution using its own workload identity, connector certificate, and credential class. The executor is not allowed to verify itself.
A signed proof compares expected postconditions and prohibited collateral effects with the independent observation, preserving uncertainty instead of converting missing data into success.
Request to consequence
02 / SEVEN BOUNDARIES
Anchor the consequence record to the exact delegation root, semantic request, deterministic policy decision, required approvals, Action Permit, target, and request digest. Proof remains attributable to the action that was actually authorized.
Define expected postconditions, prohibited collateral effects, verification deadline, rollback or compensation references, and the independent verification profile before the machine acts. The success criterion cannot be quietly rewritten after seeing the result.
Execution Fabric contributes its signed, attributable report of what the certified executor attempted and returned. The receipt is necessary lineage, but it does not get the final word about reality.
A separate verifier identity inspects the resulting target state through an independently governed, read-only path. Separation matters because the component that caused the change cannot simply attest that its own work succeeded.
Each required postcondition and prohibited collateral condition is evaluated deterministically against the signed observation. Missing, stale, contradictory, or incomplete evidence remains visible rather than being averaged into a reassuring green light.
The proof records VERIFIED, PARTIAL, MISMATCHED, or UNVERIFIABLE and binds that conclusion to the exact contract and observation. A mismatch can require containment, investigation, compensation planning, or delegation suspension.
Downstream settlement, assurance, audit, customer acceptance, and recovery processes may consume the proof. None of them may reinterpret a favorable proof as new machine authority or resurrect an expired permit.
Proof outcomes
03 / NO FORCED GREEN
The expected state is satisfied and prohibited collateral effects are not observed within the governed verification envelope. VERIFIED describes the consequence evidence, not customer acceptance, settlement, or future authority.
The available independent evidence confirms part, but not all, of the expected outcome. Partial verification remains a first-class state rather than being rounded upward into success.
A required postcondition failed or prohibited collateral effect was observed. The proof can drive investigation, containment, compensation planning, or authority suspension without pretending the original action never happened.
Missing, stale, inaccessible, contradictory, or invalid evidence produces UNVERIFIABLE. StableMind does not convert absence of proof into proof of success.
Where proof matters
04 / CONSEQUENCE
A processor acceptance or connector response is not settlement. Independent payment evidence can reconcile destination, amount, state, and timing before consequence reserve or downstream financial processes advance.
A restart, firewall change, role assignment, or production mutation can be checked against independent state after execution, including collateral conditions that must remain unchanged.
Identity, CRM, procurement, HR, and finance actions can be verified against expected records and prohibited side effects instead of trusting only the same write path that made the change.
Guardian intervention can preserve and request evidence, but containment is not claimed successful until the resulting state is independently observed. Recovery evidence cannot silently restore stale authority.
One chain, separated responsibilities
Authorization, execution, observation, and proof remain separately attributable even though they belong to one lineage.
Constitutional boundaries
06 / PROOF IS NOT POWER
The verifier uses a distinct workload identity, connector certificate, and read-only credential class. Self-attestation remains execution evidence, not independent Proof of Consequence.
Postconditions and prohibited collateral effects cannot be retrofitted after the fact to make an inconvenient result look compliant.
When required evidence is unavailable or stale, the outcome is UNVERIFIABLE. Absence is never silently promoted into VERIFIED.
Even a VERIFIED consequence cannot issue a new Action Permit, release credentials, broaden delegation, or reactivate authority that expired or was revoked.
Technical consequence verification, bilateral acceptance, invoicing, cleared payment, revenue recognition, and realized customer value remain separate states with separate evidence.
A MISMATCHED or PARTIAL result is not an inconvenience to hide. It is precisely the information a governed machine-action system must surface while response options are still available.
For evaluators
07 / ASK BETTER QUESTIONS
Public truth boundary
Examples on this page are synthetic and are labeled as such. StableMind does not claim that this public website has independently verified a customer consequence, observed a live customer environment, established settlement, closed customer acceptance, recognized revenue, or produced realized customer value. The current product lineage remains governed by the same Public Truth Contract used throughout stablemind.io.
The customer path
ActionGate remains the commercial front door. SIO10 preserves Evidence while making Consequence Capital understandable before signup, but public account activation remains reserved for SIO19 and the first governed-action experience remains SIO21. Visiting or reading this page creates zero authority and zero external proof.
Interactive evidence
The Proof Explorer lets you inspect a browser-local synthetic evidence envelope artifact by artifact, then change independent evidence quality while keeping the original authorization and executor receipt fixed.