TRUTH ENGINE Request access

A trust boundary for consequential software actions

The action may
have happened.
The application cannot prove it.

Truth Engine determines what software is justified in claiming after an external action.

Run the Playground Read the API
flagship example
$ action.send()
  1. 1Sending action to provider…
  2. 2Provider accepted and processed it.
  3. 3Response was lost.
  4. 4The application cannot prove occurrence.
provider_effects
1
response
LOST
truth
UNKNOWN
retry
not attempted

Try the difficult case

What should the software do now?

Retrying is an action. Truth Engine checks whether that action is actually authorized.

  1. 01
    Initial action

    Provider processed the request.

  2. 02
    Retry

    Requested, but not automatically permitted.

  3. 03
    Reconcile

    Read-only observation looks for evidence.

  4. 04
    Final truth

    Updated only when evidence justifies it.

ENGINE-BACKED PLAYGROUND CONNECTION REQUIRED BEFORE PUBLIC DEPLOYMENT

$ retry()
RETRY_NOT_PERMITTED Retry intent is not retry authority.
provider_effects
1
truth
UNKNOWN
$ reconcile()
  1. 1Read-only observation…
  2. 2Evidence found.
  3. 3Truth updated.
SATISFIED Available evidence now justifies the claim.
provider_effects
1
lookups
1

The model

Separating execution from truth.

Truth Engine controls what software is justified in claiming after consequential actions.

INTENT
→
AUTHORITY
→
ATTEMPT
→
EVIDENCE
→
JUSTIFIED
TRUTH

Three truth states

There are exactly three.

SATISFIED

The available evidence justifies that the intended effect occurred.

NOT_SATISFIED

The available evidence justifies that the intended effect did not occur.

UNKNOWN

The available evidence does not justify either certainty.

Reconciliation may still return UNKNOWN.

Where it sits

A truth boundary between your software and external systems.

Understanding an instruction does not create authority to execute it. Executing it does not prove what happened.

YOUR APPLICATION / AI
↓
TRUTH ENGINE
↓
EXTERNAL PROVIDER
payment · email · data API

Flagship proof

ONE EFFECT.

Response lost.

Truth remained UNKNOWN.

Blind retry denied.

Evidence reconciled.

Truth became SATISFIED.

PROVIDER EFFECTS 1

Evaluate for yourself

Don’t trust our claims. Break it.

Run the engine-backed Playground. Try retrying uncertainty. Reconcile evidence. Leave a scenario unresolved.

⌁ PlaygroundRun the flagship scenario. → { } API & DocsInspect the evaluation surface. →

What this proves — and what it does not

Be precise about the boundaries.

Evaluation environment Synthetic provider No exactly-once guarantee No production certification Reconciliation may remain UNKNOWN No scale claim
Read Trust & Limitations →