Resources

Concepts: the loop as data

The loop as data: five records every interface reads or writes.

Approved by the Board for implementation · 20 November 2026

Status

Field names are drafts and may change.

For builders, integrators and researchers.

In one paragraph

The rest of our site explains the loop in words. Developers need it as data: what records exist, what each holds, and how one leads to the next. Five records carry the loop from evidence to value. Every interface CRIS publishes will read or write one of them, so learning these five is learning the hub. Field names here are drafts (0.1) and may change.

Evidence itemObservation Meshsourcetrust scoreconsent recordheld-out flagquarantine flagProposedEᵢProofthe Provercandidatebaselineevidence setdomainscontradictionsverdict, conditionssignatureDraft 0.1↔Gate decisionC-Gateencoded purposedriftbenefit scorelineage, rollbackpromote / hold /rejectProposedCLedger entryKnowledge Ledgerwhat it isbuilds onevidencewritten bylevellink to previousDraft 0.1ΣValue recordValue Metercontributionledger linkrule appliedamountsettlementProposedΣ → valueProof and ledger fields follow the draft tables on the system pages; the others are proposals forSystems Engineering to confirm. Field names are illustrative until each format is published.
The loop as data: five records a developer will meet, with the symbol each one carries. Every interface CRIS publishes will read or write one of them.

The loop in one breath

Every CRIS system runs the same six stages: observe, verify, reason, accumulate, benefit and continue. Two gates stand between a proposed change and its use. The Correspondence Prover tests the change against reality; the C-Gate tests it against purpose. A change that fails either gate is not used, and the failure is recorded. The CRL Engine runs the loop, and the Knowledge Ledger keeps the record of everything that happens in it.

The five records

Draft 0.1. Field names may change.

RecordWritten byWhat it holdsWho reads itSpecification
Evidence itemObservation Mesh (and CIX for human observations)The observation; its source; a trust score; the consent record for a human contribution; whether it is held out for testing; whether it is quarantinedThe Correspondence Prover; the Reasoning Core (verified evidence only)Evidence format, planned
Correspondence proofCorrespondence ProverCandidate and baseline; the evidence set; the result in each domain; contradictions, with sources; the verdict (pass, staged or fail); conditions for a staged verdict; the signatureThe C-Gate; the Knowledge Ledger; the Audit SuiteProof format, draft 0.1
Gate decisionC-GateThe encoded purpose it was tested against; the drift measured; the benefit score; lineage and rollback path; promote, hold or rejectThe Knowledge Ledger; the Audit SuiteDecision format, planned
Ledger entryKnowledge LedgerWhat it is; what it builds on; the evidence; who wrote it; its level (hypothesis, supported claim or verified knowledge); a link to the previous entryThe Reasoning Core; the Value Meter; the Audit SuiteLedger entry format, draft 0.1
Value recordParticipation Value MeterThe contribution; the ledger entry that verified it; the published rule applied; the amount; settlement through a payment partnerThe contributor; the Audit Suite; public totalsValue record format, planned

Proof and ledger fields come from the draft tables on the system pages. The others are proposals for Systems Engineering to confirm, which is why their specifications are planned rather than drafted.

How one record leads to the next

An evidence item enters through the Observation Mesh, or through CIX when a person contributes, with their consent recorded alongside it. Some evidence is held out, kept back from every system so that it can test them fairly. The Reasoning Core reads only verified evidence and proposes a change. That proposal is not yet a record: it becomes one when the Prover writes a correspondence proof about it, whatever the verdict.

A passing or staged proof goes to the C-Gate, which writes a gate decision: promote, hold or reject. Every proof and every decision is written to the Knowledge Ledger as a ledger entry. When an entry verifies a person’s contribution, the Value Meter writes a value record, applying a published rule, and a licensed payment partner settles it. CRIS never holds the money.

Three ideas every developer must hold

Propose, never approve. No component approves its own output. The Reasoning Core proposes, the Prover tests against reality, the C-Gate tests against purpose. A system built to our specifications keeps that separation, and the specifications make it visible in the data: a proof is signed by the Prover and a decision by the C-Gate, never by the component that proposed the change.

Append, never overwrite. The Ledger is append-only. A correction is a new entry that points to the old one, so both stay visible. Any interface that writes to the Ledger will offer append and nothing else. Each entry links to the one before it, so any change to history would be detectable.

Levels of knowledge. A claim starts as a hypothesis, becomes a supported claim when it passes the Prover, and becomes verified knowledge when repeatedly confirmed. Only verified knowledge is ground for reasoning. Developers who read the Ledger must always check the level, and never treat a hypothesis as fact.

Symbols and their records

The book and our system pages use one symbol for each stage of the loop. Each symbol has a record.

SymbolStageRecord
EᵢObserveEvidence item
↔VerifyCorrespondence proof
IᵢReasonA proposed change, not a record until proven
ΣAccumulateLedger entry
Σ → valueBenefitValue record
CContinueGate decision

What this guide does not do

It does not show field names as final. Until each format is published as a specification at 1.0, every field table here is a draft, and names may change; the changelog records every change. It shows no real record: every example in the hub is illustrative until a CRIS system runs. And it does not describe any interface as working: see Status for what is open.

Owner: Systems EngineeringNext review: 2027-05-11