Operator · Continuity

C-Gate

Capability alone never earns succession.

Approved by the Board for implementation · 20 November 2026

In one paragraph

The C-Gate is the last test a new version of a CRIS system must pass before it replaces the old one. It checks that the new version still pursues the purpose its board approved, that it has not drifted from that purpose over many small updates, and that it benefits people and the environment. A version that is more capable but fails any of these is not promoted. Capability alone never earns succession.

It carries C, continuity, the fifth operator of the Fourth Law. Its output is a succession decision, with the new version's lineage and a tested way back to the old one.

Where it sits

The C operator has two stages of the loop. At Benefit the C-Gate asks whether a change improves outcomes for people and the environment; at Continue it asks whether the successor is still the same system in purpose. In the CRL Engine's Workflow A it is Gate 2 (step 5), after the proof and the ledger entry.

Knowledge today.
A more continuous tomorrow.
  1. Observe

    Observe the world as it is, and record where each observation came from.

    Learn more about observe →

  2. Verify

    Verify every observation, and every change a system makes to itself, against evidence it has not seen.

    Learn more about verify →

  3. Reason

    Reason only from what has been verified, in steps someone else can trace.

    Learn more about reason →

  4. Accumulate

    Accumulate what was verified, so knowledge adds up instead of being overwritten.

    Learn more about accumulate →

  5. Benefit

    Benefit people and the environment: participants share in the value they create, and every change is weighed against our public benefit purpose.

    Learn more about benefit →

  6. Continue

    Continue, changing anything except correspondence with reality and continuity of purpose.

    Learn more about continue →

  1. 1Propose a changeReasoning Core · Iᵢ
  2. 2Gather fresh evidenceObservation Mesh · Eᵢ
  3. Gate 1Proof passes?Correspondence Prover · ↔No → reject and record
  4. 4Record in ledgerKnowledge Ledger · Σ
  5. Gate 2C-Gate passes?Purpose and benefit · CNo → reject and record
  6. 6Staged rolloutSandbox, testnet, then partners
  7. 7Promote successorWith a rehearsed rollback path
Reject and recordA failure at either gate is written to the Knowledge Ledger and kept as knowledge. The next cycle restarts at step 2.
Workflow A from the CRIS document: a change becomes the successor only after two gates. Both failure paths end in the ledger, so a rejected upgrade still teaches the next cycle.

Rules for this system are proposed in the AI Constitution, Articles 7 and 8: encoded purpose and purpose tests. They are approved by the Board for implementation.

Two gates, two questions

A new version must pass two gates, and they ask different questions. The Correspondence Prover asks whether the candidate still matches reality. The C-Gate asks whether it is still the same system in purpose, and whether it benefits people.

Gate 1 · Correspondence Prover
Question
Does it still match reality?
Tests against
Held-out evidence it has never seen
Output
A correspondence proof: pass, fail or staged
Gate 2 · C-Gate
Question
Is it still the same system in purpose, and does it benefit people?
Tests against
The encoded purpose and the public benefit scorecard
Output
A succession decision: promote, hold or reject
Both gates must pass. A candidate that fails either one is recorded in the Knowledge Ledger and is not promoted.

How it works

Every succession decision follows the same six steps.

  1. 1LoadThe encoded purpose, as approved by the board
  2. 2TestRun purpose scenario suites on the candidate
  3. 3DetectGoal drift against the predecessor and the original purpose
  4. 4ScorePublic benefit, on the board’s scorecard
  5. 5DecidePromote, hold or reject; major versions go to the Audit Council
  6. 6RecordLineage; confirm the rollback path works
TriggerA candidate that passed the Correspondence Prover
OutputA succession decision, with lineage and rollback
NeverCapability alone never earns succession
The C-Gate workflow (CRIS document, Section 9).
  1. Load the encoded purpose of the current system, as approved by the board.
  2. Test the candidate with purpose scenario suites that check it still pursues that purpose.
  3. Detect goal drift against the predecessor's behaviour and the original purpose.
  4. Score public benefit impact, using the Public Benefit Office's scorecard.
  5. Decide: promote, hold or reject. Major versions also go to the Audit Council.
  6. Record lineage, and confirm the rollback path works before promotion.

The trigger is a candidate that has passed the Correspondence Prover.

Capability is not enough

A new version may be faster, larger and better on every benchmark, and still be rejected. Our rule is "no succession on capability alone" (book Chapter 10.6): a stronger version that has drifted from its purpose, or fails the benefit test, is not promoted. The chart is an illustration, not a real result.

Illustrative — not real versions or results
Purpose holds ACurrent version BStronger, purpose holds → may be promoted
Purpose drifted CMuch stronger, drifted → rejected
A more capable version that has drifted from its purpose is rejected. No succession on capability alone (book Chapter 10.6).

The encoded purpose

Every CRIS system has an encoded purpose: what it is for, written as tests it must keep passing. Our board approves it. Only the board can change it, after consulting the Public Benefit Office, the Audit Council and the participants it affects, with a written rationale and a public notice period. The system cannot change its own purpose, and neither can the engineers who build it.

How a purpose is set

  1. Board approvesThe encoded purpose of each system
  2. Written as testsScenario suites the candidate must pass
  3. C-Gate loads itAt step 1 of every decision

To change a purpose

  1. Board decidesNot the system, not its engineers
  2. Must consultPublic Benefit Office, Audit Council and affected participants
  3. EvidenceWritten rationale and a public notice period
The encoded purpose: what the gate tests against. It is set, and changed, only by the board (CRIS document, Section 10).

Drift against the original

A system rarely changes its purpose in one leap. It drifts, a little with each update. If each version were compared only with the one before it, every step could look safe while the whole moved far from where it began. That is why the C-Gate compares every new version not only with its parent but with the original purpose, and a version that has drifted beyond the tolerance the board has approved fails, however small its own step. The chart is an illustration, not a real result.

Illustrative — not real data
Drift from the parentEach step is small; alone, it would pass
Drift from the originalAdds up; later versions exceed the tolerance and fail
Small updates can add up to a large drift. The C-Gate measures both, so a version that has drifted too far from the original purpose fails, however small its own step.

Goal drift across many small updates is named in our risk register (book Chapters 2.2 and 10.3). It is the rupture the book calls goal discontinuity, and the C-Gate is the guard against it.

Rupture (Chapter 2)Guard in the loopCRIS system
Correspondence driftThe system's model slowly separates from the world.↔ Correspondence proofEach change tested against realityCorrespondence ProverObservation Mesh
Goal discontinuityIts purpose is reinterpreted at each rewrite until it survives in name only.C Purpose testA successor must keep its purposeC-Gate
Knowledge erosionVerified knowledge is overwritten by unverified assumptions.Σ Promotion rulesOnly verified claims persistKnowledge Ledger
Human severanceThe improvement loop closes around itself and people become bystanders.Eᵢ Human observationPeople stay a source of realityCIXObservation Mesh
The four ways a self-improving system can lose continuity, and the CRIS systems designed to guard against each. Select a guard to see how it is designed to work.

Read about goal discontinuity on the ruptures page.

Benefit score

As a public benefit corporation, CRIS must weigh public benefit alongside its other interests. The C-Gate builds that duty into every decision: at step 4 each new version is scored on a public benefit scorecard approved by our board and run by the Public Benefit Office. The scorecard's weightings will be shown once the board publishes them.

BoardApproves the scorecard and its weightings
Public Benefit OfficeRuns the score at every decision
  • PeopleParticipants, value shared
  • SocietyOutcomes of applied programmes
  • EnvironmentEnergy per capability gain, in metered CTU
  • ContinuityComplete proofs, gates and rollbacks
Benefit scorePart of the C-Gate report
Step 4 uses the scorecard the board approves. Weightings are not shown until the board publishes them.
Awaiting counsel. The public benefit purpose from our charter will be quoted here, in the wording approved by legal counsel.

Decisions and who makes them

The C-Gate decides one of three things: promote, hold or reject.

  1. PromotePurpose holds, no drift beyond tolerance, benefit score passes, rollback tested
    NextStaged roll-out: sandbox, testnet nodes, then limited partners
  2. HoldA question needs people, under the board’s escalation rules
    NextThe review board or the Audit Council decides, with the evidence
  3. RejectThe purpose test failed, or drift was found
    NextRecorded in the Knowledge Ledger with the reason; the next cycle reads it
What the C-Gate can decide, and what happens next.
DecisionWhenWhat happens next
PromotePurpose holds, no drift beyond tolerance, the benefit score passes and rollback is testedStaged roll-out: sandbox, testnet nodes, then limited partners
HoldA question needs people, under the board's escalation rulesThe review board or the Audit Council decides, with the evidence
RejectThe purpose test failed, or drift was foundRecorded in the Knowledge Ledger with the reason

Every release is reviewed by people: our engineering and research leads and our Public Benefit Office, sitting as the C-Gate review board. Major versions also go to an independent Audit Council.

  • Every changeThe C-Gate runs automaticallyMachine speed
  • Every releaseC-Gate review board: Systems Engineering lead, Continuity Labs lead, Public Benefit OfficeRelease speed
  • Major versionsAlso the independent Audit CouncilIndependent review
  • EscalationRules approved by the board send a decision to peopleBoard-set
  • A change of purposeThe board, after consultation and public noticeInstitutional speed
Who decides at the C-Gate: machine, release and institutional speed (CRIS document, “Who decides what”).
DecisionWho decidesMust consultEvidence
Promote a new versionC-Gate review board: Systems Engineering lead, Continuity Labs lead, Public Benefit OfficeThe Audit Council, for major versionsCorrespondence proof, C-Gate report, ledger entry
Change a system's purposeThe boardPublic Benefit Office, Audit Council, affected participantsWritten rationale, public notice period

Before any version is promoted, we prove that we can return to the one before it. The rollback is rehearsed in a test; if it does not work, the version is held.

  1. Record lineageWhich version, which proofs, which decision
  2. Rehearse rollbackRestore the predecessor in a test
  3. Rollback works?Yes → promote; no → hold
  4. Predecessor archivedWith its rollback path
No promotion without a rehearsed rollback: every step is reversible (book Chapter 10.5). The CRL Engine carries out the staged roll-out and any rollback.

The C-Gate is the second of five layers of assurance, from automated checks to the public report.

  1. Automated checksProofs and drift tests
  2. Human reviewC-Gate board, every release
  3. Independent auditAudit Council, major versions
  4. External standardThird-party certification
  5. Public reportBenefit report, every year
Five layers of assurance around the engine, from machine speed to public view.

Status and read more

This status is set by the CRIS team and links to the evidence behind it. The CRIS 2027 plan includes the first version of the C-Gate in the CRL prototype.

Current status: Status to be confirmed

LevelEvidence requiredWho sets it
ResearchConcept page; draft C-Gate report format publishedCEO
PrototypeC-Gate v1 decisions in a test CRL cycle, with purpose tests, drift checks and rollback rehearsalsCEO, on the Labs and Systems leads’ advice
PilotUsed for a partner deployment on testnet; review board decisions; Audit Council review of a major versionCEO
LiveProduction use; external assessment passedCEO, with notice to the board

For developers: Concepts: the loop as data (the gate decision format is planned).

Next steps

Owner: Head of Systems EngineeringNext review: 2027-03-01