Operator · External observation

Observation Mesh

Reality, with its sources shown.

Approved by the Board for implementation · 20 November 2026

In one paragraph

The Observation Mesh keeps every CRIS system learning from the world, not from itself. It takes in evidence from people, sensors, markets, scientific instruments and institutions; records where each piece came from and under what consent; scores how far each source can be trusted; and holds part of the evidence back, so every change a system makes to itself can be tested on evidence it has never seen.

It carries Eᵢ, external observation, the first operator of the Fourth Law, and it is the first system the CRL Engine calls in every cycle.

Where it sits

The mesh carries the first stage of the loop, Observe. What it delivers feeds every other stage.

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 →

Gives toWhat it givesWhy
CRL EngineFresh evidence for every cycle (Workflow A, step 2)No cycle starts from the system's own outputs
Correspondence ProverHeld-out evidence from the vaultEach candidate is tested on evidence it has never seen
Reasoning CoreThe training poolProposals are grounded in real observations
Knowledge LedgerProvenance and split recordsAuditors can confirm what was used and how

It receives verified human observations, with consent records, from CIX; sensor, market, instrument and institutional data from partners and public sources; and agreement and contradiction results from the Correspondence Prover, which move trust scores.

Why it matters

A system that learns only from its own outputs slowly drifts away from reality. Each version inherits the last one's mistakes and adds its own. The Observation Mesh is the anchor that stops this: every improvement cycle must start from fresh evidence about the world, with a record of where it came from.

Without the mesh

  1. v1
  2. v2
  3. v3
  4. v4

Each version trains on the last version’s outputs; errors compound and the model separates from the world.

With the mesh

  1. v1
  2. v2
  3. v3
  4. v4

Every cycle starts from fresh external evidence with provenance; a minimum share of fresh evidence is required in each cycle.

A mind that learns only from itself drifts; fresh observation keeps it anchored (book Chapter 7.1).

The book calls this the danger of self-referential learning, and it is the first rupture, correspondence drift, in action. Every cycle must therefore include a minimum share of fresh external evidence, set by our board. The mesh also guards against the fourth rupture, human severance: by treating people as a required source of evidence, it keeps people inside the improvement loop rather than outside 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 correspondence drift and human severance on the ruptures page.

Five kinds of source

The mesh listens to five kinds of source. Each brings its own risks, so each has its own main check.

Examples are illustrative, not live feeds
  • PeopleThrough CIX and partnerse.g. Local water conditionsCheck: Consent, track record
  • SensorsPhysical devicese.g. Weather stationsCheck: Calibration, uptime
  • Economic signalsMarkets and pricese.g. Food pricesCheck: Source licence, lag
  • InstrumentsScientific measuremente.g. Lab resultsCheck: Method, peer review
  • InstitutionsOfficial recordse.g. Health statisticsCheck: Publisher, date
Observation MeshRegisters, scores, quarantines, separates held-out evidence, and serves the CRL Engine
The mesh listens to five kinds of source, each with its own risks (book Chapter 7.2).
SourceExamples (illustrative)Main riskMain check
PeopleLocal water or crop conditions reported through CIXError, bias, manipulationConsent, track record, cross-checking
SensorsWeather stations, energy metersFaults, driftCalibration, uptime
Economic signalsMarket prices, trade dataLag, licensingSource licence, publication date
Scientific instrumentsLaboratory measurementsMethod errorsMethod record, peer review
InstitutionsHealth and census statisticsOut-of-date or revised figuresPublisher, version, date

How it works

Every observation passes through five steps before any system learns from it: register, score, quarantine, separate and update.

  1. 1RegisterSource, owner, consent basis
  2. 2ScoreInitial trust from type and history
  3. 3QuarantineNew evidence waits in a holding store
  4. 4SeparateHeld-out vault vs training pool
  5. 5UpdateTrust rises or falls with the Prover’s results
TriggerA new source is proposed, or a CRL cycle asks for evidence
OutputAn evidence stream the system did not author, each item with provenance
NeverHeld-out evidence is never used to train the system it tests
Every observation passes through five steps before any system learns from it (CRIS document, Section 9).
  1. Register the source and record its provenance, owner and consent basis.
  2. Score it with an initial trust level from its type and history.
  3. Quarantine new evidence in a holding store until format and source are checked.
  4. Separate held-out evidence into a vault, so it can never be used to train the system being tested.
  5. Update trust scores as the Correspondence Prover confirms or contradicts each source.

Provenance

Every observation carries a record of where it came from, who owns it, under what consent it was given, how it was measured, and how far it is currently trusted. The book's phrase is "knowing where evidence came from and how far to trust it".

Provenance record · proposed fields
Observation ID
A unique reference
Kind of source
Person, sensor, market, instrument or institution
Source
The registered source it came from
Owner
Who holds the rights
Consent
The basis and scope of consent
Captured
Time and place
Method
How it was measured
Trust
The source’s current trust score
Status
Quarantine, training pool or held-out vault
Licence
Terms of use
Tamper check
A fingerprint that shows if the record was altered
Why provenance“Knowing where evidence came from and how far to trust it” (book Chapter 7.3).
Who can see itAuditors see full records; the public sees totals only.
Personal dataFor observations from people, identity is stored separately from content, and only to the level the task needs.
Each observation carries its own record of where it came from. Field names are shown in plain words; the specification will define the exact format.

Held-out evidence

Part of the evidence goes into a vault. The system being tested never sees it; only the Correspondence Prover reads it, to check that each proposed change still matches reality. Held-out evidence is never used for training. The split is recorded, so auditors can confirm that no vault evidence was used for training.

QuarantineNew, verified for format and source
Training poolMay be used by the Reasoning Core to propose and train changes
Held-out vaultRead only by the Correspondence Prover to test candidates; never shown to the system under test
Fair testEach candidate is judged on evidence it has never seen
Held-out evidence lives in a vault the system can never train on. The split is fixed for each cycle and recorded in the Knowledge Ledger, so auditors can confirm no vault item reached training.

Trust scores

Trust is earned. Each source starts with a score based on its type and history. When other evidence confirms it, the score rises; when evidence contradicts it, the score falls and the items are reviewed. A source below the threshold set by our board is paused, and its owner is told why and how to appeal.

Illustrative — no real scores
  1. StartInitial score from source type and history
  2. ConfirmedThe Prover finds agreement with other evidence: the score rises
  3. ContradictedThe Prover finds conflict: the score falls and items are flagged
  4. Below thresholdThe source is paused; its owner is told why and may appeal
Trust is earned: a source’s score moves with the evidence, and below the board-set threshold the source is paused.
EventEffect on trustWho is told
New source registeredInitial score from type and historySource owner
Prover confirms agreementScore risesRecorded in the ledger
Prover finds a contradictionScore falls; items flagged for reviewSource owner, with reason
Score below the board-set thresholdSource pausedSource owner, with appeal route
Appeal upheldScore restoredSource owner

People and privacy

Awaiting counsel. People-and-privacy wording is a draft. Counsel and the privacy lead must approve this block before the page is published.

Many of the most valuable observations come from people. When you contribute through the Continuity Intelligence Exchange (CIX), you choose what to share, you keep ownership of your contribution, and you are told why if an observation is not accepted. Your identity is stored separately from what you observe, and only to the level the task needs. You can leave at any time. Read the participant charter. Read our privacy page.

This follows the participant protection charter in the CRIS document: consent first, ownership kept by the participant, explanations for rejected items, the right to leave, and a route to appeal.

  1. 1Join CIXVia the web or the CES app
  2. 2Consent and IDPrivacy and ownership
  3. 3ContributeAn observation or insight
  4. 4Verified?Correspondence Prover · ↔No → feedback, try again
  5. 5Enter ledgerKnowledge Ledger · Σ
  6. 6Value MeterScores the contribution
  7. 7RewardPaid; reputation rises
  8. 8Benefit reportAggregate and public
On CIX, only verified contributions earn rewards. More on CIX →

Status and read more

This status is set by the CRIS team and links to the evidence behind it. The CRIS roadmap places the first CRL prototype and the closed CIX pilot in 2027; the mesh moves up a level only as each level's evidence is published.

Current status: Status to be confirmed

LevelEvidence requiredWho sets it
ResearchConcept page and draft provenance specification publishedCEO
PrototypeThe mesh registers sources and separates held-out evidence for a test CRL cycle; ledger records existCEO, on the Labs and Systems leads’ advice
PilotLive partner sources and CIX pilot observations flow in on testnet; counsel has approved consent flows; Audit Council reviewCEO
LiveProduction sources; external audit of provenance and the vault split passedCEO, with notice to the board

For developers: Concepts: the loop as data (the evidence item record; its format is planned).

Next steps

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