EVIDENCE / CLAIM CONTAINMENT

A false claim is contained through its support graph.

Verification is claim-type-specific. When evidence contradicts a claim, every known consumer is marked atomically before stale work can advance.

CONCEPT SOURCES
SRC-V-06

CLAIM SAFETY / ATOMIC CONTAINMENTCLAIM / C-17 → C-21

Poison follows support edges to T-4, T-8, O-3.

  1. Claim C-17 is proposedClaim C-17 enters with provenance and begins epistemically unverified, authoritatively proposed, and current.
  2. Known consumers cite the proposed claimT-4, T-8, O-3 record support edges to C-17; each becomes a causal consumer rather than an awareness-only reader.
  3. New evidence disputes C-17Claim-type verification produces contradictory evidence under a recorded method, environment, result, and artifact identity.
  4. Policy or review marks C-17 poisonedAn append-only ledger event atomically commits the poison decision as a poison epoch with the known affected set, and preserves the earlier proposal and evidence history.
  5. Known consumers inherit a poisoned dependencyT-4, T-8, O-3 receive a visible poisoned-support state. Consumers may learn of the correction asynchronously; the poison epoch, not instantaneous propagation, preserves the safety property.
  6. The poison-epoch fence stops stale effectsOutputs are quarantined, active work is recalled, and any worker, result, or merge candidate that has not reconciled beyond the poison epoch is forbidden from committing.
  7. Repair and retest tasks are issuedThe manager creates scoped repair, re-derivation, and retest orders with required evidence for each affected consumer.
  8. Replacement claim C-21 is verifiedC-21 records the corrected claim, links itself as superseding C-17, and is ratified only if policy permits.
  9. Epoch N+1: revised sealed orders issuedThe official ledger decision records C-21, advances Epoch N+1, invalidates stale orders, and ensures current consumers receive revised current sealed orders.

Every evidence record carries

  • artifact ID
  • source
  • date
  • environment
  • claim tested
  • method
  • result
  • epistemic state
  • authority state
  • currency
  • consumers
  • establishes
  • does not establish
  • supersedes/superseded-by links

Epoch N+1: revised sealed orders issued. Containment ends with the replacement claim in the ledger and revised current sealed orders—not with poison classification alone.

TEACHING PLOT / C-17

False-claim containment follows support, not confidence.

Verification is claim-type-specific. When evidence contradicts a claim, every known consumer is marked atomically before stale work can advance.

EVENTS / intake → policy → ratification → frontier → order → evidence → commitment → epoch

DECISION RECORD / CLAIM C-17

Decision. Consequential statements are explicit claims with provenance, and they start unverified. Provenance is not truth.

Rejected. Model confidence, repetition, or fluent reasoning as evidence.

  • INVARIANT 02 Agent-generated claims start unverified.
  • INVARIANT 05 Model confidence and self-explanation are not evidence.

Evidence register

DECISION RECORD / KNOWN SUPPORT CONSUMERS

Decision. Support reads create causal edges that containment can later follow; awareness reads are logged but create no dependency.

Rejected. Undifferentiated reads: either every byte is a dependency and the graph is unusable, or none is and containment is blind.

  • INVARIANT 10 Unknown dependencies trigger conservative tainting, broader invalidation, retesting, or review, not invented precision.

Claim verification and poison containment

DECISION RECORD / CLAIM-TYPE VERIFIER

Decision. Verification matches the claim type: reproducible benchmarks for performance, contract tests for interfaces, archived primary sources for external-library claims.

Rejected. One generic confidence score standing in for evidence.

  • INVARIANT 05 Model confidence and self-explanation are not evidence.

Claim verification and poison containment

DECISION RECORD / POISON DECISION

Decision. Poisoning atomically commits a poison epoch and the known affected set; the false claim's history is preserved, never rewritten.

Rejected. Instantaneous global propagation. A distributed system cannot promise it, so it cannot be the safety property.

  • INVARIANT 13 A poisoned claim commits a poison epoch and affected set; consumers behind that epoch cannot commit.
  • INVARIANT 19 Superseded intent is retained and linked to its successor; history is never silently erased.

Claim verification and poison containment

DECISION RECORD / OPERATIONAL CONTAINMENT

Decision. Consumers that have not reconciled beyond the poison epoch cannot commit; newly discovered consumers are appended through later versioned events.

Rejected. Label-only poisoning with no operational fence on commitment.

  • INVARIANT 13 A poisoned claim commits a poison epoch and affected set; consumers behind that epoch cannot commit.
  • INVARIANT 14 Uncertainty blocks official commitment, not harmless local exploration.

Evidence register

DECISION RECORD / REPLACEMENT C-21

Decision. The replacement claim supersedes the poisoned one with links in both directions; consumers proceed only from the replacement support set and the current epoch.

Rejected. Silent deletion of the false claim, which would erase the audit trail with it.

  • INVARIANT 19 Superseded intent is retained and linked to its successor; history is never silently erased.

Evidence register

State 1 of 9: Claim C-17 is proposed

STATE 01 / 09

Claim C-17 is proposed

Claim C-17 enters with provenance and begins epistemically unverified, authoritatively proposed, and current.

Authority remains proposed; provenance is required, but confidence and repetition are not proof.

Read the complete textual equivalent
  1. Claim C-17 is proposed. Claim C-17 enters with provenance and begins epistemically unverified, authoritatively proposed, and current. Authority remains proposed; provenance is required, but confidence and repetition are not proof.
  2. Known consumers cite the proposed claim. T-4, T-8, O-3 record support edges to C-17; each becomes a causal consumer rather than an awareness-only reader. Support reads create an auditable evidence dependency.
  3. New evidence disputes C-17. Claim-type verification produces contradictory evidence under a recorded method, environment, result, and artifact identity. Verification changes epistemic state, not authority by itself.
  4. Policy or review marks C-17 poisoned. An append-only ledger event atomically commits the poison decision as a poison epoch with the known affected set, and preserves the earlier proposal and evidence history. The false claim is never silently deleted or rewritten.
  5. Known consumers inherit a poisoned dependency. T-4, T-8, O-3 receive a visible poisoned-support state. Consumers may learn of the correction asynchronously; the poison epoch, not instantaneous propagation, preserves the safety property. Unaffected records without a support edge remain current; newly discovered consumers are appended through later versioned events.
  6. The poison-epoch fence stops stale effects. Outputs are quarantined, active work is recalled, and any worker, result, or merge candidate that has not reconciled beyond the poison epoch is forbidden from committing. Containment is an operational consequence, not merely a new label.
  7. Repair and retest tasks are issued. The manager creates scoped repair, re-derivation, and retest orders with required evidence for each affected consumer. Salvage retains evidence and useful analysis, not stale conclusions.
  8. Replacement claim C-21 is verified. C-21 records the corrected claim, links itself as superseding C-17, and is ratified only if policy permits. Epistemic verification, authority, and currency remain independent.
  9. Epoch N+1: revised sealed orders issued. The official ledger decision records C-21, advances Epoch N+1, invalidates stale orders, and ensures current consumers receive revised current sealed orders. T-4, T-8, O-3 can proceed only from the replacement support set and current epoch.

Epoch N+1: revised sealed orders issued

Designed and built by Jeff Sylvan ·

Provenance is not truth.

Lineage records where a claim came from, who or what produced it, and which evidence was attached. Reuse still requires claim-type verification, independent reproduction where appropriate, contradiction handling, expiry, and revocation. Every record below therefore states both what it establishes and what it does not.

REC-BLACK-SKIES-INTENT-2026-07-12

Black Skies carries a ratified long-term intent of 10,000 simultaneous captains at full simulation speed.

Epistemic
unverified
Authority
ratified
Currency
current
Environment
Black Skies product planning
Claim tested
The 10,000-captain statement is an authorized direction rather than a demonstrated benchmark result.
Method
Ratified-intent record inspection
Result
The target is recorded as current ratified intent; no achievement artifact is attached.
Consumers
/black-skies/, /reconciliation/, /limits/
Supersession
Supersedes no prior record · Current unless its currency says otherwise

Establishes
The target is an authorized direction and a useful pressure test for the development control plane.

Does not establish
That this performance has been achieved or benchmarked.

Source: Black Skies product intent

REC-PUBLIC-SPECIFICATION-2026-07-13

The public specification defines the intended PlotWarden end-state model.

Epistemic
verified
Authority
ratified
Currency
current
Environment
Public source-aligned website
Claim tested
The intended graphs, authority boundary, reconciliation behavior, and safety invariants are defined publicly.
Method
Specification and terminology-contract inspection
Result
The normative terms and boundaries are present in the public specification.
Consumers
/system/, /authority/, /reconciliation/, /workers/, /evidence/
Supersession
Supersedes no prior record · Current unless its currency says otherwise

Establishes
The intended graphs, authority boundary, reconciliation behavior, and safety invariants are explicit and reviewable.

Does not establish
That every intended behavior is implemented or production-proven.

Source: PlotWarden public specification

Competitive observations expire.

Capabilities are defined before tools are compared. Official sources are preferred. Findings are dated and caveated so “different today” does not become “unique forever.”

Scroll horizontally to inspect every observation field.

ToolVersion / observedCapability definitionEvidenceFindingCaveat / source
GitHub Copilot coding agentsdocumentation observed 2026-07-12
2026-07-12		Asynchronous issue-to-pull-request coding work with review and repository safety controls.Official GitHub documentation describes assignment, branch-scoped work, pull request creation, review, and security controls.Strong overlap in scoped execution and human-reviewed repository effects; PlotWarden's thesis concerns cross-source living intent and change-in-flight reconciliation around such workers.This is a point-in-time reading of public documentation, not a complete product evaluation or proof of durable differentiation.
Open official source

Read the public specification