evidenced-by audit — batch 04

36 files scanned; 50 edges (23 in kb/notes/, 27 in kb/reference/). Prose mentions that are not edges: 2 (kb/reference/adr/020-theoretical-default-contrasts-mechanism.md:16, kb/sources/adaptation-of-agentic-ai-survey-post-training-memory-skills.ingest.md:30).

class count
CORROB-OK 0
QUALIFY 22
CORROB-UNTESTED 1
ORIGIN 15
MISLABEL 12

Contract findings (operator decisions)

  • The reference contract lists evidenced-by destinations as reference, sources, agent-memory-systems, agentic-systems, and external — not notes. Five reference edges point into kb/notes/ (ADR 068:85, ADR 079:103, ADR 082:183, where-change-candidates:33, :34), so they break the destination rule regardless of ADR 091.
  • The reference contract has no abstracted-from toward reference and no registered label for "the decision record this doc describes". Origin edges from reference docs to ADRs therefore fall back to see-also. The unregistered decision: label (used 8 times in reference docs) suggests an operator may want to register one.
  • The notes contract has no origin label toward external. Two notes cite Toulmin (Purdue OWL URL) as origin; the fix there is to ingest the source into kb/sources/ and use abstracted-from, or to rephrase the edge as attribution.

Non-QUALIFY edges

source file:line target class reason proposed fix
kb/reference/source-adoption-policy.md:43 agent-memory-systems/reviews/thalo.md CORROB-UNTESTED The body (line 18) says it is "treated as corroboration", but convergence on a toolbox is not a test the policy could have failed. The body itself admits "shared engineering priors". evidenced-by: an independent project that also chose compiler-like tooling for knowledge management; a parallel design choice consistent with shared engineering priors, not a test of this policy (and soften line 18's "corroboration" to match)
kb/notes/title-as-claim-enables-traversal-as-reasoning.md:78 Toulmin argument (Purdue OWL, external) ORIGIN The phrase calls Toulmin "the theory behind what this note describes". The note maps Toulmin's model onto titles and link words, so the source is the origin. Ingest to kb/sources/ and use abstracted-from: Toulmin's claim/grounds/warrant model, which this note maps onto claim titles and link words. Interim: evidenced-by: supplies the argument model this note adapts; attribution, not a test
kb/notes/writing-styles-are-strategies-for-managing-underspecification.md:52 Toulmin Argument (Purdue OWL, external) ORIGIN This is a term attribution ("warrant originates from Toulmin"), not an observation. Same ingest route with abstracted-from. Interim: evidenced-by: source of the "warrant" concept the explanatory style borrows; attribution, not a test
kb/notes/two-context-boundaries-govern-collection-operations.md:70 reference/adr/003-connect-skill-discovery-strategy.md ORIGIN ADR 003 (2026-03-05) predates the note (2026-03-08). The body reads the index boundary out of ADR 003's scaling concern, and "exactly what happens" presents that origin as confirmation. abstracted-from: ADR 003's index-first connect strategy and its stated scaling limit, the case the index boundary generalizes
kb/notes/use-tests-a-decomposition-locally-rationale-makes-transfer-testable.md:52 Parnas 1972 (external PDF) ORIGIN The body calls KWIC "the clean demonstration", but the claim is built on this case, so the case cannot test it. External targets do not allow abstracted-from. Ingest and use abstracted-from: Parnas's two working KWIC modularizations that running cannot tell apart. Interim: evidenced-by: illustrates the claim with Parnas's two working KWIC modularizations; the claim was built from this case, so it does not test it
kb/notes/verification-needs-a-typed-target-before-it-needs-an-oracle.md:64 agentic-systems/reviews/exo.md ORIGIN The Exo review and the note were created the same day (2026-07-26). The body derives "checks landed exactly where there was a class" from Exo. abstracted-from: Exo, where every declared natural-language kind carries a check and every undeclared kind carries none
kb/notes/vibe-noting.md:44 notes/evidence/vibe-noting-trace-shows-persistence-enables-revision-not-certification.md ORIGIN The trace records the episode that produced this note's current claim ("behind this note's weak claim"). The target is a note, so abstracted-from is not allowed. evidenced-by: the Commonplace episode that produced this note's current weak claim; it illustrates the claim's revision path and does not test it
kb/reference/adr/067-review-workers-read-one-prompt-and-write-one-output.md:120 reference/harness-sub-agent-model-selection-regression.md ORIGIN This incident is where the decision's split between worker claim and harness provenance came from. "shows why" presents the origin as support. see-also: the regression incident this decision's split between worker self-report and harness-observed provenance was drawn from
kb/reference/adr/079-grounding-reviews-budget-sixteen-distinct-linked-artifacts.md:103 notes/evidence/a-five-link-cap-missed-four-grounding-findings-in-twelve-reviews.md ORIGIN This is the measurement the ceiling was chosen from, not a later test. evidenced-by to notes is also outside the reference contract. rests-on: the paired outcomes, repeat baseline, and reading demand the sixteen-artifact ceiling was chosen from
kb/reference/adr/082-grounding-is-bounded-on-the-artifact-by-unquoted-sources.md:183 notes/evidence/a-five-link-cap-missed-four-grounding-findings-in-twelve-reviews.md ORIGIN This is the pre-decision measurement the ADR was built on. The destination is also not allowed for reference evidenced-by. rests-on: the paired outcomes, the uncapped arm's twenty-two-artifact reading, and the misattribution FAIL that motivated bounding on the artifact
kb/reference/collection-prototypes.md:77 reference/adr/069-collection-contract-bundles-become-one-time-prototypes.md ORIGIN The ADR is the decision this doc describes, not an observation about it. see-also: the decision record that set clone-only semantics and retired text-contract profiles
kb/reference/collections-and-types.md:122 reference/adr/018-types-are-path-references-to-instruction-docs.md ORIGIN The ADR is the decision establishing what this doc describes. see-also: the decision record establishing path-valued type identity, restored for every type by the two-root search path
kb/reference/proposals/factored-dependency-pairs-for-review-freshness.md:55 reference/adr/038-type-conformance-reviews-use-the-type-spec-as-the-gate.md ORIGIN The phrase itself says the proposal "generalizes" this instance. see-also: the shipped first instance of the factoring pattern this proposal generalizes from
kb/reference/validation-contract.md:123 reference/adr/046-verbatim-quotes-are-validated-against-their-cited-source.md ORIGIN The ADR "surfaced the class" the doc describes, so it is the origin. see-also: the decision that added the second referential check and surfaced this class
kb/reference/where-change-candidates-come-from-in-commonplace.md:33 notes/evidence/commonplace-as-a-reflective-system.md ORIGIN The phrase names it as "the trace this note generalizes beyond". The destination (notes) is also not allowed for reference evidenced-by. see-also: the ADR 026 allocation trace this description generalizes from
kb/reference/where-change-candidates-come-from-in-commonplace.md:34 notes/evidence/tag-readme-trace-observed-causal-connection.md ORIGIN This is the full version of the opening trace the doc summarizes, so it is the source case. The destination is also not allowed. see-also: the full walkthrough of the agent-assisted drafting this doc's opening trace summarizes
kb/notes/world-models-assess-explanatory-reach-through-action-conditioned.md:38 sources/why-ai-systems-dont-learn-and-what-to-do-about-it.ingest.md MISLABEL The phrase says the source "does not ground this note's … claim". It is an adjacent framing, not evidence. see-also: proposes observation and action learning as interacting subsystems under meta-control; adjacent to, not evidence for, the reach-assessment claim
kb/reference/adr/018-types-are-path-references-to-instruction-docs.md:118 reference/adr/009-link-relationship-semantics.md MISLABEL This is a neighbouring ADR the decision leaves unchanged, not evidence. see-also: the linking vocabulary embedded in cp-skill-write/SKILL.md, unchanged by this ADR
kb/reference/adr/068-collection-contracts-stop-enumerating-available-types.md:85 notes/indexes-lower-recall-when-they-suppress-retrieval-that-would-find-more.md MISLABEL The target is a theory note the decision relies on. Reference evidenced-by does not allow notes as a destination. rests-on: the recall failure the## Typesgate reproduces on the write path
kb/reference/proposals/archive/atomic-step-grounding-bounds-unquoted-sources-per-note.md:202 reference/adr/079-grounding-reviews-budget-sixteen-distinct-linked-artifacts.md MISLABEL The target is the status-quo baseline the option would keep, not evidence for it. see-also: the interim ceiling and corpus measurement this option would leave in place as a non-binding ceiling
kb/reference/proposals/archive/user-level-uv-tool-installation-for-commonplace-commands.md:40 uv tools docs (external) MISLABEL The target is the mechanism under evaluation, not evidence about it. see-also: the external tool-installation mechanism evaluated
kb/reference/proposals/artifact-function-as-a-routing-field.md:175 reference/collections-never-own-frontmatter-semantics.md MISLABEL The target is a design constraint the proposal must respect, not an observation. see-also: constraint — any persisted function field must keep one type-owned meaning across collections
kb/reference/proposals/generalized-validation-invalidation-and-imperative-extension.md:80 reference/validation-contract.md MISLABEL This is the shipped contract the proposal may extend (a companion), not evidence. see-also: the shipped owner/mechanism boundary this proposal may eventually extend
kb/reference/source-adoption-policy.md:44 agent-memory-systems/reviews/arscontexta.md MISLABEL This is an alternative the policy declines, not evidence. see-also: the cognitive-science alternative the policy acknowledges but does not adopt wholesale
kb/reference/types/adr.md:70 reference/where-change-candidates-come-from-in-commonplace.md MISLABEL The target argues that ADRs are evidence. That is the reverse of evidence for this type spec. see-also: argues that instrumented ADRs are the ongoing evidence stream for how the system's organization changes
kb/reference/validation-contract.md:124 reference/adr/024-schema-severity-is-per-constraint-fail-by-default.md MISLABEL The target is the decision that defines the schema severities, not evidence. see-also: the decision setting how the schema source assigns its own severities
kb/reference/where-change-candidates-come-from-in-commonplace.md:35 reference/tag-readme-trace-as-self-improving-loop.md MISLABEL The target is a second reading of the same trace (a companion), not an independent observation. see-also: reads the same agent-assisted drafting as the computational part of jointly performed Search
kb/reference/where-change-candidates-come-from-in-commonplace.md:37 reference/freshness-architecture.md MISLABEL The target is a related shipped mechanism, not evidence. see-also: the newer, systematic staleness-noticing mechanism, still limited to review-pair targets

QUALIFY edges

  • kb/notes/the-deployed-system-not-the-model-is-the-unit-of-learning.md:66 → sources/co-harness-co-evolving-harness-and-model-weights.ingest.md
  • kb/notes/the-deployed-system-not-the-model-is-the-unit-of-learning.md:67 → sources/meta-harness-end-to-end-optimization-of-model-harnesses.ingest.md
  • kb/notes/the-deployed-system-not-the-model-is-the-unit-of-learning.md:68 → sources/machine-studying.ingest.md
  • kb/notes/the-deployed-system-not-the-model-is-the-unit-of-learning.md:69 → sources/position-ai-agents-in-scientific-teams-as-human-agent-systems.ingest.md
  • kb/notes/topology-isolation-and-verification-form-a-causal-chain-for-reliable.md:79 → sources/knowledge-centric-self-improvement-2607.19592.ingest.md
  • kb/notes/treat-continual-learning-as-representational-form-coevolution.md:59 → agent-memory-systems/reviews/meta-harness.md
  • kb/notes/treat-continual-learning-as-representational-form-coevolution.md:60 → yoonholee.com/meta-harness/paper.pdf (the link text says "Ingest:" but the link goes to the external PDF, so check which one was meant)
  • kb/notes/treat-continual-learning-as-representational-form-coevolution.md:61 → sources/verbalizable-representations-global-workspace-llms.ingest.md
  • kb/notes/unformalized-improvements-need-a-pre-formal-stage-in-the-loop.md:59 → notes/evidence/commonplace-as-a-reflective-system.md
  • kb/notes/universal-software-factory-needs-a-declared-universality-axis.md:62 → sources/di-giovanni-padella-universal-software-factory-1983.ingest.md
  • kb/notes/use-tests-a-decomposition-locally-rationale-makes-transfer-testable.md:65 → reference/design-rationale-management.md
  • kb/notes/verifiable-subroles-before-reviewer-identity.md:63 → sources/agentic-code-reasoning.ingest.md (uses -- instead of an em dash as the separator)
  • kb/notes/warranted-autonomy-is-bounded-by-oracle-domain.md:50 → notes/evidence/commonplace-as-a-reflective-system.md
  • kb/notes/world-models-assess-explanatory-reach-through-action-conditioned.md:35 → sources/meta-v-jepa-world-model.ingest.md
  • kb/notes/world-models-assess-explanatory-reach-through-action-conditioned.md:36 → sources/meta-v-jepa-2-world-model.ingest.md
  • kb/notes/world-models-assess-explanatory-reach-through-action-conditioned.md:37 → sources/meta-v-jepa-2-publication.ingest.md
  • kb/reference/commonplace-as-an-instrument.md:70 → sources/the-anatomy-of-a-design-theory-gregor-jones-2007.ingest.md
  • kb/reference/proposals/calibrating-semantic-gates-against-labelled-fixtures.md:118 → sources/lifecycle-llm-judge-recommendation-explanations.ingest.md
  • kb/reference/proposals/calibrating-semantic-gates-against-labelled-fixtures.md:119 → sources/improving-ai-skills-with-autoresearch-evals-skills-203525743436.ingest.md
  • kb/reference/proposals/deterministic-write-context-assembly.md:136 → reference/collections-and-types.md
  • kb/reference/proposals/generalized-validation-invalidation-and-imperative-extension.md:82 → reference/adr/038-type-conformance-reviews-use-the-type-spec-as-the-gate.md
  • kb/reference/proposals/trajectory-aware-evaluation-of-transforming-agent-workflows.md:111 → sources/agent-harness-large-language-model-agents-survey.ingest.md