KB maintenance

Type: kb/types/tag-readme.md

How an agent-operated KB stays healthy as it grows. Detection, operations, and the dynamics that govern quality over time. For how the KB is built, see tags. For document structure, see document-system.

Dynamics

• maintenance-capacity-must-match-harmful-artifact-inflow — stable quality requires prevention, containment, detection, and repair capacity to keep pace with risk-weighted harmful-artifact inflow; gross generation is only a proxy
• traversal-improves-the-graph — every traversal is a read-write opportunity; agents should log improvement opportunities during reading, then process them separately to avoid context-switching
• title-as-claim-exposes-commitments-enabling-popperian-maintenance — claim titles make maintenance cheap: scan the index, ask "do I still believe this?", open only the doubtful ones

Detection

• quality-signals-for-kb-evaluation — composite oracle from graph-topology, content-proxy, and LLM-hybrid signals; the evaluation layer the learning loop needs
• notes-need-quality-scores-to-scale-curation — recomputable note scores (type, inbound links, review vetting, recency) filter /connect candidates and truncate budget-bounded listings as the KB grows
• semantic-review-catches-content-errors-that-structural-validation-cannot — four semantic checks (enumeration completeness, grounding alignment, boundary-case coverage, internal consistency) that require LLM adversarial reading
• link-graph-plus-timestamps-enables-make-like-staleness-detection — existing links encode dependencies; comparing note and target timestamps flags staleness without new annotation
• indexes-lower-recall-when-they-suppress-retrieval-that-would-find-more — an apparently complete index lowers route-specific recall when it suppresses retrieval that would find more
• a-derived-copy-of-recomputable-truth-must-be-checked-or-absent — a copy of mechanically recomputable information must be validator-checked against its source or not exist; hand-maintained-and-trusted is the forbidden middle
• seven documentation cases left routing and synthesis — a bounded Commonplace sweep where direct source access removed exact-fact prose while checked discovery and cross-component synthesis survived
• final task success does not establish intended-path health — identical terminal outcomes can conceal broken prescribed paths; maintenance needs independent path events
• domain-pricing-routes-an-exception-to-idealization-assessment — separates truth verdicts from repair dispositions: pricing routes a defeated claim to idealization assessment, and a retained idealization carries an adequacy record later passes can attack

Operations

• maintenance-operations-catalogue-should-stage-stable-procedures — staging catalogue for periodic operations before they are turned into reusable procedures
• periodic-kb-hygiene-should-be-externally-triggered-not-embedded-in-routing — periodic audits belong in externally triggered operations, not always-loaded routing docs
• gate-learning-from-accepted-edits — proposal: turn accepted edit diffs into review-gate candidates with a promotion/rollback lifecycle and budget-bounded loading; the oracle constraint is an-accepted-edit-verifies-the-change-not-the-rule

Related Tags

• tags — parent area: architecture and design of the KB itself
• document-system — type system and validation that maintenance operations check against
• links — linking methodology that staleness detection and quality signals operate on

Other tagged notes

• A five-link cap missed four grounding findings in twelve reviews - A paired Commonplace assay found five capped-versus-uncapped grounding outcome divergences; one reproduced as reviewer noise, while four appeared only after fuller reading reached 6–16 linked artifacts
• A linked note discharges its own grounding, so a citing note owes representation, not re-grounding - A cited source imposes a grounding obligation; a claim-titled note that already passed its own grounding review imposes only a representation obligation — with the preconditions that keep the distinction and why it is not a paraphrase ledger
• A note is an atomic step relative to the check that reads it - Two independent bounds on a note: one claim sized to the reader's bounded context, and one checkable inference sized to the checker's single pass — for the grounding check the unit is the unquoted source
• A quotes-route rollout grounded more claim uses without earning claim identifiers - A Commonplace grounding rollout recorded 30% grounded claim uses under a paraphrased claims ledger and 75% under verbatim quotes or pinned snapshots, with no case needing claim identifiers; the non-random cohorts make the gap descriptive
• Ad hoc explanation can be rational when error is cheap and local - Explains why a disposable local guess can rationally select the next probe when error is cheap and contained, while retained explanations need reach checks
• An enforced tag-README combines a MOC pattern with checked membership - A Commonplace tag-README can inherit Milo's contextual mapping pattern while validation checks only its declared membership relations, not editorial quality.
• An independent pass tightened three of four Pirolli grounding verdicts - A Commonplace grounding case changed three of four support verdicts after separating source reconstruction from target-claim judgment, making bilateral isolation a candidate control rather than a proven cause.
• Attempted recovery identifies informational gaps, not provenance or authority - Recovery failure shows content is missing from the tested source; causal provenance and live authority require independent evidence, and only pairs with unique content on both sides are bidirectionally irrecoverable
• Brainstorming: how explanatory-reach informs KB design - Deutsch's reach, registered here as explanatory-reach, applied to KB notes — a maintenance risk signal, not a retrieval signal, because high-explanatory-reach revisions break downstream reasoning silently
• Cheap adoption and weak retirement accumulate cost - Explains why locally cheap structural additions become routing, maintenance, and migration debt when retirement needs distributed evidence and lacks an equally operative path.
• Claim modality is the inference form of the refuter - The three claim modes are refuter-defined images of deduction, induction, and comparative abduction; grounds the mode list's closure for empirical claims and gives vacuity and genre drift precise readings
• Commitment, not derivation, creates new ground truth - Derivation — claims recoverable from the source, nothing added — leaves the source as ground truth; what adds unentailed resolutions becomes ground truth at commit, repaired by supersession
• Criteria edits invalidate verdicts; process edits invalidate artifacts - Editing quality criteria invalidates verdicts; editing production processes calls for artifact regeneration — verdict freshness includes the artifact and criteria but excludes its production process
• Index completeness does not determine editorial orientation - Complete generated listings establish membership, but their inputs do not determine topic-specific grouping, role phrases, or reading order without editorial judgment
• Single-artifact review bundles still cut Claude costs substantially after cache-aware weighting - April 2-4, 2026 review telemetry reweighted with Anthropic Opus 4.6 prompt-caching prices still shows a substantial cost drop from the single-artifact bundle refactor
• Superseded choices need a historical witness; refuted beliefs lose subject-matter standing - Retention after supersession follows remaining truth role rather than maintenance operation: preserve a witness to the choice event, while a refuted belief loses subject-matter standing
• Title as claim makes overlap between notes visible - When note titles are claims, overlap between notes is visible at the index level — similar assertions are obvious without opening files; topical titles hide overlap behind different labels for the same territory