Obvious distillation cases

First staging pass for the library-side distill* migration. Corpus: kb/notes/**/*.md and kb/reference/**/*.md, excluding sources and generated reports. Sweep result on this branch: 110 files, 377 matching lines.

This is not a full classification table. Its job is to find the notes where the control regime is already obvious enough to schedule an early wave, and to mark the notes that should not be touched by a quick rewrite.

Mapping to the migration plan's waves (the two files cut the work differently — this file stages by obviousness, the plan by dependency): the "first semantic wave" below was Wave-1-shaped AMP content executed early (done, commit 80f47b41); the "second semantic wave" below is the library subset of plan Wave 2 — unblocked now that Wave 0 is complete, independent of plan Wave 1; the "deliberately not first-wave" list is plan Wave 1 proper (entangled, review-gated — not batch work); the META section is plan Wave 3.

Rule for the first pass

An "obvious" case is one where the replacement term follows from the note's own argument without needing the promoted definition notes in hand:

  • Discovery / abstraction when a trace, episode, accepted edit, or repeated behavior becomes a reusable rule.
  • Derivation / selection when already-recorded reasoning is reshaped into a task-facing artifact without adding a new substantive claim.
  • META / infrastructure when the occurrence is a tag, definition page, link label, filename, or navigation surface.

Mixed cases stay out of the first pass even if one sentence looks easy.

First semantic wave: obvious discovery / abstraction

These should be fixed before the broad derivation rename because they are the original mistake: using distillation for evidence-to-rule generalization.

Status: applied to the six semantic notes and the linked agent-memory-README.md summary. The distillation tag on an-accepted-edit-verifies-the-change-not-the-rule.md and old vocabulary in linked filenames remain deferred to the META/infrastructure wave.

Note Current usage Replacement direction
trace-extracted-memory-earns-authority-per-operation-not-at-capture.md Distill rung generalizes a verified fact into a rule Rename the rung to abstraction / discovery-stage generalization; keep the per-operation authority theory
an-outcome-check-licenses-replay-a-rule-needs-the-process-verified.md "distilling a rule" transfers a mechanism to new contexts Reword as abstracting/generalizing a rule; the control regime is process evidence plus boundary/testing, not derivation
an-accepted-edit-verifies-the-change-not-the-rule.md accepted edit verifies instance, rule extraction sits at the distill rung Reword as rule extraction / conjecture; likely migrate the distillation tag to discovery after tag infrastructure is ready
abstract-an-experience-only-when-you-can-state-the-boundary.md already argues about abstraction but cites distillation as the trace-to-rule mechanism Replace distill language with abstraction / discovery lifecycle; preserve the boundary-statability oracle
spec-mining-as-codification.md system "distills stochastic regularities into deterministic code" Reword to mine/extract discovered regularities, then codify them
research/adaptation-agentic-ai-analysis.md Distill when signal carries reusable judgment but not exact procedure Split: reusable judgment from traces is discovery/abstraction; exact procedure can later be derived or codified

Small linked surface to update with this cluster:

  • agent-memory-README.md - its "verification, distillation, and consultation" summary should follow the renamed trace ladder.

Residues found while checking the applied wave (2026-07-17): sweeps grep distill and miss the single-l spelling — a-citation-cannot-assert-more-fidelity-than-its-capture-preserved.md had "verify, distil, consult" (fixed with the ladder rename); the surviving case is integrated as a straggler row in the second-wave scope below. Future sweeps use the single-l distil pattern.

Second semantic wave: obvious derivation / selection

These are not the mistake. They are mostly use-shaped rewrites of existing reasoning, procedures, or traces for a bounded consumer. Under the no-successor-term direction there are no derivation/selection definitions to wait for: they wait on Wave 0's structure note and label grammar instead, and then reword into plain language, citing the structure note (or carrying a Derived from:-style label) only where the strong recomputable-from-source semantics are actually claimed. The table's "Replacement direction" column predates that decision — read "reframe as derivation/selection" as "reword in plain English, with a citation or label where the cache/fallback semantics matter."

Note Current usage Replacement direction
skills-derive-from-methodology-through-distillation.md methodology -> skill in same medium Retitle around derivation; preserve source-retention, residue, and edge-case fallback as derivation/cache semantics
skills-are-instructions-plus-routing-and-execution-policy.md skills/instructions are distilled procedures Split stable-core discovery from derived procedure body where needed; most references become derivation/proceduralization
distillation-status-determines-directory-placement.md procedures distilled for execution belong in kb/instructions/ Reframe directory boundary as execution-shaped procedural derivation
maintenance-operations-catalogue-should-stage-distillation-into.md stable manual operations are distilled into instructions Reframe as staging until stable, then deriving an execution procedure
ad-hoc-prompts-extend-the-system-without-schema-changes.md repeated prompts become skills through distillation Reframe as discovery of a stable procedure, then derivation of the reusable skill
capability-placement-should-follow-autonomy-readiness.md stable procedures are distilled into instructions/skills Reframe as promotion plus derivation into the appropriate execution surface
agents-md-should-be-organized-as-a-control-plane.md stable procedures are distilled into instructions/skills Same as capability placement: promotion plus derivation
scenario-decomposition-drives-architecture.md operational instructions distilled from deeper methodology Reframe as derived procedures with explicit fallback to methodology
session-history-should-not-be-the-default-next-context.md execution-boundary compression targeted to next stage Reframe as selection plus derived handoff artifact; preserve "store more than you load"
conversation-vs-prompt-refinement-in-agent-to-agent-coordination.md prompt refinement is distillation of caller knowledge Reframe as deriving a focused handoff prompt from caller context and target task
bounded-context-orchestration-model.md compaction of scheduler state Reframe as selection plus derivation for the orchestrator's context budget
decomposition-heuristics-for-bounded-context-scheduling.md saved intermediates are often distillations Reframe as derived intermediate artifacts shaped for later reuse
llm-mediated-schedulers-are-a-degraded-variant-of-the-clean-model.md compaction applied to scheduler conversation state Reframe as derived compaction / handoff artifact
feasibility-is-the-heaviest-forks-net-load.md parent absorbs work and hands down a leaner task Reframe as frontloaded selection/derivation for sub-agent feasibility
pointer-design-tradeoffs-in-progressive-disclosure.md pointers and abstracts are distillations Reframe as derived pointers / precomputed selection aids
charting-the-knowledge-access-problem-beyond-rag.md TOCs, abstracts, overviews as navigation distillates Reframe as derived navigation views for read/skip decisions; keep synthesis cases separate
epiplexity-by-example-what-entropy-and-complexity-miss.md distillation raises extractable structure for a bounded observer Reframe as observer-shaped derivation/selection unless the note needs the external ML term
reverse-compression-is-when-llm-output-expands-without-adding.md distillation as productive compression preserving structure Reframe as derivation or structure-preserving compression
../reference/where-change-candidates-come-from-in-commonplace.md cp-skill-ingest folds connect findings into a durable ingest report Reframe as selecting/folding discovered candidates into the report; likely derivation where the report preserves the findings
../reference/design-rationale-management.md a design constraint can motivate extracting a use-shaped artifact from larger material Reframe as selection plus derivation, unless the artifact adds a new design commitment
../reference/proposals/factored-dependency-pairs-for-review-freshness.md source-as-gate checks consistency with the source snapshot a note distills Reframe as source-derived consistency / derivation-gate language

META and infrastructure: defer

These are obvious, but they belong to the infrastructure wave, not the first semantic wave:

  • distillation-README.md, learning-theory-README.md, tags-README.md, README.md, tag fields, and covered_by fields.
  • definitions/distillation.md and cross-definition references in definitions/constraining.md, definitions/codification.md, definitions/context-engineering.md, definitions/directed-reading.md, definitions/text-contract.md.
  • Distilled into: footer labels and backlink grammar.
  • Filenames containing distill*.
  • Link-only references where the target note is itself being renamed.
  • Reference infrastructure and history: ../reference/link-vocabulary.md, ADR 011's inline-gloss decision, ADR 005's historical "notes distill from which" comment, ADR 020's catalogue-sync wording, ADR 021's path-audit examples and old filename discussion, and ../reference/tag-readme-trace-observed-causal-connection.md's historical mention of distillation-README.md.

These should move after the semantic notes settle, otherwise generated navigation and backlinks will hide unresolved vocabulary decisions.

Deliberately not first-wave

These notes have distill* uses that are semantically important or mixed enough to need the promoted receiving vocabulary before editing:

  • distillation-is-transformation-not-selection.md - pivotal mixed note; its anti-selection argument survives, but its trace-to-rule evidence routes to discovery.
  • constraining-and-distillation-both-trade-generality-for-reliability.md - foundational tradeoff note; likely needs a rewrite around derivation/selection rather than local substitutions.
  • distilled-artifacts-need-source-tracking.md - lineage argument still matters, but "distilled artifact" is broader than the new derivation term unless the scope is restated.
  • a-derived-copy-of-recomputable-truth-must-be-checked-or-absent.md - already partially migrated; should be aligned only after the source-tracking note is rewritten.
  • evolving-understanding-needs-re-distillation-not-composition.md - holistic rewrite can be derivation, synthesis, or authored judgment depending on what the narrative adds.
  • memory-management-policy-is-learnable-but-oracle-dependent.md - explicitly contrasts selection and distillation; must be reinterpreted under the new split, not patched.
  • minimum-viable-vocabulary-is-the-naming-set-that-most-reduces.md and information-value-is-observer-relative.md - information-theoretic framing may need selection/derivation/discovery all at once.
  • legal-drafting-solves-the-same-problem-as-context-engineering.md and soft-bound-traditions-as-sources-for-context-engineering-strategies.md - external analogies; preserve loose analogical force until the internal terms settle.
  • task-fitted-structure-costs-cross-task-reuse.md - uses distillation as a contrast case; only safe after the tradeoff note is rewritten.
  • semantic-review-catches-content-errors-that-structural-validation.md - discusses the old three-operation framework as a discovered error; may need historical wording rather than replacement.
  • ../reference/proposals/automated-note-refinement-as-search-over-source-bundle.md - explicitly asks whether automated refinement subsumes re-distillation; hold until the evolving-understanding note and source-bundle proposal can share the new vocabulary.
  • skills-derive-from-methodology-through-distillation.md - skipped in the second semantic wave: its methodology case claims source fallback, but its judgment and raw-domain-artifact cases add substantive choices and generalizations that need an entangled rewrite.
  • session-history-should-not-be-the-default-next-context.md - skipped in the second semantic wave: the note explicitly leaves unresolved whether Slate-style compressed episodes are narrow handoffs or synthesis-bearing trace artifacts.
  • charting-the-knowledge-access-problem-beyond-rag.md - skipped in the second semantic wave: its navigation pointers are precomputed views, but the same vocabulary also covers contradiction-resolving synthesized narratives that can add substantive claims.
  • epiplexity-by-example-what-entropy-and-complexity-miss.md - skipped in the second semantic wave: its value claim combines rearrangement, compression, and observer-relative accessibility without fixing which transformation controls the resulting artifact.
  • reverse-compression-is-when-llm-output-expands-without-adding.md - skipped in the second semantic wave: its productive-inverse contrast inherits the old term's broad scope and does not choose between selection, source-derived reshaping, and structure-preserving compression.
  • areas-exist-because-useful-operations-require-reading-notes-together.md - skipped in the fourth batch: current COLLECTION.md no longer contains the area assignment, split-threshold, or areas: rules named by the legacy footer.
  • discovery-is-seeing-the-particular-as-an-instance-of-the-general.md - skipped in the fourth batch: current cp-skill-connect no longer contains the Phase 5 abstraction-opportunity logging attributed by the legacy footer.

Execution order (current)

  1. ~~Apply the first semantic wave: the obvious discovery / abstraction cluster.~~ Done — commit 80f47b41; residues integrated as stragglers below.
  2. ~~Promote the Wave 0 receiving surfaces.~~ Done — structure note 7e9a63c8, label semantics 0dc7fb4c, discovery lifecycle 7ac83249; Wave 0 is complete.
  3. Apply the second semantic wave (plan Wave 2's obvious library subset) under the execution contract below. Decided: this runs independently of plan Wave 1 — it depended only on Wave 0.
  4. The deliberately-not-first-wave notes are plan Wave 1: entangled, review-gated, not batch work — out of scope for this file's executor.
  5. META, tags, filenames, and Distilled into: footers are plan Wave 3 / Wave 1 classification — untouched here.

Execution contract for the second wave

Authority: migration-plan.md is canonical; this file is the staging inventory for its Wave 2 obvious library subset. On any conflict, the plan wins.

Scope: exactly the rows of the second-wave table plus the stragglers list. Do not touch the deliberately-not-first-wave notes, tag fields, filenames, Distilled into: footers, or any file in the META list.

Rewording rule: plain English — there is no successor technical term. Where a passage genuinely claims the strong semantics (the artifact is recomputable from a retained source; stale when the source changes), cite kb/notes/theory-and-methodology-form-a-two-layer-execution-system.md or use the derived-from / Derived into: labels per kb/reference/link-vocabulary.md. Where the passage just describes use-shaping, ordinary words (reshape, work out, compile, condense, extract) suffice. "Derive/derived" in ordinary English is fine and claims nothing.

Judgment latitude: rows hedged with "likely / unless / where needed" get the classifying question from the link grammar — could another agent reconstruct the passage's substantive claims from the stated source? If a row turns out mixed or non-obvious in situ, skip it and log it under the deliberately-not-first-wave list with one line of reasoning; never force a classification to finish a batch.

Procedure: work in batches of a few notes; run commonplace-validate notes (plus reference for the two reference-collection rows) after each batch; commit each batch separately, staging explicit paths (never git add -A). No renames in this wave.

Completion sweep: after the last batch, rg -in 'distil' <touched files> must show only link paths to not-yet-renamed files (behind neutral anchor text). Sweep with the single-l pattern — distill misses British spellings, which is how the stragglers below survived the first wave.

Stragglers (found post-first-wave; execute with the second wave)

Note Current usage Replacement direction
claw-learning-loops-must-improve-action-capacity-not-just-retrieval.md:30 AWM "distils successful traces into workflow prose" (single-l) Plain rewording of the external system's operation ("turns successful traces into workflow prose" / "extracts workflows from successful traces") — description of an ampliative loop, no strong claim to carry

Third batch (reclassified from Wave 1, 2026-07-17; same execution contract)

Note Current usage Replacement direction
minimum-viable-vocabulary-is-the-naming-set-that-most-reduces.md use-shaping talk (ln ~20, 22, 39); distillation tag stays for Wave 3 Plain rewording: domain knowledge reshaped for a newcomer under a minimal context budget
information-value-is-observer-relative.md use-shaping talk (desc + ln ~37, 38, 42, 60) Plain rewording: reshaping for a specific observer; description becomes "grounding use-shaped reshaping and discovery"
legal-drafting-solves-the-same-problem-as-context-engineering.md loose external analogy (ln ~47, 78) Plain rewording: task-targeted compression; preserve the analogical looseness

Convert the footer label only where the classification below is high/medium confidence; adjust the context phrase to match the label's semantics. Skip-and-log anything that reads mixed in situ. Do not touch the needs-read rows (they stay with the Wave 1 worksheet): unified-calling-conventions, rlm-has-the-model-write…, session-history-should-not….

Source note Footer target New label
areas-exist-because… COLLECTION.md rules Derived into: (quick confirm in situ)
definitions/reach-assessment formal-systems split-out Derived into:
first-principles-reasoning… review-explanatory-reach task Derived into:
title-as-claim-exposes… COLLECTION.md checklist Derived into:
discovery-is-seeing… cp-skill-connect Derived into: (medium — confirm the skill content is worked out from the note)