llm-context-base epistemic report

Type: agentic-system-analyses/types/agentic-system-epistemic-report.md

Source-and-claim boundary

The question is how this template acquires, checks and licenses retained content for later answers and behavior. Scope and exclusions are see boundary; sources are see Source register. The assessment concerns shipped prompts, templates, configuration and the metadata converter, not the external assistant's inference or tool implementation. All new inspections below concern SRC-1 at the frozen revision. Prompt routes have architectural status doctrine only: executable host delivery and observed adherence are not supplied. The converter alone supplies an inspected executable transformation. No target execution was performed.

Knowledge and warrant claims are EPI-CLM-decision-learning, EPI-CLM-standard-enforcement, EPI-CLM-knowledge-memory, and the supplied RT-CLM-review-policy. Context-efficiency claims and host-permission limits are see RT-RTE-context-review and RT-CLM-host-permissions.

Entry point or operation	Source path, SRC-1	Coverage and provenance	Limit and conclusion prevented
Generic bootstrap and startup	AGENTS.md, _config/config.md, _meta/instructions/general.md	Direct reads; see RT-RTE-session-start	Framework-mode bypass; no observed delivery/adherence
Other five bootstraps and tool discovery	.claude/CLAUDE.md, .cursor/rules/bootstrap.mdc, .cursorrules, .github/copilot-instructions.md, .windsurfrules	Supplied runtime and boundary findings, not new direct reads; see RT-RTE-session-start	External host discovery not inspected; no cross-host reliability conclusion
Capture, web clip triage and source preservation	_meta/instructions/agent-write.md, _meta/instructions/general.md, assets/clipper-template.json, _sources/README.md	Direct reads; see RT-RTE-capture-file	Clipper implementation external; no semantic fidelity or factual endorsement established
Query, missing-document response and decision precedent	_meta/instructions/knowledge-query.md, _meta/templates/decision.md	Direct reads; see RT-RTE-knowledge-query	No user corpus or outcome trace; retrieval quality and improved decisions unobserved
Standards, completion and links	_config/standard.md, _meta/instructions/definition-of-done.md, _meta/templates/knowledge-article.md	Direct reads; EPI-RTE-completion-check, EPI-RTE-completion-status, EPI-RTE-completion-links	Accuracy is a requested criterion without specified evidence procedure
Lint evaluation and cleanup/withdrawal	_meta/instructions/knowledge-lint.md, _config/standard.md	Direct reads; EPI-RTE-lint-check, EPI-RTE-lint-action	Seven prescribed structural/age checks, not factual verification
Immediate rule/preference capture	_meta/instructions/general.md, _config/context.md, _config/config.md	Direct reads; EPI-RTE-pattern-capture, see RT-BAP-instruction-load	No required per-rule evidence acceptance or observed later activation
Optimization, optional hooks, permissions	_meta/instructions/optimization-review.md, docs/claude-code-setup.md, .claude/settings.json	Direct reads; see RT-RTE-context-review, RT-CLM-host-permissions	Hook recipes are not installed scripts; logs of paths/line counts do not license source truth
Blog review, publication and tracking	_meta/instructions/blog-posts/instructions.md	Direct read; EPI-RTE-blog-review, EPI-RTE-blog-publish	Newly elaborated runtime omission; platform implementation and factual review criteria absent
Standalone Python dispatch and rewrite	migrate-to-yaml-frontmatter.py	Complete direct read, static trace; EPI-RTE-metadata-transform, see RT-RTE-metadata-migrate	No run evidence, parser, backup or semantic comparator
Project indexing, local capture and exports	1-Projects/README.md, .obsidian/app.json, _output/README.md, docs/linked-code-repos.md	General-module direct read plus supplied boundary/runtime coverage; see RT-RTE-capture-file, RT-RTE-knowledge-query	Renderers, linked repositories and plugins external; no independent evaluator inferred

The tracked tree was enumerated and has no additional shipped CLI driver, registered-tool implementation or hook script. This is not an absence claim about external intelligence layers. Available supporting documentation excluded by the boundary is not represented as inaccessible evidence. No additional material operation was identified outside the frozen functional boundary. Coverage of indirectly read paths depends on the supplied findings, not a fresh complete-source audit.

Missing evidence is local: original-source truth and extraction fidelity prevent a warrant-preservation verdict; evidence-consuming accuracy decisions prevent identification of accepted newly produced claims; user corpus and authentic outcome records prevent retrieval/decision effectiveness conclusions; host/model identity and parameter behavior prevent claims about parametric learning; actual update and later-consumer traces prevent exercised adaptation or self-improvement attribution; intervention comparisons prevent causal benefit attribution. The template's sample or placeholder artifacts do not supply these traces.

Epistemic-object inventory

Object ID	Candidate truth-apt content or none	Claimed role	Evidence source ID and local anchor	Gap/limit
EPI-OBJ-knowledge-article	Processes, references, extracted points and saved synthesis; particular propositions not determinable in an empty user corpus	see record	SRC-1, _meta/templates/knowledge-article.md	Draft/active labels do not establish factual warrant
EPI-OBJ-sources	Imported original content; truth and original provenance unknown	see record	SRC-1, _sources/README.md	Optional preservation, not endorsement
EPI-OBJ-decision-outcome	Reports of what happened and discovered constraints	see record	SRC-1, _meta/templates/decision.md	Pending/example fields are not observed evidence
EPI-OBJ-context	Personal details and reports of preferences; preferences also prescribe behavior	see record	SRC-1, _config/context.md	User assertions versus inferred patterns not uniformly distinguished
RT-OBJ-instruction-modules	Mainly non-truth-apt policy; guidance also asserts context-efficiency mechanisms	see record	SRC-1, _meta/instructions/optimization-review.md	Policy obedience and mechanism truth are different questions
EPI-OBJ-lint-report	Structural, date and count claims about files	see record	SRC-1, _meta/instructions/knowledge-lint.md	Model-followed recipe, no observed checking
EPI-OBJ-blog-post	Possible factual prose and real examples; proposition content not determinable before a user draft exists	see record	SRC-1, _meta/instructions/blog-posts/instructions.md	Presentation review is not source verification
EPI-OBJ-metadata	Summary/date/source fields may describe the file; tags/status also direct selection	see record	SRC-1, migrate-to-yaml-frontmatter.py	Symbolic consumer behavior can differ from surface text preservation
RT-CMP-ai-assistant	Distributed-parametric content opaque	see record	SRC-1, _config/config.md	Parameters, exact version and endpoint mutability uninspected; no parametric epistemic verdict

Authority-route ledger

Fields owned by the declaring records say see; the ledger supplies architectural status and content-edge assessment rather than recopying canonical findings. Repeated citations split consequential functions, not identities. None of the doctrine rows is an observed route.

route ID	route function	architectural status	object/candidate ID	content/update relation	transition or check target	evaluator/condition and domain	activation and timing	possible or observed result	implemented force	epistemic authority and scope	operational authority: behavior permitted, blocked, or changed	behavioral-authority path: consumer, channel, force, horizon	evidence source ID and local anchor	claim IDs or none	mismatch marker or none	gap/limit
RT-RTE-capture-file	content transformation	doctrine only	EPI-OBJ-sources → EPI-OBJ-knowledge-article	truth-apt transformation: acquisition/import	External clips/user assertions enter captures	see RT-RTE-capture-file	see RT-RTE-capture-file	Prescribed draft/extracted working copy	see RT-RTE-capture-file	Source warrant unknown; attribution is not verification	see RT-RTE-capture-file	see RT-RTE-capture-file	SRC-1, _meta/instructions/agent-write.md	RT-CLM-review-policy	none	External source truth, semantic extraction fidelity and execution unknown
RT-RTE-capture-file	retention	doctrine only	EPI-OBJ-sources and EPI-OBJ-knowledge-article	no content change	Original/working-copy persistence	see RT-RTE-capture-file	see RT-RTE-capture-file	Prescribed filing; optional original retention	see RT-RTE-capture-file	No added warrant from disk or Git	see RT-RTE-capture-file	see RT-RTE-capture-file	SRC-1, _sources/README.md	RT-CLM-review-policy	none	Preservation defaults off; full source may be discarded
RT-RTE-knowledge-query	content transformation	doctrine only	EPI-OBJ-knowledge-article	truth-apt transformation: indeterminate	Selected files → synthesized answer	see RT-RTE-knowledge-query	see RT-RTE-knowledge-query	Prescribed cited synthesis, or nothing-found response	see RT-RTE-knowledge-query	May reshape or derive; conjecture requires evidence of non-entailment	see RT-RTE-knowledge-query	see RT-RTE-knowledge-query	SRC-1, _meta/instructions/knowledge-query.md	EPI-CLM-knowledge-memory	none	No input/output pair to distinguish reshaping, entailment or ampliation
RT-RTE-knowledge-query	operational admission/selection/consumption	doctrine only	EPI-OBJ-decision-outcome	no content change	Relevant past outcomes → new decision deliberation	see RT-RTE-knowledge-query	see RT-RTE-knowledge-query	Prescribed precedents and failure constraints	see RT-RTE-knowledge-query	Outcome report licenses bounded precedent consideration, not causal explanation/transfer	see RT-RTE-knowledge-query	see RT-RTE-knowledge-query	SRC-1, _meta/instructions/knowledge-query.md	EPI-CLM-decision-learning	none	No observed dependence or improved decision; no universal superseded-content ban
EPI-RTE-completion-check	check/evidence production	doctrine only	EPI-OBJ-knowledge-article	no content change	Knowledge accuracy/completeness and type-specific completion criteria	Assistant/user checklist; accuracy evidence procedure unspecified	see EPI-RTE-completion-check	Checklist judgment requested	see EPI-RTE-completion-check	Criterion named, but evidence-consuming factual acceptance not established	Supports subsequent status proposal	Assistant → checklist judgment → completion; advisory, current task	SRC-1, _meta/instructions/definition-of-done.md	EPI-CLM-knowledge-memory	none	Accuracy criterion not an implemented verifier
EPI-RTE-completion-status	disposition/acceptance	doctrine only	EPI-OBJ-knowledge-article	non-truth-apt policy/content update: lifecycle label	All checklist items → active/complete	Assistant with user nudges; useful/findable intended use	see EPI-RTE-completion-status	Requested status and filing change	see EPI-RTE-completion-status	Operational completion only; no observed accepted ampliative claim	File becomes designated in-use/complete	Future assistant → file status → reliance/selection; advisory, later sessions	SRC-1, _meta/instructions/definition-of-done.md	EPI-CLM-knowledge-memory	none	Active is not a truth guarantee
EPI-RTE-completion-links	lifecycle integration	doctrine only	EPI-OBJ-knowledge-article	non-truth-apt policy/content update: links and organization	Completed item → related links and proper directory	Assistant; relevance and checklist condition	see EPI-RTE-completion-links	Requested connected/filed artifact	see EPI-RTE-completion-links	Conditional integration after prescribed completion, not observed epistemic acceptance	Makes completed item navigable	Later reader → links/directory → discovery; advisory, later queries	SRC-1, _meta/instructions/definition-of-done.md	EPI-CLM-knowledge-memory	none	Links are also a completion prerequisite; graph connection can precede acceptance
EPI-RTE-lint-check	check/evidence production	doctrine only	EPI-OBJ-lint-report	truth-apt transformation: entailed derivation	Dates, type-field presence, inbound links, sizes, names and link targets	Assistant following specified searches/counts; filesystem structural domain	see EPI-RTE-lint-check	Requested seven-check report	see EPI-RTE-lint-check	If correctly interpreted, findings follow file measurements; no source-truth license	Reports feed recommendations and optimization trigger	Assistant/user → report → advisory action; current maintenance session	SRC-1, _meta/instructions/knowledge-lint.md	EPI-CLM-standard-enforcement	none	No execution; inconsistent metadata coverage; not a complete schema check
EPI-RTE-lint-action	disposition/acceptance	doctrine only	EPI-OBJ-knowledge-article	non-truth-apt policy/content update: cleanup and retirement	Flagged records → review/update/archive/link repair	Assistant recommends; user/manual resolution on broken links	see EPI-RTE-lint-action	Requested triage/discard/archive or in-place format fixes	see EPI-RTE-lint-action	Structural fitness/freshness disposition, not falsification of factual claims	Retirement affects relevance and freezes stale-check exemption	Assistant → status/links/report → advisory selection; later queries	SRC-1, _meta/instructions/knowledge-lint.md	EPI-CLM-standard-enforcement	none	No universal enforced withdrawal or rollback cascade
EPI-RTE-blog-review	check/evidence production	doctrine only	EPI-OBJ-blog-post	no content change	Draft → visual, screenshot, keyword and social-post checklist	Self-review or peer reviewer; presentation readiness domain	see EPI-RTE-blog-review	Requested readiness judgment and pushback on absent visuals	see EPI-RTE-blog-review	Licenses presentation readiness, not factual truth of examples	Supports publishing continuation	Author/reviewer → checklist → advisory; current draft	SRC-1, _meta/instructions/blog-posts/instructions.md	none	none	No factual correctness or evidence acceptance criterion specified
EPI-RTE-blog-publish	operational admission/selection/consumption	doctrine only	EPI-OBJ-blog-post	non-truth-apt policy/content update: published label and URL	Reviewed draft → primary publication/cross-post tracking	External author/assistant and platform; deployment uninspected	see EPI-RTE-blog-publish	Requested posted content, status published and links	see EPI-RTE-blog-publish	Publication licenses no additional epistemic warrant	Makes content publicly consumable	Author/platform → publish and cross-post channels → requested action; durable audience exposure	SRC-1, _meta/instructions/blog-posts/instructions.md	none	none	Review ownership does not require independent factual evaluator
EPI-RTE-pattern-capture	behavior/policy adaptation	doctrine only	EPI-OBJ-context	non-truth-apt policy/content update: operating convention/preference	Conversation rule/preference → context or instruction edit	Assistant identifies established patterns; ask if ambiguous; user can override	see EPI-RTE-pattern-capture	Requested immediate update and later reload	see EPI-RTE-pattern-capture	User preference may license personalization; inferred patterns lack mandatory acceptance test	Changes requested future behavior without ordinary draft gate	see RT-BAP-instruction-load	SRC-1, _meta/instructions/general.md	EPI-CLM-knowledge-memory	none	Factual personal-detail edges separately acquired; no observed activation
RT-RTE-context-review	behavior/policy adaptation	doctrine only	RT-OBJ-instruction-modules	non-truth-apt policy/content update: instruction restructuring	Measured organization → user-approved leaner instructions	see RT-RTE-context-review	see RT-RTE-context-review	Prescribed proposal, possible rejection or approved edits	see RT-RTE-context-review	Size/repeated reads warrant review, not demonstrated speed or reliability	see RT-RTE-context-review	see RT-BAP-instruction-load	SRC-1, _meta/instructions/optimization-review.md	EPI-CLM-knowledge-memory	none	Successful learning/reflection/self-improvement remain independently unestablished
EPI-RTE-metadata-transform	content transformation	implemented	EPI-OBJ-metadata	truth-apt transformation: indeterminate	Legacy bullet metadata and surrounding text → frontmatter rewrite	Fixed regex and formatter; matching Markdown domain	see EPI-RTE-metadata-transform	Statically deduced rewrite; can lose pre-title content/duplicate metadata	see EPI-RTE-metadata-transform	No validation or preservation guarantee; bounded witness below	Direct overwrite supplies changed text to future file consumers	Operator → converter → file → later reader/parser; enforcing rewrite, durable	SRC-1, migrate-to-yaml-frontmatter.py	EPI-CLM-standard-enforcement	none	Intended reshaping, but loss and YAML interpretation defeat general semantic preservation

System-claim versus route comparison

Claim ID	Claimed operation or warrant	Claim source ID/anchor and evidence layer	Doctrine/design support	Implemented route IDs	Observed-run support	Causal support and design limits	Supported conclusion	Mismatch/unknown
RT-CLM-review-policy	see record	SRC-1, _meta/instructions/agent-write.md; doctrine/design	see record	EPI-RTE-metadata-transform is an alternate direct rewrite, not a review gate	None supplied	see record	Inbox/draft enables review opportunity	Immediate pattern edits, explicit destinations and migration defeat universal per-content approval
EPI-CLM-standard-enforcement	see record	SRC-1, _config/standard.md; doctrine/design	Required fields plus lint recipe	EPI-RTE-metadata-transform converts text, does not validate	None supplied	No compliance experiment; lint tests only type-field presence in limited roots	A requested metadata convention and partial check	“Always enforced” and missing-field coverage exceed specified check
EPI-CLM-decision-learning	see record	SRC-1, _meta/instructions/knowledge-query.md; doctrine/design	Outcome template and proactive precedent query	None in target; see RT-RTE-knowledge-query	None supplied	Outcomes alone do not establish explanation, transfer or improvement	Prescribed future use of operational experience	Improved capacity, causal attribution and actual decision dependence unknown
EPI-CLM-knowledge-memory	see record	SRC-1, PHILOSOPHY.md; doctrine/design	Capture, completion, querying and evolution guidance	Converter only supplies bounded file maintenance	None supplied	No authentic update/consumer trace or causal comparison	Durable text supports a prescribed institutional-memory and adaptation path	Accuracy, complete retention, instant retrieval and successful improvement not demonstrated

Bounded conclusion

Acquisition and retention: see RT-RTE-capture-file, EPI-OBJ-sources and EPI-OBJ-knowledge-article. External statements enter drafts and extracted working copies. Source attribution records origin, not truth. Originals are selectively preserved only when enabled; the default can discard originals. Warrant preservation is unknown, while loss of auditability is a supported design risk. Later query selection can omit most files and the later sections of large files. Freshness warnings and citations do not restore omitted evidence or endorse claims.

Production and checks: see RT-RTE-knowledge-query, EPI-RTE-completion-check, EPI-RTE-lint-check and EPI-RTE-blog-review. Synthesis/idea reactions are candidate production, but no concrete input/output pair establishes a particular reshaping, entailed claim or non-entailed conjecture. Lint predicates can entail structural claims from correctly measured file premises; those premises and interpretations have not been observed. Blog review checks presentation. None supplies a general factual oracle. The accuracy item in completion is a counterexample to “no accuracy criterion exists,” but not evidence that such a criterion was applied against evidence.

Disposition and integration: see EPI-RTE-completion-status, EPI-RTE-completion-links, EPI-RTE-lint-action and EPI-RTE-blog-publish. Useful/findable completion is a named intended use with type-specific criteria. Prescribed checklist satisfaction permits active/complete status and final filing; related links support discovery, sometimes before status change. No supplied evidence individuates an evidence-consuming acceptance of a newly produced ampliative claim. Archive/superseded labels request changed reliance and require a reason, but query explicitly deprioritizes archived records rather than enforcing universal rejection of retired claims. Publication is an operational consequence, not epistemic acceptance. There is no supported downstream retraction guarantee.

Behavioral use: see EPI-RTE-pattern-capture, RT-BAP-instruction-load and RT-RTE-context-review. Retained preferences, conventions and instructions request future assistant behavior through startup and task-triggered file reads. This is a behavior-shaping path even without producing a truth-apt proposition. Immediate pattern capture differs from user-approved optimization restructuring. Human override/approval is route-specific; initiating a session or capture is not approval of every sentence. No observed update-to-consumer dependence or successful improvement is established.

Independent properties: the optimization record supplies separately bounded theory-builder conditions and improvement objective. Whole textual guidance is addressable and persistent by design; consumption is prescribed; content-directed criticism and criticism-fed iteration remain unobserved. The decision-outcome route separately prescribes use of experience, without proving improved capacity or causal mechanism. Host model mutability is see RT-CMP-ai-assistant, not inferred from file adaptation. These observations neither confer nor deny a single system-wide learning, reflection, autonomy or self-improvement status.

Executable exception: see EPI-RTE-metadata-transform. Static code supports direct metadata rewrite and a concrete loss witness, not factual validation or a successful runtime migration. The converter's printed success labels cannot establish YAML fidelity, source truth or replay safety. No observed run or causal experiment closes the uncertainties above.

Shared records

Operative objects

EPI-OBJ-knowledge-article

Label: Knowledge article

Source-native identity: retained knowledge documents shaped by _meta/templates/knowledge-article.md, including working versions of captures/syntheses. Natural-language body and summary on Markdown/filesystem, with symbolic metadata selectors separately assessed in EPI-OBJ-metadata. These are outputs/read-back inputs of RT-RTE-capture-file and RT-RTE-knowledge-query, not duplicates of RT-OBJ-instruction-modules. SRC-1, doctrine/design, _meta/templates/knowledge-article.md:

[Content here] --- _meta/templates/knowledge-article.md

The placeholder gives a content shape, not a completed proposition or its production provenance. Individual truth-apt claims in user documents are not determinable here.

EPI-OBJ-sources

Label: Preserved sources

Source-native identity: original documents moved to _sources/. Natural-language originals (or externally encoded PDFs) reside in files and are Git-versionable; opaque binary source content is not supplied. Distinct from working articles; no supplied operative-object counterpart after comparison with RT-OBJ-instruction-modules. Closest route is RT-RTE-capture-file. SRC-1, doctrine/design, _sources/README.md:

Sources here are immutable - they're reference copies, not working documents --- _sources/README.md

This is requested immutability, not a filesystem invariant. Originating-source truth and full reprocessing fidelity remain unknown. The preservation default is evidenced in _config/config.md:

• Preserve Important Sources: no --- _config/config.md

EPI-OBJ-decision-outcome

Label: Decision outcome

Source-native identity: the Outcome part of decision Markdown, natural-language reports and constraints with status/date fields, stored on filesystem. Distinct from operating instructions and generic knowledge articles; closest supplied route RT-RTE-knowledge-query consumes it but is not the object. SRC-1, doctrine/design, _meta/templates/decision.md:

--- _meta/templates/decision.md

[Any constraints or failure modes you didn't anticipate that should inform future decisions?] --- _meta/templates/decision.md

No specified independent measurement, expected-answer oracle or causal comparison validates the reported outcome or its proposed lesson. Pending fields and examples do not establish any completed evaluation route.

EPI-OBJ-context

Label: Personal context

Source-native identity: _config/context.md, profile and learned conventions in _config/config.md. Natural-language facts/preferences stored in Markdown on filesystem; dates/phases have selector roles. These accumulated personal facts differ from the operating modules in RT-OBJ-instruction-modules; closest supplied route RT-RTE-session-start reloads them. SRC-1, doctrine/design, _config/context.md:

This file is populated by your AI during training. As it learns about you, it logs personal details here so future sessions can reference them without asking again. You can also add things manually at any time. --- _config/context.md

Assertion provenance is not uniformly separated from inferred preference. Facts are truth-apt over the person/time named; instructions such as response preferences have non-truth-apt policy force.

EPI-OBJ-lint-report

Label: Lint report

Source-native identity: the report prescribed in knowledge lint and its logged issue counts. Natural-language findings plus symbolic counts/dates, returned in chat and summarized in WIKI-LOG.md; no actual report supplied. Unlike RT-OBJ-instruction-modules, this is a requested evaluation product, not evaluator guidance. SRC-1, doctrine/design, _meta/instructions/knowledge-lint.md:

After running all checks, compile a summary: --- _meta/instructions/knowledge-lint.md

EPI-OBJ-blog-post

Label: Blog post

Source-native identity: template-shaped blog draft, visual plan, full draft and social-post adaptation. Natural-language prose with code/data/screenshots, stored in Markdown and potentially published externally. No user instance or image semantics available. The closest supplied RT-RTE-capture-file covers general creation but not the public publishing consumer; no supplied object counterpart. SRC-1, doctrine/design, _meta/instructions/blog-posts/instructions.md:

• Code / data blocks — real examples, not hypothetical. Readers skim for code. --- _meta/instructions/blog-posts/instructions.md

EPI-OBJ-metadata

Label: File metadata

Source-native identity: legacy bullet fields and YAML frontmatter in content files. Symbolic keys/status/tags/dates drive routing and format interpretation; summary/source strings carry natural-language descriptions. Filesystem storage, versionable when committed. Distinct from RT-OBJ-instruction-modules; closest supplied RT-RTE-metadata-migrate transforms it. SRC-1, implementation, migrate-to-yaml-frontmatter.py:

METADATA_LINE = re.compile(r'^- **(\w+):** (.*)$') --- migrate-to-yaml-frontmatter.py

Routes

EPI-RTE-completion-check

Label: Completion checklist evaluation

Part of: RT-RTE-capture-file

Closest supplied records: RT-RTE-capture-file. Distinct consequential function: completion checklist evaluation, not the whole supplied invocation/container. No same-function supplied route declaration.

Assistant/user consumes drafted content and type-specific checklist → requested completion judgment. Domain: completeness, accuracy and usability for knowledge; rationale/options/owner for decisions; shipped criteria for projects. Assistant proposes, user can decline; no independent oracle or specified accuracy-evidence protocol. Open completion request, not a controlled experiment. SRC-1, doctrine/design, _meta/instructions/definition-of-done.md.

• Immediate return: Checklist judgment / readiness nudge
• Later read-back: inapplicable — this check itself retains no product; the disposition route can write status
• Delegated visibility: inapplicable — no delegated worker is specified
• Selection predicate: Draft content and declared type
• Invalidation or expiry: inapplicable — no durable check result expiry is specified
• Activation or effect: Requested readiness assessment; actual checking unobserved
• Evidence limits: Static source only; no observed host adherence, evidence-consuming acceptance or causal benefit
• design conclusion status: claimed
• operation conclusion status: uninspected

Content is complete and accurate --- _meta/instructions/definition-of-done.md

EPI-RTE-completion-status

Label: Completion status disposition

Part of: RT-RTE-capture-file

Closest supplied records: RT-RTE-capture-file. Distinct consequential function: completion status disposition, not the whole supplied invocation/container. No same-function supplied route declaration.

Checklist judgment → active/complete status and final filing. Assistant writes through host tools; user can accept/reject nudges. Trigger is all checklist items done; admission requests useful/findable content for intended future use, not an independently recorded factual acceptance. No answer oracle. Open maintenance mode; user can edit or externally revert committed files, no automatic rollback. SRC-1, doctrine/design, _meta/instructions/definition-of-done.md.

• Immediate return: Status/filing response
• Later read-back: Changed status/files may reach later queries; no observed read-back
• Delegated visibility: inapplicable — no delegated worker is specified
• Selection predicate: All checklist items judged done
• Invalidation or expiry: Status may later be changed; no universal expiry
• Activation or effect: Requested in-use/complete designation
• Evidence limits: Static source only; no observed host adherence, evidence-consuming acceptance or causal benefit
• design conclusion status: claimed
• operation conclusion status: uninspected

When ALL checklist items are done: --- _meta/instructions/definition-of-done.md

EPI-RTE-completion-links

Label: Completion link integration

Part of: RT-RTE-capture-file

Closest supplied records: RT-RTE-capture-file. Distinct consequential function: completion link integration, not the whole supplied invocation/container. No same-function supplied route declaration.

Completed article → related links/proper directory, conditional on prescribed completion. Assistant selects relevant connections; user can decline filing or change links. Related links are themselves checklist items, so operational organization may precede epistemic acceptance. Intended future use is findability; no new factual oracle or autonomous acceptance engine. Open filing; edits/Git are external recovery. SRC-1, doctrine/design, _meta/instructions/definition-of-done.md.

• Immediate return: Connected/filed artifact
• Later read-back: Links/directories may guide later readers
• Delegated visibility: inapplicable — no delegated worker is specified
• Selection predicate: Connections exist and type-specific completion criteria
• Invalidation or expiry: Links can become broken/retired; lint requests repair
• Activation or effect: Requested navigation/integration, not observed acceptance
• Evidence limits: Static source only; no observed host adherence, evidence-consuming acceptance or causal benefit
• design conclusion status: claimed
• operation conclusion status: uninspected

Related links added if connections exist --- _meta/instructions/definition-of-done.md

EPI-RTE-lint-check

Label: Lint predicate evaluation

Closest supplied records: RT-RTE-context-review. Distinct consequential function: lint predicate evaluation, not the whole supplied invocation/container. No same-function supplied route declaration.

User lint request/cadence → assistant reads files/searches dates, metadata, links, sizes and names → seven-check report/count log. Assistant follows deterministic predicates through host tools but also interprets scope and output. User may dispute findings. Reference domain is actual file existence/text/date/line counts, not an expected factual answer. Open health check; no measured accuracy run. SRC-1, doctrine/design, _meta/instructions/knowledge-lint.md.

• Immediate return: Report with flags/counts
• Later read-back: Logged issue counts may be read from wiki log; no observed later effect
• Delegated visibility: inapplicable — no delegated worker is specified
• Selection predicate: Seven declared checks and root-specific scans
• Invalidation or expiry: Active age threshold 90 days; archived/superseded exempt; review marker 30 days
• Activation or effect: Requested structural evidence and recommendations
• Evidence limits: Static source only; no observed host adherence, evidence-consuming acceptance or causal benefit
• design conclusion status: claimed
• operation conclusion status: uninspected

How: Read each file's first 20 lines. If no type: YAML frontmatter field found, flag it. --- _meta/instructions/knowledge-lint.md

Under the declared predicates, a file with status: active, an accurately read update date 100 days before today, and no exemption entails a stale-active flag (>90). This is a doctrine-level conditional deduction, not execution or a freshness-to-truth inference. Metadata check can pass mere type: presence while summary/tags/status/date are missing.

EPI-RTE-lint-action

Label: Lint cleanup disposition

Closest supplied records: RT-RTE-context-review. Distinct consequential function: lint cleanup disposition, not the whole supplied invocation/container. No same-function supplied route declaration.

Lint flags → assistant recommendations or specified in-place path-format fixes; user/manual resolution for broken targets → edited/archived/linked records. Software measurements and assistant judgment propose actions; no independent factual oracle. Archive/superseded reasons follow standard. Review/update/archive can be rejected by user; no automatic restore/withdrawal cascade, committed Git edits potentially reversible. Open maintenance, triggered by flags. SRC-1, doctrine/design, _meta/instructions/knowledge-lint.md.

• Immediate return: Cleanup recommendation or requested edit
• Later read-back: Changed metadata/links reach later query selection; archived records generally deprioritized
• Delegated visibility: inapplicable — no delegated worker is specified
• Selection predicate: Flag category; manual resolution on broken links
• Invalidation or expiry: Terminal statuses exempt stale-active check, not removed from all retrieval
• Activation or effect: Requested cleanup and changed selection relevance
• Evidence limits: Static source only; no observed host adherence, evidence-consuming acceptance or causal benefit
• design conclusion status: claimed
• operation conclusion status: uninspected

→ Action: Convert format issues in-place; broken links need manual resolution --- _meta/instructions/knowledge-lint.md

SRC-1, _config/standard.md, supplies retirement reasons:

• superseded — replaced by a newer file or approach. Requires failure_reason: explaining what replaced it. --- _config/standard.md

The downstream selection limit is SRC-1, _meta/instructions/knowledge-query.md:

• Status filter: archived files can usually be deprioritized unless explicitly asked about history --- _meta/instructions/knowledge-query.md

EPI-RTE-blog-review

Label: Blog presentation review

Closest supplied records: RT-RTE-capture-file. Distinct consequential function: blog presentation review, not the whole supplied invocation/container. No same-function supplied route declaration.

Template draft → self-review or peer review → readiness judgment. Reviewer checks visuals, screenshot accessibility/caption, keyword use and social section. User/reviewer can request revision; zero visuals prompts pushback. Domain is presentation readiness; “real examples” asks for authenticity without an evidence procedure. No factual answer oracle, open publication preparation. SRC-1, doctrine/design, _meta/instructions/blog-posts/instructions.md.

• Immediate return: Review/pushback
• Later read-back: inapplicable — no persisted review report or later reviewer invocation is specified
• Delegated visibility: inapplicable — no delegated worker is specified
• Selection predicate: Visual plan and publishing checklist
• Invalidation or expiry: inapplicable — no retained result expiry specified
• Activation or effect: Requested review before publishing
• Evidence limits: Static source only; no observed host adherence, evidence-consuming acceptance or causal benefit
• design conclusion status: claimed
• operation conclusion status: uninspected

If a post has zero visual elements, push back and ask what could be added. --- _meta/instructions/blog-posts/instructions.md

EPI-RTE-blog-publish

Label: Blog publishing consumption

Closest supplied records: RT-RTE-capture-file. Distinct consequential function: blog publishing consumption, not the whole supplied invocation/container. No same-function supplied route declaration.

Reviewed draft → primary platform post → published status/date → cross-post adaptation → retained URLs. External author/assistant/platform executes; no target-owned posting client. User can stop publication; reviewer identity/mandatory human approval not specified. Published status is operational admission, not factual acceptance. Open publishing task; no rollback/unpublish procedure supplied. SRC-1, doctrine/design, _meta/instructions/blog-posts/instructions.md.

• Immediate return: Published/cross-posted artifact and links requested
• Later read-back: Status and URLs may be consumed by later file readers or audiences
• Delegated visibility: inapplicable — no delegated worker is specified
• Selection predicate: Publishing flow after review
• Invalidation or expiry: Dates updated and URLs tracked; no automatic withdrawal
• Activation or effect: Requested publication and tracking
• Evidence limits: Static source only; no observed host adherence, evidence-consuming acceptance or causal benefit
• design conclusion status: claimed
• operation conclusion status: uninspected

1. Track — Update the post file with the published URL and any cross-post links. --- _meta/instructions/blog-posts/instructions.md

EPI-RTE-pattern-capture

Label: Conversation pattern capture

Closest supplied records: RT-RTE-session-start and RT-RTE-capture-file. Distinct consequential function: conversation pattern capture, not the whole supplied invocation/container. No same-function supplied route declaration.

Conversation-established rule/preference/detail → assistant immediate write to context/config/module → future startup/task read. Assistant identifies pattern; ask if ambiguous, user can reject or override. Direct rule edits differ from approval-gated restructuring. Trigger is explicit rule or inferred usage pattern; policy change or imported personal assertion persists as text. No required source-truth oracle. Open personal use; user edits and available Git history offer nonautomatic recovery. SRC-1, doctrine/design, _meta/instructions/general.md.

• Immediate return: Immediate file update requested
• Later read-back: Context/conventions/modules requested inputs to later sessions
• Delegated visibility: inapplicable — no delegated worker is specified
• Selection predicate: Established conversation pattern; ask when ambiguous
• Invalidation or expiry: User override or replacement; no general expiry
• Activation or effect: Requested durable policy/personality adaptation; no observed later dependence
• Evidence limits: Static source only; no observed host adherence, evidence-consuming acceptance or causal benefit
• design conclusion status: claimed
• operation conclusion status: uninspected

If it's ambiguous whether something is worth capturing, ask: "We worked out [X] — should I add that to the instructions?" Then capture it in the same exchange, not at the end of the session. --- _meta/instructions/general.md

See RT-BAP-instruction-load for later policy consumer/channel/force/horizon. Preference capture need not formulate a theory; learning benefit is uninspected. For inferred conventions, the textual unit is addressable and persistent by design, consumption is prescribed, and no content-directed criticism episode or criticism-fed iteration is supplied. No autonomous or reflective theory-builder verdict follows. This is separate from the optimization theory account in RT-RTE-context-review.

EPI-RTE-metadata-transform

Label: Metadata text transformation

Part of: RT-RTE-metadata-migrate

Operator invokes utility → fixed regex scans/collects legacy fields → formatter builds frontmatter/title/body → direct overwrite. This is the transformation part of the supplied batch route, not a new invocation. Caller initiates mutation, code decides matches, no per-file user veto, independent oracle or expected-output comparison. Fixed batch mode. No backup/transaction; Git recovery only if available. SRC-1, implementation, migrate-to-yaml-frontmatter.py.

• Immediate return: Converted label after write
• Later read-back: Rewritten text becomes later reader/parser input
• Delegated visibility: inapplicable — no delegated worker is specified
• Selection predicate: Leading frontmatter skip; first H1 followed by matching bullet fields
• Invalidation or expiry: Rerun skips leading frontmatter; no semantic invalidation
• Activation or effect: Implemented direct overwrite, not executed
• Evidence limits: Static source only; no target execution or comparator; semantic preservation not guaranteed
• implementation conclusion status: wired
• operation conclusion status: uninspected

body_lines = lines[i:]

--- migrate-to-yaml-frontmatter.py

Static concrete witness, not a target fixture or execution: input Caution: obsolete instructions.\n# Guide\n\n- **Type:** knowledge\n- **Summary:** yes\n\n---\n\nBody.\n. title_index selects # Guide; the metadata regex collects type and summary; body_lines starts after the separator; the builder emits ---\ntype: knowledge\nsummary: yes\n---\n\n# Guide\n\nBody.\n. The pre-title caution is lost. An unquoted summary: yes is also not forced to remain a string for all YAML consumers. The direct write return merely says converted. This refutes universal lossless reshaping; it does not establish an ampliative conjecture. For well-formed restricted inputs the utility may only reshape, but the general edge is indeterminate without a checked source/consumer domain. Duplicate field names additionally overwrite earlier values in the dict. SRC-1, implementation:

        fields[m.group(1).lower()] = m.group(2).strip()

--- migrate-to-yaml-frontmatter.py

title_line = lines[title_index]
new_content = '\n'.join(yaml) + '\n\n' + title_line + '\n'

--- migrate-to-yaml-frontmatter.py

No supersession of the supplied migration container is needed: its warning about leading material and absence of validation is consistent with this narrower transformation record.

Claims

EPI-CLM-decision-learning

Label: Decision experience learning

Claimed operation: outcome-bearing past decisions become operational experience that informs later choices. Closest supplied route RT-RTE-knowledge-query describes delivery, but no supplied claim record declares this learning assertion. SRC-1, doctrine/design, _meta/instructions/knowledge-query.md:

The system doesn't just remember what you decided — it learns from how those decisions played out. Over time, your decision records become operational experience, not just documentation. --- _meta/instructions/knowledge-query.md

Supported scope: requested precedent consumption. Learned capacity and causal benefit uninspected; outcome success does not validate a mechanism or its transfer.

EPI-CLM-standard-enforcement

Label: Universal metadata enforcement

Claimed operation: every content file has required Tier 1 fields and lint flags missing fields. Distinct from RT-CLM-review-policy, which concerns review, not schema enforcement; closest supplied migration route warns it is no validator. SRC-1, doctrine/design, _config/standard.md:

Tier 1 — Universal (always enforced) Required on every content file. Non-negotiable. Lint will flag missing Tier 1 fields. --- _config/standard.md

Specified lint scope and type: presence test in EPI-RTE-lint-check are insufficient for that universal guarantee. This elaborates, rather than supersedes, the supplied runtime account's metadata warning.

EPI-CLM-knowledge-memory

Label: Institutional knowledge memory

Claimed operation: captures/processes persist and are queryable, while usage improves the system and prior decisions inform future ones. Closest supplied RT-CLM-review-policy concerns admission only; this distinct claim describes persistent knowledge/use/adaptation. SRC-1, doctrine/design, PHILOSOPHY.md:

• Past decisions inform future decisions (precedent chain)
• Patterns emerge across journal entries and knowledge articles
• Stale content gets surfaced automatically (lint)

• The system improves as you use it (training period) --- PHILOSOPHY.md

• Processes are captured once and queryable forever --- PHILOSOPHY.md

Prompt paths support intended persistence/consumption, not universal retention or factual correctness. The “automatically” lint wording does not supply a scheduler; startup checks old inbox items but full lint is request/cadence guidance. No causal improvement or infinite retention evidence is supplied. Completion's useful/findable criterion and accuracy requirement limit reliance as requested policy, not demonstrated acceptance.

Behavioral-authority paths

none declared in this member; see RT-BAP-instruction-load and the ledger's route-specific consumers.