Nemsy epistemic report

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

Source-and-claim boundary

This assesses the whole shipped document-to-Wiki-to-answer loop fixed in boundary, using its Source register. SRC-1 supplies implementation and doctrine/design, not observed operation. The question is whether imported documents and model responses acquire warranted reliance, rather than merely being stored and reused. Exclusions and dependency boundaries are see boundary; no development-plane or roadmap claim is assessed.

Public/design claims are RT-CLM-wiki-grounding and EPI-CLM-knowledge-accumulation. The former is a generation constraint; the latter promises accumulation and requests related-page updates, not a formal proof or universal truth oracle.

Entry point or operation	Source path	Coverage and records
CLI default chat, slash commands, ingest/query/lint and reasoning alternate	SRC-1: src/nemsy/cli.py	Supplied runtime dispatch findings: RT-RTE-conversation-chat, RT-RTE-document-ingest, RT-RTE-wiki-answer, RT-RTE-reasoning-query, RT-RTE-wiki-lint; direct focused read of smoke dispatch
Agent generation, parser, archive, save and context loader	SRC-1: src/nemsy/agent.py	Direct complete read; all five task prompts and their consumers; EPI-RTE-summary-cleanup plus runtime routes
Web chat/query/lint/ingest/save	SRC-1: src/nemsy/web.py	Direct read of these handlers; independent prompt construction checked against RT-RTE-wiki-answer, RT-RTE-insight-save and RT-RTE-wiki-lint
Wiki listing, citation resolution and file viewing	SRC-1: src/nemsy/web.py, frontend/src/components/MessageContent.tsx	Direct component/resolver reads plus correction-round Chat callback and file endpoint reads; EPI-RTE-citation-display; browser save/history boundary otherwise supplied RT-RTE-insight-save
Status, source list, balance and settings	SRC-1: src/nemsy/web.py, src/nemsy/cli.py	Endpoint/function enumeration directly searched; supplied runtime account covers counters and inert settings POST. Accounting, not truth acceptance; detailed rendering uninspected, preventing a complete UI-correctness claim
Hash/status, persistence, deletion and guidance	SRC-1: src/nemsy/vault.py	Direct hash/status read; write/delete/guidance responsibilities otherwise supplied RT-RTE-document-ingest and runtime account. No inference that deleting a page retracts its copies
LLM dispatch and provider state	SRC-1: src/nemsy/llm.py	Direct message construction, chat and stream calls; text-only request without registered tools; reasoning/text extraction otherwise supplied RT-CMP-deepseek and RT-RTE-reasoning-query
Page schema and smoke evaluators	SRC-1: src/nemsy/schema.py, tests/test_smoke.py, src/nemsy/cli.py	Direct complete schema/test reads; EPI-RTE-smoke-check, EPI-RTE-smoke-disposition
Public intent	SRC-1: README.md	Direct opening/structure read; EPI-CLM-knowledge-accumulation; remainder not used as evidence

Missing user documents and actual generated pages prevent proposition-level fidelity, entailment or novelty judgments. Missing service traces prevent claims of compliance or activated reliance. Missing interventions/comparisons prevent benefit and component-effect attribution. Opaque provider internals prevent an account of hidden model evaluation or training. The available source is sufficient to distinguish implemented request, admission and consumption; it does not license a system-complete negative about every possible evaluator or external human workflow.

Epistemic-object inventory

Object ID	Candidate truth-apt content or none	Claimed role	Evidence source ID and local anchor	Gap/limit
EPI-OBJ-source-content	see record	see record	SRC-1: src/nemsy/agent.py	see record
EPI-OBJ-summary-content	see record	see record	SRC-1: src/nemsy/agent.py	see record
EPI-OBJ-questions-content	see record	see record	SRC-1: src/nemsy/agent.py	see record
EPI-OBJ-answer-content	see record	see record	SRC-1: src/nemsy/agent.py	see record
EPI-OBJ-insight-content	see record	see record	SRC-1: src/nemsy/agent.py	see record
EPI-OBJ-lint-content	see record	see record	SRC-1: src/nemsy/agent.py	see record
EPI-OBJ-index-content	see record	see record	SRC-1: src/nemsy/agent.py	see record
RT-CMP-deepseek	not determinable at individual parametric proposition level	see record	SRC-1: src/nemsy/llm.py	Remote weights and their provenance uninspected; no inference from Wiki updates to weight learning

Authority-route ledger

Each compact entry separates a consequential function. “See” leaves the declaring record's endpoints, timing and evidence limits with that record. Architectural status below describes source wiring, not exercised behavior. Retention is not treated as lifecycle integration.

Route ID: RT-RTE-document-ingest Route function: content transformation Architectural status: implemented Object/candidate ID: EPI-OBJ-source-content → EPI-OBJ-summary-content Content/update relation: truth-apt transformation: acquisition/import Transition or check target: Local source text enters generation Evaluator/condition and domain: Caller supplies content; no source-veracity check Activation and timing: see RT-RTE-document-ingest Possible or observed result: Source and Wiki context delivered Implemented force: Permissive input admission Epistemic authority and scope: Source warrant unknown; import does not endorse source Operational authority: Allows document text to inform a proposed summary Behavioral-authority path: Configured model, request text, advisory, one ingest Evidence source ID and local anchor: see RT-RTE-document-ingest Claim IDs: none Mismatch marker: none Gap/limit: No supplied run; delivery is not demonstrated activation.

Route ID: RT-RTE-document-ingest Route function: content transformation Architectural status: implemented Object/candidate ID: EPI-OBJ-summary-content Content/update relation: truth-apt transformation: indeterminate Transition or check target: Document plus selected Wiki to generated SUMMARY Evaluator/condition and domain: Model requested to extract and relate; domain is supplied text Activation and timing: see RT-RTE-document-ingest Possible or observed result: Candidate Markdown Implemented force: Generation, not verification Epistemic authority and scope: No checked entailment or evidence-consuming acceptance Operational authority: Makes a candidate available to parser Behavioral-authority path: Parser, response string, permissive, this ingest Evidence source ID and local anchor: see RT-RTE-document-ingest Claim IDs: EPI-CLM-knowledge-accumulation Mismatch marker: none Gap/limit: Could faithfully reshape, derive or conjecture; actual output needed. Non-entailment is not established merely by generation.

Route ID: EPI-RTE-summary-cleanup Route function: content transformation Architectural status: implemented Object/candidate ID: EPI-OBJ-summary-content; EPI-OBJ-questions-content Content/update relation: truth-apt transformation: non-ampliative reshaping Transition or check target: see EPI-RTE-summary-cleanup Evaluator/condition and domain: see EPI-RTE-summary-cleanup Activation and timing: see EPI-RTE-summary-cleanup Possible or observed result: see EPI-RTE-summary-cleanup Implemented force: Enforcing text selection, not truth validation Epistemic authority and scope: No increased warrant Operational authority: Changes which returned spans survive Behavioral-authority path: Parser, regex/string operations, enforcing, saved page Evidence source ID and local anchor: see EPI-RTE-summary-cleanup Claim IDs: none Mismatch marker: none Gap/limit: No supplied run; delivery is not demonstrated activation.

Route ID: RT-RTE-document-ingest Route function: disposition/acceptance Architectural status: implemented Object/candidate ID: EPI-OBJ-summary-content Content/update relation: no content change Transition or check target: Presence of SUMMARY delimiter, not accuracy Evaluator/condition and domain: Python regex in response-format domain Activation and timing: see RT-RTE-document-ingest Possible or observed result: None on missing delimiter; text eligible otherwise Implemented force: Syntactic veto only Epistemic authority and scope: Not epistemic acceptance Operational authority: Allows write without per-content semantic review Behavioral-authority path: Writer, parse result, permissive, current write Evidence source ID and local anchor: see RT-RTE-document-ingest Claim IDs: EPI-CLM-knowledge-accumulation Mismatch marker: none Gap/limit: No supplied run; delivery is not demonstrated activation.

Route ID: RT-RTE-document-ingest Route function: retention Architectural status: implemented Object/candidate ID: EPI-OBJ-summary-content; EPI-OBJ-questions-content Content/update relation: no content change Transition or check target: Candidate page to filesystem Evaluator/condition and domain: see RT-RTE-document-ingest Activation and timing: see RT-RTE-document-ingest Possible or observed result: see RT-RTE-document-ingest Implemented force: Durable candidate use Epistemic authority and scope: No added license from storage Operational authority: Page becomes eligible for recency read-back Behavioral-authority path: Wiki loader then model, file text, permissive then advisory, later requests Evidence source ID and local anchor: see RT-RTE-document-ingest Claim IDs: none Mismatch marker: none Gap/limit: No supplied run; delivery is not demonstrated activation.

Route ID: RT-RTE-document-ingest Route function: lineage/freshness/recovery Architectural status: implemented Object/candidate ID: EPI-OBJ-source-content Content/update relation: no content change Transition or check target: Source hash equality; not source truth or summary correctness Evaluator/condition and domain: Stripped UTF-8 SHA-256 comparison, direct inspected vault code Activation and timing: see RT-RTE-document-ingest Possible or observed result: new/changed/done/empty Implemented force: CLI skip/force policy; Web differs Epistemic authority and scope: Hash equality only, not endorsement Operational authority: Unchanged input may skip regeneration even after a failed parse Behavioral-authority path: CLI pending selector, JSON hash, enforcing unless force, later ingest Evidence source ID and local anchor: see RT-RTE-document-ingest Claim IDs: none Mismatch marker: none Gap/limit: see RT-RTE-document-ingest; page/result-independent recording limits 'done'.

Route ID: RT-RTE-wiki-answer Route function: operational admission/selection/consumption Architectural status: implemented Object/candidate ID: EPI-OBJ-index-content; EPI-OBJ-summary-content; EPI-OBJ-answer-content; EPI-OBJ-insight-content Content/update relation: truth-apt transformation: non-ampliative reshaping Transition or check target: Index plus count-bounded recency pages enter request Evaluator/condition and domain: Python mtime/count selector; model relevance interpretation Activation and timing: see RT-RTE-wiki-answer Possible or observed result: Subset concatenated with file labels Implemented force: Selection enforcing; interpretation advisory Epistemic authority and scope: Retained page status supplies no correctness license Operational authority: Allows retained text to inform answers Behavioral-authority path: see RT-BAP-wiki-context Evidence source ID and local anchor: see RT-RTE-wiki-answer Claim IDs: RT-CLM-wiki-grounding Mismatch marker: none Gap/limit: No supplied run; delivery is not demonstrated activation.

Route ID: RT-RTE-wiki-answer Route function: content transformation Architectural status: implemented Object/candidate ID: EPI-OBJ-answer-content Content/update relation: truth-apt transformation: indeterminate Transition or check target: Wiki text and question to answer/citations Evaluator/condition and domain: Model; open questions without a reference oracle Activation and timing: see RT-RTE-wiki-answer Possible or observed result: Answer including possible abstention or ARCHIVABLE marker Implemented force: Returned directly, no semantic veto Epistemic authority and scope: Grounding policy only Operational authority: Answer delivered before independent checking Behavioral-authority path: Human reader, text/SSE, advisory, current question Evidence source ID and local anchor: see RT-RTE-wiki-answer Claim IDs: RT-CLM-wiki-grounding Mismatch marker: none Gap/limit: Entailment versus conjecture needs actual inputs/output; reasoning alternate has the separately recorded tuple-consumer defect.

Route ID: RT-RTE-conversation-chat Route function: operational admission/selection/consumption Architectural status: implemented Object/candidate ID: EPI-OBJ-answer-content Content/update relation: no content change Transition or check target: Wiki snapshot/history to subsequent chat turn Evaluator/condition and domain: see RT-RTE-conversation-chat Activation and timing: see RT-RTE-conversation-chat Possible or observed result: see RT-RTE-conversation-chat Implemented force: Context admission Epistemic authority and scope: Conversation reuse is not independent corroboration Operational authority: Prior text can influence next reply Behavioral-authority path: see RT-BAP-wiki-context Evidence source ID and local anchor: see RT-RTE-conversation-chat Claim IDs: RT-CLM-wiki-grounding Mismatch marker: none Gap/limit: No supplied run; delivery is not demonstrated activation.

Route ID: RT-RTE-answer-archive Route function: disposition/acceptance Architectural status: implemented Object/candidate ID: EPI-OBJ-answer-content Content/update relation: no content change Transition or check target: Caller archive flag, not answer truth Evaluator/condition and domain: Caller operation choice; no expected-answer oracle Activation and timing: see RT-RTE-answer-archive Possible or observed result: Eligible answer archived irrespective of ARCHIVABLE Implemented force: Permissive flag Epistemic authority and scope: Not evidence-consuming epistemic acceptance Operational authority: Authorizes write before caller review of that generated answer Behavioral-authority path: Archive helper, flag and answer text, permissive, current query Evidence source ID and local anchor: see RT-RTE-answer-archive Claim IDs: none Mismatch marker: none Gap/limit: No supplied run; delivery is not demonstrated activation.

Route ID: RT-RTE-answer-archive Route function: retention Architectural status: implemented Object/candidate ID: EPI-OBJ-answer-content Content/update relation: truth-apt transformation: non-ampliative reshaping Transition or check target: Strip ARCHIVABLE marker, add query metadata and index Evaluator/condition and domain: see RT-RTE-answer-archive Activation and timing: see RT-RTE-answer-archive Possible or observed result: see RT-RTE-answer-archive Implemented force: Filesystem retention Epistemic authority and scope: No added warrant Operational authority: Eligible for later Wiki read-back Behavioral-authority path: Wiki loader/model, Markdown, advisory, later invocations Evidence source ID and local anchor: see RT-RTE-answer-archive Claim IDs: none Mismatch marker: none Gap/limit: No supplied run; delivery is not demonstrated activation.

Route ID: RT-RTE-insight-save Route function: content transformation Architectural status: implemented Object/candidate ID: EPI-OBJ-insight-content Content/update relation: truth-apt transformation: indeterminate Transition or check target: History to rewritten insight page Evaluator/condition and domain: Model; requested conclusions/background/links, no supplied fresh Wiki Activation and timing: see RT-RTE-insight-save Possible or observed result: see RT-RTE-insight-save Implemented force: Generation Epistemic authority and scope: Neither format nor a 'background and derivation' heading proves entailment Operational authority: Proposes a replacement representation of the chosen history Behavioral-authority path: Save parser, returned Markdown, permissive, one save Evidence source ID and local anchor: see RT-RTE-insight-save Claim IDs: none Mismatch marker: none Gap/limit: Could reshape or add content; actual history/output needed.

Route ID: RT-RTE-insight-save Route function: disposition/acceptance Architectural status: implemented Object/candidate ID: EPI-OBJ-insight-content Content/update relation: no content change Transition or check target: Nonempty history and caller save request Evaluator/condition and domain: Human chooses operation/history boundary; code/model choose content/title Activation and timing: see RT-RTE-insight-save Possible or observed result: see RT-RTE-insight-save Implemented force: Operation approval, not particular-content approval Epistemic authority and scope: No semantic acceptance decision Operational authority: Allows generated page write without post-generation confirmation Behavioral-authority path: Writer, parsed text/title, permissive, current save Evidence source ID and local anchor: see RT-RTE-insight-save Claim IDs: none Mismatch marker: none Gap/limit: No supplied run; delivery is not demonstrated activation.

Route ID: RT-RTE-insight-save Route function: retention Architectural status: implemented Object/candidate ID: EPI-OBJ-insight-content Content/update relation: no content change Transition or check target: Generated insight and index to disk Evaluator/condition and domain: see RT-RTE-insight-save Activation and timing: see RT-RTE-insight-save Possible or observed result: see RT-RTE-insight-save Implemented force: Durable candidate Epistemic authority and scope: No added warrant Operational authority: Later context eligible; cached CLI snapshot unchanged Behavioral-authority path: Wiki loader/model, Markdown, advisory, later loads Evidence source ID and local anchor: see RT-RTE-insight-save Claim IDs: none Mismatch marker: none Gap/limit: No supplied run; delivery is not demonstrated activation.

Route ID: RT-RTE-wiki-lint Route function: check/evidence production Architectural status: implemented Object/candidate ID: EPI-OBJ-lint-content Content/update relation: truth-apt transformation: indeterminate Transition or check target: Contradictions, obsolete claims, orphans and gaps in supplied Wiki window Evaluator/condition and domain: Model diagnosis; no independent source/reference oracle; graph coverage windowed Activation and timing: see RT-RTE-wiki-lint Possible or observed result: ISSUES/SUGGESTIONS/NEXT_SOURCES text requested Implemented force: Advisory to user, not enforcing against a page Epistemic authority and scope: Diagnostic candidates, not established contradictions Operational authority: No installed page correction or changed reliance state Behavioral-authority path: User, report/SSE, advisory, current maintenance request Evidence source ID and local anchor: see RT-RTE-wiki-lint Claim IDs: EPI-CLM-knowledge-accumulation Mismatch marker: none Gap/limit: Missing outputs prevent evaluating an individual diagnosis; no automatic lint-result withdrawal/rejection consumer in these handlers.

Route ID: EPI-RTE-citation-display Route function: lineage/freshness/recovery Architectural status: implemented Object/candidate ID: EPI-OBJ-answer-content; resolved retained pages: MEM-OBJ-source-page, MEM-OBJ-query-page, MEM-OBJ-insight-page Content/update relation: truth-apt transformation: non-ampliative reshaping Transition or check target: see EPI-RTE-citation-display Evaluator/condition and domain: see EPI-RTE-citation-display Activation and timing: see EPI-RTE-citation-display Possible or observed result: see EPI-RTE-citation-display Implemented force: Navigational, not evidential endorsement Epistemic authority and scope: A clickable page is not a verified support relation Operational authority: Allows reader to inspect a resolved page Behavioral-authority path: Browser/user, numbered links then file API, advisory, current display Evidence source ID and local anchor: see EPI-RTE-citation-display Claim IDs: none Mismatch marker: none Gap/limit: No supplied run; delivery is not demonstrated activation.

Route ID: EPI-RTE-smoke-check Route function: check/evidence production Architectural status: implemented Object/candidate ID: EPI-OBJ-summary-content; EPI-OBJ-answer-content Content/update relation: no content change Transition or check target: see EPI-RTE-smoke-check Evaluator/condition and domain: see EPI-RTE-smoke-check Activation and timing: see EPI-RTE-smoke-check Possible or observed result: see EPI-RTE-smoke-check Implemented force: Assertions halt test on failure Epistemic authority and scope: Only the asserted structural/substring properties Operational authority: Can fail the test invocation, not gate normal writes Behavioral-authority path: pytest, assertions, enforcing, selected test run Evidence source ID and local anchor: see EPI-RTE-smoke-check Claim IDs: none Mismatch marker: none Gap/limit: No supplied run; delivery is not demonstrated activation.

Route ID: EPI-RTE-smoke-disposition Route function: disposition/acceptance Architectural status: implemented Object/candidate ID: EPI-OBJ-summary-content; EPI-OBJ-answer-content Content/update relation: no content change Transition or check target: see EPI-RTE-smoke-disposition Evaluator/condition and domain: see EPI-RTE-smoke-disposition Activation and timing: see EPI-RTE-smoke-disposition Possible or observed result: see EPI-RTE-smoke-disposition Implemented force: Smoke command exit status Epistemic authority and scope: No truth or production-admission license Operational authority: Reports pass/fail, does not roll back already written pages Behavioral-authority path: CLI/caller, pytest return code, enforcing exit, smoke invocation Evidence source ID and local anchor: see EPI-RTE-smoke-disposition Claim IDs: none Mismatch marker: none Gap/limit: No supplied run; delivery is not demonstrated activation.

No lifecycle-integration entry is claimed: these retention paths do not first establish acceptance against an evidence-consuming content criterion. Index appending connects files navigationally, but does not convert admission into epistemic acceptance. No behavior/policy-adaptation entry is inferred merely from suggestions or the name “insight.”

System-claim versus route comparison

Claim ID	Claimed operation or warrant	Claim source ID/anchor and layer	Doctrine/design support	Implemented route IDs	Observed-run support	Causal support and design limits	Supported conclusion	Mismatch/unknown
RT-CLM-wiki-grounding	see record	see record	see record	RT-RTE-wiki-answer, RT-RTE-conversation-chat, RT-BAP-wiki-context	None supplied	None supplied; prompt compliance and citation entailment unchecked	Wiki-only generation is requested	Confident style and citations do not prove warrant; partial context cannot prove whole-Wiki absence
EPI-CLM-knowledge-accumulation	see record	see record	see record	RT-RTE-document-ingest, RT-RTE-insight-save, RT-RTE-answer-archive, RT-RTE-wiki-lint	None supplied	No comparison; related-page edits are not implemented by a general edit-plan executor	Accumulation by caller-triggered generation/storage and index append	Broader autonomous maintenance or improved knowledge quality unestablished

Bounded conclusion

Nemsy imports local documents with unknown source warrant, generates candidate pages and answers, and reuses recent retained text. See RT-RTE-document-ingest, RT-RTE-wiki-answer, RT-RTE-answer-archive and RT-RTE-insight-save. These operations are wired, not observed. A source hash controls redundant work, not reliability. The parser can remove text and admit a malformed or unsupported summary; EPI-RTE-summary-cleanup demonstrates its narrow transformation without executing the target.

There is no demonstrated produced accepted ampliative claim in these routes. Without actual source/output pairs their generation relation is indeterminate, not automatically novel conjecture. Prompted synthesis and a derivation heading do not establish entailed derivation. No warranted premises plus checked interpretation establish a new conclusion here. No actual user source is available to decide that more specifically.

Checking is heterogeneous. The SUMMARY delimiter checks parseability. RT-RTE-wiki-lint requests model criticism of a limited page window, but its suggestions have no implemented edit/withdrawal consumer. EPI-RTE-smoke-check asserts limited structural and substring criteria; EPI-RTE-smoke-disposition consumes test success only for a smoke command's message/exit. The test writes precede checks and are not rolled back. These are real limited evaluators, counterexamples to “no checks,” but not semantic acceptance of retained claims.

Actual operational authority lies in writes and later context delivery: caller-selected operations permit code to write model-selected content, and the model later receives it as Wiki text. See RT-BAP-wiki-context. Schema, freshness and link rendering do not add truth warrant. The browser can resolve a claimed citation and deliver current retained Wiki page text to a later human browser consumer without checking that it supports the answer; EPI-RTE-citation-display. Regeneration/overwrites and the caller-trusted deletion helper described in the runtime account supply no claim-level retraction propagation to archived answers, insights or a cached CLI Wiki snapshot.

Provider parameters, exact version and endpoint fixity are see RT-CMP-deepseek. Hidden learning or evaluation is not determinable. Persisted insights afford future consumption; they do not alone demonstrate improved capacity. Lint has no retained criticism-to-next-round revision loop in the inspected functions. Thus this account establishes neither an exercised self-improvement pathway nor autonomous evidence-responsive theory revision. These limits do not negate the separate implemented acquisition, generation, diagnostic and retention routes, and give Nemsy no single epistemic score.

Shared records

Components

none declared in this member

Operative objects

EPI-OBJ-source-content

Label: Imported source text

SRC-1: src/nemsy/agent.py. Natural-language assertions in caller UTF-8 documents; filesystem substrate before import, request string during ingestion. External claims are acquired, not produced. Directory guidance is separate instruction text in the same request. Particular propositions and source warrant unknown without a supplied Vault. Distinct identity: operative content object, not the progression represented by RT-RTE-document-ingest; no operative-object counterpart is declared in the supplied runtime. Evidence passages: see RT-RTE-document-ingest and the task/cleanup passages below.

EPI-OBJ-summary-content

Label: SUMMARY page body

SRC-1: src/nemsy/agent.py. Natural-language generated claims and links in a response string, then Markdown file. Requested type/source/date metadata is symbolic labeling; the unvalidated body supplies future model-interpreted content. No particular proposition or faithful compression is demonstrated. Distinct identity: operative content object, not the progression represented by RT-RTE-document-ingest; no operative-object counterpart is declared in the supplied runtime. Evidence passages: see RT-RTE-document-ingest and the task/cleanup passages below.

EPI-OBJ-questions-content

Label: QUESTIONS appendix

SRC-1: src/nemsy/agent.py. Natural-language questions, requested to distinguish original-source clues from external-verification needs; normally not truth-apt questions themselves, although embedded premises/labels may assert claims. Response substring appended to the source Markdown, not a research-result evaluator. Distinct identity: operative content object, not the progression represented by RT-RTE-document-ingest; no operative-object counterpart is declared in the supplied runtime. Evidence passages: see RT-RTE-document-ingest and the task/cleanup passages below.

EPI-OBJ-answer-content

Label: Answer and archived query body

SRC-1: src/nemsy/agent.py. Natural-language answers/abstentions with [[page]] references and optional ARCHIVABLE marker in returned strings, console/SSE and optional query Markdown. Wiki-only and synthesis criteria are instructions, not checked facts. Archive metadata/marker stripping are symbolic formatting, not new warrant. Distinct identity: operative content object, not the progression represented by RT-RTE-wiki-answer; no operative-object counterpart is declared in the supplied runtime. Evidence passages: see RT-RTE-wiki-answer and the task/cleanup passages below.

EPI-OBJ-insight-content

Label: Conversation insight page

SRC-1: src/nemsy/agent.py. Natural-language conclusions, background and linked knowledge generated from selected history, stored as Markdown after code-fence removal. Symbolic title extraction controls filename. No new Wiki is loaded for this call, and the saved content is not reviewed before write. Distinct identity: operative content object, not the progression represented by RT-RTE-insight-save; no operative-object counterpart is declared in the supplied runtime. Evidence passages: see RT-RTE-insight-save and the task/cleanup passages below.

EPI-OBJ-lint-content

Label: Wiki lint report

SRC-1: src/nemsy/agent.py. Natural-language diagnostic assertions and suggested actions returned from a model request; response-string substrate. Contradiction/orphan/obsolescence targets differ, but none has an independent formal evaluator in this route. No retained report or installed correction in the handler. Distinct identity: operative content object, not the progression represented by RT-RTE-wiki-lint; no operative-object counterpart is declared in the supplied runtime. Evidence passages: see RT-RTE-wiki-lint and the task/cleanup passages below.

EPI-OBJ-index-content

Label: Wiki index entries

SRC-1: src/nemsy/agent.py. Markdown index mixes navigational [[stem]] identifiers and natural-language summary labels plus date. Filesystem substrate; loader preferentially supplies index to later requests. Label 'summary' or 'query result' denotes provenance category, not acceptance. Repeated appends are not a deduplicated truth registry. Distinct identity: operative content object, not the progression represented by RT-RTE-document-ingest; no operative-object counterpart is declared in the supplied runtime. Evidence passages: see RT-RTE-document-ingest and the task/cleanup passages below.

entry = f"- [[{stem}]] — {summary_line} （{date.today()}）\n"

--- src/nemsy/agent.py

[值得进一步探究的问题列表，每条标注：【原文已有线索】或【需外部文献验证】]

--- src/nemsy/agent.py

1. 仅当答案是对多个页面的深度综合分析、且结论不能从任何单一页面直接读出时， 才在末尾单独一行标注 ARCHIVABLE: true； 简单检索、单页引用、或仅整理已有内容不标注 --- src/nemsy/agent.py

Routes

EPI-RTE-summary-cleanup

Label: SUMMARY selection and appendix cleanup Part of: RT-RTE-document-ingest

SRC-1: src/nemsy/agent.py, implementation. This is the deterministic text-selection part of ingestion, not another end-to-end ingestion declaration. Python owns response-to-page progression: regex selects SUMMARY, removes an enclosing fence, deletes a heading-matched suffix, and appends separately extracted QUESTIONS. The caller/model proposal and admission limits are see RT-RTE-document-ingest. No semantic oracle or human review occurs in this cleanup.

• implementation conclusion status: wired
• operation conclusion status: uninspected
• Immediate return: Cleaned summary string proceeds to the page writer; missing SUMMARY returns None.
• Later read-back: Surviving text reaches later Wiki loads after the enclosing ingest writes it; removed spans are not separately retained by this parser.
• Delegated visibility: inapplicable — local string handling launches no worker or provider call.
• Selection predicate: SUMMARY delimiter and optional QUESTIONS delimiter; full enclosing code fence; suffix beginning at the matched pending-questions heading.
• Invalidation or expiry: inapplicable — transient intermediate string, not an independently versioned object.
• Activation or effect: Enforces textual deletion/recomposition before persistence, not accuracy or premise preservation.
• Evidence limits: Deduction from inspected code, no execution; arbitrary matched suffixes can contain substantive claims.

Concrete input trace, a deduction in the implementation layer: with response ---SUMMARY---\nA\n## 待探索问题\nB and no QUESTIONS delimiter, SUMMARY extraction first yields A\n## 待探索问题\nB. There is no enclosing fence. The first cleanup regex (requiring a separator) does not match; the second removes the heading and everything following it. The final saved body is A. Thus a real claim represented by B would be lost by position alone, not judged false. Conversely ---SUMMARY---\nA\n---QUESTIONS---\nQ retains A and appends Q under the questions heading. The route is non-ampliative reshaping/deletion, not a truth check; no semantic-preservation guarantee follows.

summary_match = re.search(r"---SUMMARY---\n(.*?)(?=---QUESTIONS---|$)", response, re.DOTALL) --- src/nemsy/agent.py

summary_content = re.sub( r"\n{0,2}## 待探索问题\b.*$", "", summary_content, flags=re.DOTALL, ).rstrip() --- src/nemsy/agent.py

if questions_text: summary_content = summary_content + "\n\n---\n\n## 待探索问题\n\n" + questions_text --- src/nemsy/agent.py

EPI-RTE-citation-display

Label: Answer citation rendering and resolution

SRC-1: frontend/src/components/MessageContent.tsx, frontend/src/views/Chat.tsx, src/nemsy/web.py, implementation. Distinct from RT-RTE-wiki-answer, which produces answers, this downstream browser/backend mechanism rewrites link markers, enumerates references and resolves a requested title to a file. The component returns reference titles through its callback; the resolver matches lowercased stems. Direct correction-round inspection confirms Chat's callback opens the resolved file endpoint in a new tab, which returns current UTF-8 text in JSON subject to the runtime's bounded lexical Vault check. This supplies inspectability and later human-facing read-back, not a proposition-to-evidence test. It overlaps MEM-RTE-browser-read at citation resolution and file delivery, but separately includes answer-marker rendering; neither record supersedes the other. Retained generated pages can be MEM-OBJ-source-page, MEM-OBJ-query-page or MEM-OBJ-insight-page.

• implementation conclusion status: wired
• operation conclusion status: uninspected
• Immediate return: Numbered reference display; resolver returns path metadata or 404; file endpoint returns UTF-8 text.
• Later read-back: Generated source summaries, archived query answers and saved insights can reach a later human browser consumer when a citation resolves to their retained Wiki file; the file endpoint returns current text in JSON. This route need not write content to deliver accumulated material. Viewing arbitrary static source text alone does not establish retained memory.
• Delegated visibility: inapplicable — browser callbacks/HTTP handlers, not a model or worker.
• Selection predicate: All regex-recognized double-bracket titles; deduplicated by trimmed title in appearance order; resolver uses first matching case-insensitive stem.
• Invalidation or expiry: Resolution uses the current list, with no pinned page version or retained historical evidence target.
• Activation or effect: Human can inspect a linked page; rendering does not test reference existence and resolution does not test support or entailment.
• Evidence limits: Direct reads establish marker rendering, Chat callback, resolver and file-content return, not observed interaction, human reliance or source/bundle equivalence. Read-back is conditional on resolution to a retained generated page, not established for every viewable file.

Concrete implementation deduction: A [[Page]] B [[Page]] produces one reference numbered 1; both occurrences become %%REF:1:Page%% before Markdown rendering. Whether Page supports A or B is never an input to these functions. The same renderer accepts a nonexistent Page, whose later resolution may return 404.

const title = match[1].trim() if (!seen.has(title)) { seen.set(title, counter++) }

--- frontend/src/components/MessageContent.tsx

return idx !== undefined ? %%REF:${idx}:${title.trim()}%% : title.trim() --- frontend/src/components/MessageContent.tsx

if note.stem.lower() == target: return {"title": title, "abs_path": str(note), "rel_path": str(note.relative_to(wiki_path))} --- src/nemsy/web.py

.then((d) => { if (d?.abs_path) window.open(/api/file?path=${encodeURIComponent(d.abs_path)}, '_blank') }) --- frontend/src/views/Chat.tsx

return {"path": path, "content": p.read_text(encoding="utf-8")} --- src/nemsy/web.py

EPI-RTE-smoke-check

Label: Smoke assertion evaluators

SRC-1: tests/test_smoke.py, implementation/test definitions. No counterpart after comparison with the runtime's routes: it mentions smoke tests but declares no evaluator record. Tests call ingest and query, then Python assertions assess paths, existence, metadata, length and answer substrings. Test authors provide a known source page and question; this is a bounded fixture-based mode, distinct from open user queries. The expected-answer oracle is weak: query permits any of DeepSeek, API or 暂无, not an exact reference answer. Ingest's oracle is structural metadata and length, not correspondence to source assertions. Archive checking asks for any dated query file and checks the latest one, not a guaranteed question-specific file. These checks are not an acceptance criterion for normal application knowledge writes.

• implementation conclusion status: wired
• operation conclusion status: uninspected
• Immediate return: Assertion pass or pytest failure; tests print diagnostic output.
• Later read-back: Tests write into the configured Wiki and require manual cleanup; generated pages can later be loaded irrespective of assertions.
• Delegated visibility: Live model calls receive fixture source and current Wiki through the ordinary agent routes.
• Selection predicate: Explicit pytest selection or smoke command, not an application admission hook.
• Invalidation or expiry: Fixture source file is removed by cleanup; generated Wiki files require manual cleanup, with no assertion-triggered rollback.
• Activation or effect: A failed assertion stops its test; passing does not independently establish grounded answers or source fidelity.
• Evidence limits: No supplied execution result. Structural checks, permissive disjunction and any-date-file archive lookup cannot establish semantic correctness, causal benefit or complete archive attribution.

assert "DeepSeek" in result or "API" in result or "暂无" in result --- tests/test_smoke.py

assert note.metadata["type"] == "source" assert "date" in note.metadata assert "tags" in note.metadata

--- tests/test_smoke.py

query_files = list(queries_dir.glob(f"*{date.today()}.md")) --- tests/test_smoke.py

EPI-RTE-smoke-disposition

Label: Smoke exit-result consumption

SRC-1: src/nemsy/cli.py, implementation. Distinct consumer of EPI-RTE-smoke-check: the smoke driver starts pytest, then prints success on zero or exits 1 otherwise. No counterpart route in the supplied runtime. Test code decides pass/fail, caller initiates the bounded smoke invocation and may decide subsequent action; no automatic product installation/admission or knowledge rollback consumes the result. The target is this test invocation's exit status, not Wiki truth.

• implementation conclusion status: wired
• operation conclusion status: uninspected
• Immediate return: Success message or failure message and process exit 1.
• Later read-back: inapplicable — the driver retains no acceptance artifact or revised policy for later runs.
• Delegated visibility: pytest subprocess runs the selected tests; this is not a model-selected agent delegation.
• Selection predicate: Explicit nemsy-smoke invocation, fixed tests/test_smoke.py.
• Invalidation or expiry: inapplicable — immediate process-result consumption, no persistent accepted status.
• Activation or effect: Changes the command's terminal outcome; no rollback of test-generated Wiki changes.
• Evidence limits: No observed run; a green message's implemented license is only the selected assertion suite.

if result.returncode == 0: console.print("\n[green]✓ 烟雾测试通过！[/green]") else: console.print("\n[red]✗ 烟雾测试失败，请检查输出。[/red]") sys.exit(1)

--- src/nemsy/cli.py

Claims

EPI-CLM-knowledge-accumulation

Label: Continuous accumulation and autonomous induction

SRC-1: README.md, src/nemsy/agent.py, doctrine/design. README calls Nemsy a self-building personal knowledge assistant and describes continuous accumulation and autonomous induction. BASE_SYSTEM requests updating related pages rather than merely summarizing and asks for contradiction marking. This differs from RT-CLM-wiki-grounding, which limits answers' information source. In inspected implementation, ingest creates one summary page and appends index; lint advises rather than automatically applying repairs. The model can include links/contradiction labels inside that new page, a genuine partial realization; it has no general arbitrary-page edit execution grant. “Self-building” is therefore not evidence of exercised self-improvement, independent truth acceptance or autonomous maintenance of the whole Wiki.

本项目基于 Karpathy 的 LLM Wiki 理念，由 DeepSeek 超长上下文模型驱动，以 Obsidian Vault 为知识源，为唯一用户提供持续积累、自主归纳的个人知识服务。 --- README.md

1. 知识积累优先：每次摄取新资料都要更新相关 Wiki 页面，不只是生成摘要
2. 交叉引用：若内容与已有 Wiki 页面有明确关联，用 [[页面名]] 标注；无需强行寻找关联
3. 矛盾标注：用 > ⚠️ 矛盾 格式标注与已有知识相悖的内容

--- src/nemsy/agent.py

Evidenced absences

none declared in this member

Behavioral-authority paths

none declared in this member