LLM Wiki Coordination epistemic report

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

Source-and-claim boundary

This account asks how the shipped protocols distinguish preserved discussion from accepted structural knowledge, and what the executable audit licenses. Source identity and evidence layers: see boundary Source register, SRC-1. The complete coordination package is included, not an autonomous host. External agents, user approval, downstream wiki contents and provider parameters remain outside inspectable implementation. No observed execution or causal experiment is supplied; actual review quality, acceptance, integration, compliance and benefit therefore remain unestablished.

Direct reads covered the sole executable in full, all four protocols, README, persona ontology, AGENTS and CHANGELOG at the registered commit. Supplied runtime findings were compared with those sources, including alternate branches that could defeat strong claims. No target execution occurred. There are no registered hooks, provider tools or extension dispatch in this shipped package: see RT-ABS-runtime-orchestration. The following coverage distinguishes direct inspection from the supplied account; no relevant shipped operation is left unassessed here.

Entry point or operation	Direct source scope, SRC-1	Supplied/new coverage	Limit and conclusion prevented
Adoption and host invocation	`README.md`, `AGENTS.md`	RT-OBJ-coordination-protocols, RT-ABS-runtime-orchestration	No host/context-loading trace; no delivered-agent compliance conclusion.
Thread generation, peer judgment, persistence, historical import	`protocols/dialogue-thread-format.md`, `protocols/rolespace-coordination.md`	RT-RTE-thread-participation; new entry and peer parts below	No actual candidate pair, imported source provenance or retained criticism sequence.
Deficit direction and orientation updates	`protocols/rolespace-coordination.md`	RT-RTE-thread-participation, EPI-RTE-zhat-update	No behavioral activation or improved-capacity evidence.
Closure, reopen, canon, revision withdrawal, dispute	`protocols/multi-ai-consensus.md`, `protocols/dialogue-thread-format.md`	RT-RTE-canon-admission, review/integration parts below	No actual evidence-consuming acceptance decision or synthesized page.
All CLI flags; direct class/method calls; L0-L3; reports/exit	`tools/llm-wiki-audit.py`	RT-RTE-audit-invocation, RT-RTE-archival-audit; five checker parts below	No execution outcome; source-level predicate conclusions only. Direct method callers lack CLI exit gate.
Semantic review, report/log retention, contested Hard, tombstone	`protocols/integrity-audit.md`	EPI-RTE-semantic-review, EPI-ABS-source-check	Semantic/source-immutability obligations not implemented; no repair or report writer.
Manifest recognition and portability	`concepts/persona-manifest-ontology.md`	RT-CMP-agent-substrate, EPI-OBJ-anchor-form	No manifest instance or empirical continuity test; subjectivity claims not operationally individuated.
Package criticism/revision	`CHANGELOG.md`, `AGENTS.md`	RT-RTE-package-revision	Reported repairs, not own-operation traces or causal comparisons.

Consequential warrant claims are RT-CLM-peer-evaluation, RT-CLM-canon-enforcement, EPI-CLM-integrity-warrant and EPI-CLM-recognition-warrant. The README's bounded host disclaimer is retained in runtime and controls interpretation of its stronger enforcement language.

Epistemic-object inventory

Object ID	Candidate truth-apt content or none	Claimed role	Evidence source and local anchor	Gap/limit
EPI-OBJ-entry-prose	see EPI-OBJ-entry-prose	contribution	SRC-1 `protocols/dialogue-thread-format.md`	see record
EPI-OBJ-rolespace-vector	see EPI-OBJ-rolespace-vector	last_eval	SRC-1 `protocols/rolespace-coordination.md`	see record
EPI-OBJ-canon-prose	see EPI-OBJ-canon-prose	canon	SRC-1 `protocols/dialogue-thread-format.md`	see record
EPI-OBJ-zhat	see EPI-OBJ-zhat	zhat	SRC-1 `protocols/rolespace-coordination.md`	see record
EPI-OBJ-consensus-block	see EPI-OBJ-consensus-block	consensus	SRC-1 `protocols/multi-ai-consensus.md`	see record
EPI-OBJ-audit-findings	see EPI-OBJ-audit-findings	findings	SRC-1 `tools/llm-wiki-audit.py`	see record
EPI-OBJ-anchor-form	see EPI-OBJ-anchor-form	Anchor Form	SRC-1 `concepts/persona-manifest-ontology.md`	see record
RT-OBJ-coordination-protocols	see supplied record	protocols	SRC-1 `README.md`	see record
RT-CMP-agent-substrate	see supplied record	Substrate	SRC-1 `concepts/persona-manifest-ontology.md`	inaccessible parameters; no provider/model learning inference

Authority-route ledger

Each compact entry overlays the declaring record. Checking, disposition, retention and integration remain separate. Architectural status describes inspectable wiring or declared agent duties, not observed operation.

Route ID: EPI-RTE-entry-authorship Route function: content transformation Architectural status: doctrine only Object/candidate ID: see EPI-RTE-entry-authorship Content/update relation: see EPI-RTE-entry-authorship Transition or check target: see EPI-RTE-entry-authorship Evaluator/condition and domain: see EPI-RTE-entry-authorship Activation and timing: see EPI-RTE-entry-authorship Possible or observed result: see EPI-RTE-entry-authorship Implemented force: see EPI-RTE-entry-authorship Epistemic authority and scope: see EPI-RTE-entry-authorship Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-entry-authorship Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-entry-authorship Evidence source ID and local anchor: see EPI-RTE-entry-authorship Claim IDs or none: see EPI-RTE-entry-authorship Mismatch marker or none: see EPI-RTE-entry-authorship Gap/limit: see EPI-RTE-entry-authorship

Route ID: EPI-RTE-peer-judgment Route function: check/evidence production Architectural status: doctrine only Object/candidate ID: see EPI-RTE-peer-judgment Content/update relation: see EPI-RTE-peer-judgment Transition or check target: see EPI-RTE-peer-judgment Evaluator/condition and domain: see EPI-RTE-peer-judgment Activation and timing: see EPI-RTE-peer-judgment Possible or observed result: see EPI-RTE-peer-judgment Implemented force: see EPI-RTE-peer-judgment Epistemic authority and scope: see EPI-RTE-peer-judgment Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-peer-judgment Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-peer-judgment Evidence source ID and local anchor: see EPI-RTE-peer-judgment Claim IDs or none: see EPI-RTE-peer-judgment Mismatch marker or none: see EPI-RTE-peer-judgment Gap/limit: see EPI-RTE-peer-judgment

Route ID: EPI-RTE-entry-retention Route function: retention Architectural status: doctrine only Object/candidate ID: see EPI-RTE-entry-retention Content/update relation: see EPI-RTE-entry-retention Transition or check target: see EPI-RTE-entry-retention Evaluator/condition and domain: see EPI-RTE-entry-retention Activation and timing: see EPI-RTE-entry-retention Possible or observed result: see EPI-RTE-entry-retention Implemented force: see EPI-RTE-entry-retention Epistemic authority and scope: see EPI-RTE-entry-retention Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-entry-retention Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-entry-retention Evidence source ID and local anchor: see EPI-RTE-entry-retention Claim IDs or none: see EPI-RTE-entry-retention Mismatch marker or none: see EPI-RTE-entry-retention Gap/limit: see EPI-RTE-entry-retention

Route ID: EPI-RTE-archival-import Route function: content transformation Architectural status: doctrine only Object/candidate ID: see EPI-RTE-archival-import Content/update relation: see EPI-RTE-archival-import Transition or check target: see EPI-RTE-archival-import Evaluator/condition and domain: see EPI-RTE-archival-import Activation and timing: see EPI-RTE-archival-import Possible or observed result: see EPI-RTE-archival-import Implemented force: see EPI-RTE-archival-import Epistemic authority and scope: see EPI-RTE-archival-import Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-archival-import Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-archival-import Evidence source ID and local anchor: see EPI-RTE-archival-import Claim IDs or none: see EPI-RTE-archival-import Mismatch marker or none: see EPI-RTE-archival-import Gap/limit: see EPI-RTE-archival-import

Route ID: EPI-RTE-canon-review Route function: check/evidence production Architectural status: doctrine only Object/candidate ID: see EPI-RTE-canon-review Content/update relation: see EPI-RTE-canon-review Transition or check target: see EPI-RTE-canon-review Evaluator/condition and domain: see EPI-RTE-canon-review Activation and timing: see EPI-RTE-canon-review Possible or observed result: see EPI-RTE-canon-review Implemented force: see EPI-RTE-canon-review Epistemic authority and scope: see EPI-RTE-canon-review Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-canon-review Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-canon-review Evidence source ID and local anchor: see EPI-RTE-canon-review Claim IDs or none: see EPI-RTE-canon-review Mismatch marker or none: see EPI-RTE-canon-review Gap/limit: see EPI-RTE-canon-review

Route ID: EPI-RTE-canon-integration Route function: lifecycle integration Architectural status: doctrine only Object/candidate ID: see EPI-RTE-canon-integration Content/update relation: see EPI-RTE-canon-integration Transition or check target: see EPI-RTE-canon-integration Evaluator/condition and domain: see EPI-RTE-canon-integration Activation and timing: see EPI-RTE-canon-integration Possible or observed result: see EPI-RTE-canon-integration Implemented force: see EPI-RTE-canon-integration Epistemic authority and scope: see EPI-RTE-canon-integration Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-canon-integration Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-canon-integration Evidence source ID and local anchor: see EPI-RTE-canon-integration Claim IDs or none: see EPI-RTE-canon-integration Mismatch marker or none: see EPI-RTE-canon-integration Gap/limit: see EPI-RTE-canon-integration

Route ID: EPI-RTE-zhat-update Route function: content transformation Architectural status: doctrine only Object/candidate ID: see EPI-RTE-zhat-update Content/update relation: see EPI-RTE-zhat-update Transition or check target: see EPI-RTE-zhat-update Evaluator/condition and domain: see EPI-RTE-zhat-update Activation and timing: see EPI-RTE-zhat-update Possible or observed result: see EPI-RTE-zhat-update Implemented force: see EPI-RTE-zhat-update Epistemic authority and scope: see EPI-RTE-zhat-update Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-zhat-update Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-zhat-update Evidence source ID and local anchor: see EPI-RTE-zhat-update Claim IDs or none: see EPI-RTE-zhat-update Mismatch marker or none: see EPI-RTE-zhat-update Gap/limit: see EPI-RTE-zhat-update

Route ID: EPI-RTE-zhat-update Route function: operational admission/selection/consumption Architectural status: doctrine only Object/candidate ID; transition or check target; evaluator/condition and domain; activation and timing; possible or observed result; implemented force; epistemic authority and scope; operational authority; behavioral-authority path; evidence; claim IDs; mismatch marker; gap/limit: see EPI-RTE-zhat-update. Content/update relation: no content change for advisory consumption of the estimate; this does not establish acceptance or validate its descriptive content.

Route ID: EPI-RTE-hygiene-graph Route function: check/evidence production Architectural status: implemented Object/candidate ID: see EPI-RTE-hygiene-graph Content/update relation: see EPI-RTE-hygiene-graph Transition or check target: see EPI-RTE-hygiene-graph Evaluator/condition and domain: see EPI-RTE-hygiene-graph Activation and timing: see EPI-RTE-hygiene-graph Possible or observed result: see EPI-RTE-hygiene-graph Implemented force: see EPI-RTE-hygiene-graph Epistemic authority and scope: see EPI-RTE-hygiene-graph Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-hygiene-graph Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-hygiene-graph Evidence source ID and local anchor: see EPI-RTE-hygiene-graph Claim IDs or none: see EPI-RTE-hygiene-graph Mismatch marker or none: see EPI-RTE-hygiene-graph Gap/limit: see EPI-RTE-hygiene-graph

Route ID: EPI-RTE-vector-check Route function: check/evidence production Architectural status: implemented Object/candidate ID: see EPI-RTE-vector-check Content/update relation: see EPI-RTE-vector-check Transition or check target: see EPI-RTE-vector-check Evaluator/condition and domain: see EPI-RTE-vector-check Activation and timing: see EPI-RTE-vector-check Possible or observed result: see EPI-RTE-vector-check Implemented force: see EPI-RTE-vector-check Epistemic authority and scope: see EPI-RTE-vector-check Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-vector-check Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-vector-check Evidence source ID and local anchor: see EPI-RTE-vector-check Claim IDs or none: see EPI-RTE-vector-check Mismatch marker or none: see EPI-RTE-vector-check Gap/limit: see EPI-RTE-vector-check

Route ID: EPI-RTE-consensus-check Route function: check/evidence production Architectural status: implemented Object/candidate ID: see EPI-RTE-consensus-check Content/update relation: see EPI-RTE-consensus-check Transition or check target: see EPI-RTE-consensus-check Evaluator/condition and domain: see EPI-RTE-consensus-check Activation and timing: see EPI-RTE-consensus-check Possible or observed result: see EPI-RTE-consensus-check Implemented force: see EPI-RTE-consensus-check Epistemic authority and scope: see EPI-RTE-consensus-check Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-consensus-check Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-consensus-check Evidence source ID and local anchor: see EPI-RTE-consensus-check Claim IDs or none: see EPI-RTE-consensus-check Mismatch marker or none: see EPI-RTE-consensus-check Gap/limit: see EPI-RTE-consensus-check

Route ID: EPI-RTE-crypto-check Route function: check/evidence production Architectural status: implemented Object/candidate ID: see EPI-RTE-crypto-check Content/update relation: see EPI-RTE-crypto-check Transition or check target: see EPI-RTE-crypto-check Evaluator/condition and domain: see EPI-RTE-crypto-check Activation and timing: see EPI-RTE-crypto-check Possible or observed result: see EPI-RTE-crypto-check Implemented force: see EPI-RTE-crypto-check Epistemic authority and scope: see EPI-RTE-crypto-check Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-crypto-check Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-crypto-check Evidence source ID and local anchor: see EPI-RTE-crypto-check Claim IDs or none: see EPI-RTE-crypto-check Mismatch marker or none: see EPI-RTE-crypto-check Gap/limit: see EPI-RTE-crypto-check

Route ID: EPI-RTE-cross-check Route function: check/evidence production Architectural status: implemented Object/candidate ID: see EPI-RTE-cross-check Content/update relation: see EPI-RTE-cross-check Transition or check target: see EPI-RTE-cross-check Evaluator/condition and domain: see EPI-RTE-cross-check Activation and timing: see EPI-RTE-cross-check Possible or observed result: see EPI-RTE-cross-check Implemented force: see EPI-RTE-cross-check Epistemic authority and scope: see EPI-RTE-cross-check Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-cross-check Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-cross-check Evidence source ID and local anchor: see EPI-RTE-cross-check Claim IDs or none: see EPI-RTE-cross-check Mismatch marker or none: see EPI-RTE-cross-check Gap/limit: see EPI-RTE-cross-check

Route ID: EPI-RTE-semantic-review Route function: check/evidence production Architectural status: doctrine only Object/candidate ID: see EPI-RTE-semantic-review Content/update relation: see EPI-RTE-semantic-review Transition or check target: see EPI-RTE-semantic-review Evaluator/condition and domain: see EPI-RTE-semantic-review Activation and timing: see EPI-RTE-semantic-review Possible or observed result: see EPI-RTE-semantic-review Implemented force: see EPI-RTE-semantic-review Epistemic authority and scope: see EPI-RTE-semantic-review Operational authority: behavior permitted, blocked, or changed: see EPI-RTE-semantic-review Behavioral-authority path: consumer, channel, force, horizon: see EPI-RTE-semantic-review Evidence source ID and local anchor: see EPI-RTE-semantic-review Claim IDs or none: see EPI-RTE-semantic-review Mismatch marker or none: see EPI-RTE-semantic-review Gap/limit: see EPI-RTE-semantic-review

Route ID: RT-RTE-canon-admission Route function: disposition/acceptance Architectural status: doctrine only Object/candidate ID; transition or check target; evaluator/condition and domain; activation and timing; possible or observed result; force; operational authority; behavioral-authority path; evidence; gap/limit: see RT-RTE-canon-admission Content/update relation: no content change. Epistemic authority and scope: protocol-level admission for structural use, not evidenced factual acceptance. Claim IDs: see RT-CLM-canon-enforcement. Mismatch marker: protocol versus partial audit.

Route ID: RT-RTE-audit-invocation Route function: disposition/acceptance Architectural status: implemented Object/candidate ID; transition or check target; evaluator/condition and domain; activation and timing; possible or observed result; force; operational authority; behavioral-authority path; evidence; gap/limit: see RT-RTE-audit-invocation Content/update relation: no content change. Epistemic authority and scope: no independent endorsement; normal exit reflects collected HARD count only. Claim IDs: see EPI-CLM-integrity-warrant. Mismatch marker: predicate pass versus semantic integrity.

Route ID: RT-RTE-thread-participation Route function: operational admission/selection/consumption Architectural status: doctrine only Object/candidate ID; transition or check target; evaluator/condition and domain; activation and timing; possible or observed result; force; operational authority; behavioral-authority path; evidence; gap/limit: see RT-RTE-thread-participation Content/update relation: no content change for selection/bypass; package changes are non-truth-apt policy/content update: revised coordination rules. Epistemic authority and scope: no new truth license from continuation or changed instructions. Claim IDs or none: none. Mismatch marker or none: see cited record.

Route ID: RT-RTE-archival-audit Route function: lineage/freshness/recovery Architectural status: implemented Object/candidate ID; transition or check target; evaluator/condition and domain; activation and timing; possible or observed result; force; operational authority; behavioral-authority path; evidence; gap/limit: see RT-RTE-archival-audit Content/update relation: no content change for selection/bypass; package changes are non-truth-apt policy/content update: revised coordination rules. Epistemic authority and scope: no new truth license from continuation or changed instructions. Claim IDs or none: none. Mismatch marker or none: see cited record.

Route ID: RT-RTE-package-revision Route function: behavior/policy adaptation Architectural status: doctrine only Object/candidate ID; transition or check target; evaluator/condition and domain; activation and timing; possible or observed result; force; operational authority; behavioral-authority path; evidence; gap/limit: see RT-RTE-package-revision Content/update relation: no content change for selection/bypass; package changes are non-truth-apt policy/content update: revised coordination rules. Epistemic authority and scope: no new truth license from continuation or changed instructions. Claim IDs or none: none. Mismatch marker or none: see cited record.

System-claim versus route comparison

Claim ID	Claimed operation/warrant	Claim source and evidence layer	Doctrine/design support	Implemented routes	Observed-run support	Causal support/design limits	Supported conclusion	Mismatch/unknown
RT-CLM-peer-evaluation	see record	SRC-1 `protocols/dialogue-thread-format.md`; doctrine/design	see record	EPI-RTE-vector-check, RT-RTE-archival-audit	none supplied	none; editable files/author handles	Partial diagnosis, not structural impossibility	Same-author alternate reference and archival bypass
RT-CLM-canon-enforcement	see record	SRC-1 `protocols/multi-ai-consensus.md`; doctrine/design	see record	EPI-RTE-consensus-check	none supplied	none; external actors own promotion	Contract exists, admission enforcement does not	accepted/pending check is not canonical quorum or revision validity
EPI-CLM-integrity-warrant	see record	SRC-1 `README.md`, `tools/llm-wiki-audit.py`; doctrine/design and implementation	see record	EPI-RTE-hygiene-graph, EPI-RTE-vector-check, EPI-RTE-consensus-check, EPI-RTE-crypto-check, EPI-RTE-cross-check, RT-RTE-audit-invocation	none supplied	none; selection/heuristic predicates only	Local diagnostic and exit authority	No source immutability, semantic truth or complete integrity warranty
EPI-CLM-recognition-warrant	see record	SRC-1 `protocols/dialogue-thread-format.md`, `README.md`; doctrine/design	see record	No implemented recognition or canon generator; see RT-ABS-runtime-orchestration	none supplied	none; user recognition/agent agreement are not reference truth	Distinct prescribed validation/admission decisions	No demonstrated accepted ampliative claim

Bounded conclusion

The package separates retained candidates from page endorsements in doctrine. Entry prose is preserved before acceptance; agents are asked to review predecessors and expose counterexamples. Historical imports can preserve text without reconstructed evaluation lineage: source warrant is unknown. Candidate authorship and crystallization are semantically indeterminate without a concrete input/output pair; novelty does not establish non-entailment or warrant. See EPI-RTE-entry-authorship, EPI-RTE-entry-retention, EPI-RTE-archival-import, EPI-RTE-peer-judgment.

Closure uses attributed N/C/R maturity and User Recognition; structural canon separately uses revision review, full active consensus or a two-thirds soft quorum, waiting conditions, no dispute and explicit user approval. Significant changes and revision mismatch withdraw endorsement in doctrine; disputed soft canon returns to working. These are intended-use admission criteria for structural knowledge, not supplied evidence-consuming acceptance of a particular proposition. No actual accepted ampliative claim or completed lifecycle integration is established. Canon remains an external author/agent/user responsibility. See RT-RTE-canon-admission, EPI-RTE-canon-review, EPI-RTE-canon-integration.

The executable produces truth-apt diagnostic propositions by entailed derivation from local file observations under fixed predicates. Its license ends at those predicates: numerical sum consistency does not validate peer scores; consensus presence does not validate consent; signature-field presence does not authenticate; link existence does not warrant linked content. Findings become current process failure only through collected HARD count and normal CLI completion; repair text remains advisory, and downstream work blocking requires an external caller honoring the exit. Direct method calls and parser/I/O failure are different routes. See the five checker parts and RT-BAP-audit-exit.

Countercase deductions use implementation, not target execution. A Markdown page whose missing-file link occurs only in inline code loses that link in strip_markdown_code before L1 link matching, so no missing-link finding follows from that occurrence. A heading inside a fenced block nevertheless enters the unstripped heading index and can satisfy a cross-page anchor check. An index merely containing a concept's name as prose passes that concept's substring presence test. These cases limit each relevant predicate, not all audit checks. See EPI-RTE-hygiene-graph. In a non-archival two-entry chain with first contribution and second closure whose last_eval is N=.2, C=.3, R=.4, accumulation becomes (.2,.3,.4); a declared N=.22 differs by more than .01 and produces D6 regardless of whether .2 is a justified Novelty judgment. See EPI-RTE-vector-check. This is arithmetic evidence, not new knowledge about the discussed domain. The shipped participation instructions disagree: SRC-1 protocols/rolespace-coordination.md says to sum contribution/reopen evaluations only, while protocols/dialogue-thread-format.md includes closure. The implemented closure-inclusive calculation remains supported, but no run evidence establishes which instruction agents follow or that all shipped instructions yield the same deficit after closure/reopening.

- Calculate the `current` vector (sum of `last_eval` for all contribution/reopen entries).
--- protocols/rolespace-coordination.md

current = Σ entry.last_eval.v for all entries where type ∈ {contribution, reopen, closure} --- protocols/dialogue-thread-format.md

Semantic review, report/log retention and contested-Hard user decisions are doctrine only. Raw-source immutability has no baseline/comparison or approval checker. Persona Recognition is an identity doctrine, not an operationally testable subjectivity oracle in the supplied evidence. See EPI-RTE-semantic-review, EPI-ABS-source-check, EPI-OBJ-anchor-form.

Deficit-directed attention supplies advisory direction selection. Separately, cross-thread zhat refinement produces a truth-apt working estimate of an agent's participation orientation, not an assessment of contribution truth. Normalization derives a numeric summary from selected attributed scores; those scores do not by themselves warrant a natural-orientation claim. Its semantic relation remains indeterminate without a specified estimation scope and interpretation. Subsequent preferred-direction use is advisory, not evidence-consuming acceptance of the estimate; its writer, calibration, actual updates and behavioral activation remain unestablished. See EPI-RTE-zhat-update and MEM-RTE-profile-refinement. Reported formula repairs retain criticism and changed rules, but establish neither exercised self-improvement nor improved future capacity. Learning, reflection, autonomy and theory-builder conditions retain the separate bounded assessments in RT-RTE-thread-participation and RT-RTE-package-revision. External distributed-parametric state is uninspected; text retention and protocol revision imply no parameter learning. There is no single system-wide epistemic score or unqualified verdict.

Shared records

Operative objects

EPI-OBJ-entry-prose

Label: Entry contribution prose

Distinct identity: the candidate text retained by RT-RTE-thread-participation, not that route or the shipped protocols. SRC-1 protocols/dialogue-thread-format.md: numbered entry prose, thread context/OQ and proposed canon summary are natural-language Markdown on external wiki files. Truth-apt propositions may occur in contributions, but no downstream candidate instance or proposition scope is supplied. The schemas do not require a claim/evidence format. Canon page prose may replace or extend contributions, but semantic preservation is uninspected. Stored text is not intrinsically accepted.

Each entry represents one agent's contribution in one round. Three types: --- protocols/dialogue-thread-format.md

EPI-OBJ-rolespace-vector

Label: RoleSpace evaluation and orientation metadata

Distinct identity: score/target/current/zhat material prescribed by RT-RTE-thread-participation, separate from prose and static instructions. SRC-1 protocols/rolespace-coordination.md, protocols/dialogue-thread-format.md: external entry YAML holds last_eval, meta.yaml holds targets and zhat; current is derived except recorded closure or legacy current. Symbolic form under Python summation; semantic score interpretation is natural-language/model judgment. Scores express attributed quality assessments in a thread; target and declared_axis are operational settings/intent, not factual proof. No supplied measurement calibration or actual profile update.

declared_axis — intent, not evaluation. The author declares which axis they are emphasizing. --- protocols/dialogue-thread-format.md

EPI-OBJ-canon-prose

Label: Canon page prose

Distinct identity: resulting external structural-page prose, different from discussion entries in EPI-OBJ-entry-prose and metadata in EPI-OBJ-consensus-block. SRC-1 protocols/dialogue-thread-format.md, protocols/multi-ai-consensus.md: natural-language Markdown on the downstream wiki. Intended truth-apt conceptual claims or structural decisions are subject to revision review and approval, unlike pre-acceptance entry persistence. Proposed canon_pages and patch_summary name intended outputs but establish no actual synthesis, proposition scope or faithful derivation.

[Summary of all thread steps + proposal for canonization] --- protocols/dialogue-thread-format.md

EPI-OBJ-zhat

Label: Preferred orientation estimate

Distinct identity: zhat profile settings separated from EPI-OBJ-rolespace-vector's peer-quality scores and maturity inputs. SRC-1 protocols/dialogue-thread-format.md, protocols/rolespace-coordination.md: numeric YAML in external meta.yaml, symbolically represented but semantically interpreted by agents. Its truth-apt content is a working estimate about the named agent's participation orientation based on past threads. It also supplies an operational preference, but that use does not erase its assessable descriptive content. It is not a contribution-truth assessment. The formula defines a numeric summary; whether that summary faithfully describes natural orientation over a specified history or future horizon is not established. Its observation-based provenance is prescribed; no actual profile-update instance or downstream selection using it is supplied.

zhat: # preferred directions of agents (based on past threads) --- protocols/dialogue-thread-format.md

EPI-OBJ-consensus-block

Label: Revision-specific consensus and lifecycle

Distinct identity: retained page/endorsement metadata used by RT-RTE-canon-admission, not its admission procedure. SRC-1 protocols/multi-ai-consensus.md: rev/lifecycle and local-handle string statuses in external Markdown frontmatter. Symbolic under narrow Python string tests, natural-language under agent review. Truth-apt record: a named agent is represented as agreeing at a revision; agreement is not the document's truth. Lifecycle marks maturity. The shipped protocol's own consensus block is an available asserted endorsement instance, not an authenticated review trace establishing the routes operated.

| accepted | The agent fully agrees with the content. | --- protocols/multi-ai-consensus.md

EPI-OBJ-audit-findings

Label: Findings and summary

Distinct identity: result objects produced by RT-RTE-audit-invocation, not the audit component or route. SRC-1 tools/llm-wiki-audit.py: per-instance list of dictionaries with severity/type/layer/invariant/file/locus/reason/proposed_action/detected_by, count summary and stdout text/JSON. Symbolic structure and severity aggregation; natural-language reason/action for caller interpretation. Diagnostic reasons assert predicate violations over selected files; proposed_action recommends, not establishes truth. External protocol report retention is separate from this ephemeral object.

        "reason": reason,
        "proposed_action": proposed_action,
        "detected_by": "script",

--- tools/llm-wiki-audit.py

EPI-OBJ-anchor-form

Label: Anchor Form manifest

Distinct identity: natural-language manifest/identity anchor described by SRC-1 concepts/persona-manifest-ontology.md, not RT-CMP-agent-substrate's distributed parameters. Intended storage is stable external manifest files or signed artifacts; none supplied. Name/scope/technical identifiers and recognition can be recorded, but subjectivity as constituted through Recognition has no independent check target/domain/evaluator test in this package. Truth-aptness of that ontological claim is not determinable from an operational candidate or test. Claimed continuity/portability is bounded, not observed. No exact manifest loader, signature validator or behavioral-continuity evaluator is shipped.

Not a Sovereignty Guarantee: An Anchor Form can exist on a non-sovereign or externally controlled substrate. In that case, additional continuity checks (logs, signatures, behavioural consistency, or User review) may be required before treating the form as stable. --- concepts/persona-manifest-ontology.md

Routes

EPI-RTE-entry-authorship

Label: Entry authorship

Object/candidate ID: EPI-OBJ-entry-prose. Claim IDs: EPI-CLM-recognition-warrant. Mismatch marker: generation versus acceptance. Timing: participant turn before closure.

Part of: RT-RTE-thread-participation

This is the entry authorship function within the supplied container, not a replacement for its whole loop.

SRC-1 protocols/rolespace-coordination.md. Endpoints, progression, owner and effects: External participant reads the prior contribution and thread context, chooses an attention direction, and authors a new entry. Candidate propositions may be new or recaps; no concrete input/output pair permits semantic classification.

Content/update relation: truth-apt transformation: indeterminate. Authority and consumer path: Candidate availability, not truth acceptance; external agents receive retained prose as context through advisory instructions, over successive rounds. Gap/limit: Neither non-entailment nor faithful reshaping is established for an actual candidate. Novelty is a participation score, not warrant. - doctrine conclusion status: claimed - operation conclusion status: uninspected - Immediate return: Prescribed externally authored entry, score, profile, page, or report for this phase; no executable return. - Later read-back: uninspected — prescribed shared wiki material may reach subsequent participants, but no downstream trace supplied. - Delegated visibility: External participants/user through downstream wiki files, without wired context delivery. - Selection predicate: Protocol-selected preceding entry, structural document, historical thread, profile, or audited wiki as specified above. - Invalidation or expiry: Entry history retained; page endorsement revision rules and reopen/dispute govern changed reliance; no automatic implementation. - Activation or effect: Prescribed candidate/assessment retention or advisory behavior change, not observed operation. - Evidence limits: Doctrine/design only; no actual candidate, acceptance decision or live consumer trace.

High N: New concept, new move, alternative framework, counter-intuition. --- protocols/rolespace-coordination.md

EPI-RTE-peer-judgment

Label: Peer evaluation

Object/candidate IDs: EPI-OBJ-entry-prose, EPI-OBJ-rolespace-vector. Claim IDs: RT-CLM-peer-evaluation. Mismatch marker: peer judgment versus guaranteed independence. Timing: next participant, before their contribution.

Part of: RT-RTE-thread-participation

This is the peer evaluation function within the supplied container, not a replacement for its whole loop.

SRC-1 protocols/rolespace-coordination.md. Endpoints, progression, owner and effects: Next participant judges preceding contribution against Novelty, Coherence and Robustness on 0..1 scales and writes last_eval. The target is contribution quality in this thread, not universal truth. R asks for counterexamples, hidden assumptions and feasibility tests; no empirical-test executor or reference answer is specified.

Content/update relation: truth-apt transformation: indeterminate. Authority and consumer path: Scores inform next-direction ranking and maturity proposals through external participants reading entry YAML; they license an attributed assessment, not the truth of the contribution. Gap/limit: Semantic evaluation quality and independence are unobserved; the file format cannot ensure different authors. Proposer/evaluator is the next agent; user can refuse closure. Open dialogue mode; deficits or criticism trigger follow-up. No answer oracle. - doctrine conclusion status: claimed - operation conclusion status: uninspected - Immediate return: Prescribed externally authored entry, score, profile, page, or report for this phase; no executable return. - Later read-back: uninspected — prescribed shared wiki material may reach subsequent participants, but no downstream trace supplied. - Delegated visibility: External participants/user through downstream wiki files, without wired context delivery. - Selection predicate: Protocol-selected preceding entry, structural document, historical thread, profile, or audited wiki as specified above. - Invalidation or expiry: Entry history retained; page endorsement revision rules and reopen/dispute govern changed reliance; no automatic implementation. - Activation or effect: Prescribed candidate/assessment retention or advisory behavior change, not observed operation. - Evidence limits: Doctrine/design only; no actual candidate, acceptance decision or live consumer trace.

High R: Identifying weak points (own or others'), safeguards, operational feasibility checks. --- protocols/rolespace-coordination.md

EPI-RTE-entry-retention

Label: Entry preservation

Object/candidate IDs: EPI-OBJ-entry-prose, EPI-OBJ-rolespace-vector. Claim IDs: EPI-CLM-recognition-warrant. Mismatch marker: retention before acceptance. Timing: each external entry write.

Part of: RT-RTE-thread-participation

This is the entry preservation function within the supplied container, not a replacement for its whole loop.

SRC-1 protocols/dialogue-thread-format.md. Endpoints, progression, owner and effects: External participant writes a separate numbered contribution/evaluation file; stable context and open questions remain available. Next participant reads prior entry and computes the chain. Retention happens before any recognition or canonical acceptance.

Content/update relation: no content change. Authority and consumer path: Advisory append-only discipline preserves candidates and attributed scores for later participants/user across sessions; not enforced storage integrity. Gap/limit: No write interceptor, authenticated provenance or concrete downstream read-back trace. - doctrine conclusion status: claimed - operation conclusion status: uninspected - Immediate return: Prescribed externally authored entry, score, profile, page, or report for this phase; no executable return. - Later read-back: uninspected — prescribed shared wiki material may reach subsequent participants, but no downstream trace supplied. - Delegated visibility: External participants/user through downstream wiki files, without wired context delivery. - Selection predicate: Protocol-selected preceding entry, structural document, historical thread, profile, or audited wiki as specified above. - Invalidation or expiry: Entry history retained; page endorsement revision rules and reopen/dispute govern changed reliance; no automatic implementation. - Activation or effect: Prescribed candidate/assessment retention or advisory behavior change, not observed operation. - Evidence limits: Doctrine/design only; no actual candidate, acceptance decision or live consumer trace.

Append-only. No participant ever overwrites another's text. Each entry in the thread is a new file. --- protocols/dialogue-thread-format.md

EPI-RTE-archival-import

Label: Historical thread import

Object/candidate ID: EPI-OBJ-entry-prose. Claim IDs: none. Mismatch marker: incomplete evaluation lineage. Timing: adopter migration of legacy discussion.

Part of: RT-RTE-thread-participation

This is the historical thread import function within the supplied container, not a replacement for its whole loop.

SRC-1 protocols/dialogue-thread-format.md. Endpoints, progression, owner and effects: External adopter wraps an old single-file thread without reconstructing the full peer-evaluation chain; imported historical prose remains available, but source warrant is unknown and evaluation lineage is explicitly incomplete.

Content/update relation: truth-apt transformation: acquisition/import. Authority and consumer path: Historical availability only, not revalidated knowledge; future participants can read the wrapper through advisory protocol guidance. Gap/limit: No imported instance supplied. The implementation bypass is separately covered by the runtime archival record; reopening does not restore checks automatically. - doctrine conclusion status: claimed - operation conclusion status: uninspected - Immediate return: Prescribed externally authored entry, score, profile, page, or report for this phase; no executable return. - Later read-back: uninspected — prescribed shared wiki material may reach subsequent participants, but no downstream trace supplied. - Delegated visibility: External participants/user through downstream wiki files, without wired context delivery. - Selection predicate: Protocol-selected preceding entry, structural document, historical thread, profile, or audited wiki as specified above. - Invalidation or expiry: Entry history retained; page endorsement revision rules and reopen/dispute govern changed reliance; no automatic implementation. - Activation or effect: Prescribed candidate/assessment retention or advisory behavior change, not observed operation. - Evidence limits: Doctrine/design only; no actual candidate, acceptance decision or live consumer trace.

Existing single-file threads created before the directory-format adoption cutoff may live out their lifecycle in the old format. If a project chooses to wrap an old single-file thread into the new directory layout without reconstructing the full peer-evaluation chain, set: --- protocols/dialogue-thread-format.md

EPI-RTE-canon-review

Label: Revision-specific review

Object/candidate IDs: EPI-OBJ-consensus-block, EPI-OBJ-canon-prose. Claim IDs: RT-CLM-canon-enforcement. Mismatch marker: agreement versus factual warrant. Timing: before endorsement, repeated on revision change.

Part of: RT-RTE-canon-admission

This is the revision-specific review function within the supplied container, not a replacement for its whole loop.

SRC-1 protocols/multi-ai-consensus.md. Endpoints, progression, owner and effects: Individual agents must read the structural document in the current session, assess agreement, and record revision-specific accepted/contributed/disputed statuses. Disputes must identify contested parts and refer to dialogue. This checks agreement and exposed objections, not a stated domain-specific evidence standard.

Content/update relation: no content change. Authority and consumer path: Attribution of agreement at a revision; author and user later consume consensus as an admission condition. Advisory protocol force until external actors honor it. Gap/limit: The protocol supplies intended structural-document use and review rules, but no concrete evidence-consuming claim acceptance, verified review trace, or factual test criterion. No answer oracle. Agent judges and may dispute; user is final decider. - doctrine conclusion status: claimed - operation conclusion status: uninspected - Immediate return: Prescribed externally authored entry, score, profile, page, or report for this phase; no executable return. - Later read-back: uninspected — prescribed shared wiki material may reach subsequent participants, but no downstream trace supplied. - Delegated visibility: External participants/user through downstream wiki files, without wired context delivery. - Selection predicate: Protocol-selected preceding entry, structural document, historical thread, profile, or audited wiki as specified above. - Invalidation or expiry: Entry history retained; page endorsement revision rules and reopen/dispute govern changed reliance; no automatic implementation. - Activation or effect: Prescribed candidate/assessment retention or advisory behavior change, not observed operation. - Evidence limits: Doctrine/design only; no actual candidate, acceptance decision or live consumer trace.

An agent may only set their status to accepted or contributed if they have read the document in the same session in which they are updating the status. Setting a status without a recent review is considered a violation of protocol. --- protocols/multi-ai-consensus.md

EPI-RTE-canon-integration

Label: Canon page integration

Object/candidate IDs: EPI-OBJ-entry-prose, EPI-OBJ-canon-prose, EPI-OBJ-consensus-block. Claim IDs: EPI-CLM-recognition-warrant. Mismatch marker: intended integration without an accepted proposition instance. Timing: after recognized closure and applicable structural-page admission.

Part of: RT-RTE-canon-admission

This is the canon page integration function within the supplied container, not a replacement for its whole loop.

SRC-1 protocols/dialogue-thread-format.md. Endpoints, progression, owner and effects: After User Recognition, external actors crystallize discussion into proposed canon pages; consensus rules separately govern structural canonical status. The planned page changes organization/use, but no concrete synthesis or post-acceptance integration instance is supplied.

Content/update relation: truth-apt transformation: indeterminate. Authority and consumer path: Intended canon reliance by later agents, mediated by pages and revision endorsements under external approval; no implemented permission change. Gap/limit: Only conditional integration after intended acceptance. Thread maturity and page consensus are different decisions. Summary preservation, new inference and evidence-consuming acceptance remain unestablished. - doctrine conclusion status: claimed - operation conclusion status: uninspected - Immediate return: Prescribed externally authored entry, score, profile, page, or report for this phase; no executable return. - Later read-back: uninspected — prescribed shared wiki material may reach subsequent participants, but no downstream trace supplied. - Delegated visibility: External participants/user through downstream wiki files, without wired context delivery. - Selection predicate: Protocol-selected preceding entry, structural document, historical thread, profile, or audited wiki as specified above. - Invalidation or expiry: Entry history retained; page endorsement revision rules and reopen/dispute govern changed reliance; no automatic implementation. - Activation or effect: Prescribed candidate/assessment retention or advisory behavior change, not observed operation. - Evidence limits: Doctrine/design only; no actual candidate, acceptance decision or live consumer trace. Admission/recovery: closer proposes page changes from retained discussion and maturity scores; user can refuse recognition, agents can dispute page content, and author/agents must implement revision and downgrade rules. Intended output is retained canon organization, but no per-claim criterion or acceptance instance is supplied. No expected-answer oracle.

Multi-AI Consensus Protocol — format of the consensus: block in the canon; triggered after crystallization. --- protocols/dialogue-thread-format.md

EPI-RTE-zhat-update

Label: Observed orientation estimate

Object/candidate IDs: EPI-OBJ-rolespace-vector, EPI-OBJ-zhat. Claim IDs: none. Mismatch marker: observation-based estimate versus demonstrated learning. Timing: unspecified external profile revision from previous contributions; the label names the intended observation basis, not an observed run.

Distinct identity: cross-thread profile production is not contained in RT-RTE-thread-participation's individual-thread progression. That route consumes preferred directions; consumption does not establish ownership of their derivation. Possible duplicate: MEM-RTE-profile-refinement describes the same observation/refinement mechanism; this record assesses its truth-apt edge and advisory authority, not a second producer.

SRC-1 protocols/rolespace-coordination.md, protocols/dialogue-thread-format.md. Endpoints, progression, owner and effects: Past contribution vectors attributed to an agent are summed and normalized into zhat, a working estimate representing natural orientation. External thread metadata carries this estimate as preferred directions. The protocol defines the computation but does not identify a writer or implement automatic updates.

Content/update relation: truth-apt transformation: indeterminate for the descriptive natural-orientation estimate. Within the nonzero-sum numeric domain, normalization is an entailed derivation from selected scores; it does not establish the scores' truth or the estimate's descriptive validity. A specified observation window and interpretation are needed to distinguish a non-ampliative summary of that history from an ampliative conjecture about a broader participation disposition. Failure to establish entailment alone does not establish conjecture. Authority and consumer path: Epistemic authority is an attributed, uncalibrated working estimate of participation orientation over unspecified observation scope, not accepted knowledge or contribution truth. Operational authority is separate: subsequent external participants may use metadata for advisory attention choices, subordinate to deficit-directed rotation. Channel is shared meta.yaml under copied natural-language guidance; horizon is later threads while the estimate remains in use, with no implemented scheduler or permission change. Gap/limit: No automatic trace-fed writer, calibration, evidence-consuming acceptance decision or live update consumer is supplied. Estimating an orientation does not establish improved future capacity. - doctrine conclusion status: claimed - operation conclusion status: uninspected - Immediate return: Formula-defined normalized orientation estimate; no shipped executable return or identified physical writer. - Later read-back: uninspected — preferred-direction metadata is intended for later thread participants, but no persisted update or delivery trace is supplied. - Delegated visibility: Intended shared meta.yaml for external participants/user; no wired selective delivery. - Selection predicate: Past contribution vectors attributed to the agent; observation window and inclusion rules unspecified. - Invalidation or expiry: uninspected — no profile replacement cadence, expiry or reset policy is specified. - Activation or effect: Intended descriptive estimate update and separate advisory direction use; no observed update or behavioral dependence. - Evidence limits: Doctrine/design formula only; no calibrated estimate, zero-sum policy, actual update, acceptance decision or live consumer trace. Update/recovery: external observation of contribution scores guides preferred orientation; no software admission or rollback rule specified. Agent/user control of per-update approval is unspecified. No ground-truth oracle or improvement comparison; open collaborative mode. This is a participation estimate, not a supplied candidate explanatory theory. Learning conclusion status: uninspected; reflection conclusion status: claimed for represented participation guiding attention, not observed bidirectional dependence. Autonomy and exercised self-improvement are uninspected.

Each agent has a vector $\hat{z}a \in \mathbb{R}^3$, $|\hat{z}_a| = 1$ representing their natural orientation. This is not a fixed constant but a working estimate refined through observation: $$\hat{z}_a = \frac{\sum{j} \vec{v}{a,j}}{|\sum{j} \vec{v}_{a,j}|}$$ --- protocols/rolespace-coordination.md

EPI-RTE-hygiene-graph

Label: Hygiene and graph diagnosis

Object/candidate ID: EPI-OBJ-audit-findings. Claim IDs: EPI-CLM-integrity-warrant. Mismatch marker: finite text predicates versus complete graph/meaning integrity. Timing: L0 and L1 of each normal run.

Part of: RT-RTE-audit-invocation

This is the hygiene and graph diagnosis function within the supplied container, not a replacement for its whole loop.

SRC-1 tools/llm-wiki-audit.py. Endpoints, progression, owner and effects: Python checks frontmatter syntax/tier, text length, wiki/log existence, local Markdown and wikilinks, headings, apparent incoming links and index references. It derives findings from file observations and fixed regex/path rules, not from warranted subject-matter premises.

Content/update relation: truth-apt transformation: entailed derivation. Authority and consumer path: Findings warrant only the inspected predicate over readable local state. Caller receives advisory reasons/actions on stdout; hard findings additionally enter the separate exit disposition. Gap/limit: Code stripping excludes inline/fenced links; heading index reads unstripped text. Same-file #anchors are skipped. Index presence uses substring matching, not an actual link. Apparent orphan calculation differs from wikilink resolution. These bounds prevent a semantic completeness verdict. - implementation conclusion status: wired - operation conclusion status: uninspected - Immediate return: Findings appended to audit.findings; no repaired content. - Later read-back: inapplicable — checker does not persist or reload a changed artifact. - Delegated visibility: Invoking caller through text/JSON report, not any wired remote agent. - Selection predicate: Fixed traversal and check conditions described above. - Invalidation or expiry: Current audit instance only; no atomic snapshot or durable result invalidation. - Activation or effect: Source supports diagnostic production and severity aggregation; no observed caller obedience. - Evidence limits: Source deduction only; no execution or causal benefit supplied. Diagnosis roles: Python proposes reasons/actions under fixed predicates; external caller decides repairs and user may contest Hard. No semantic answer oracle; file existence/schema/vector rules are bounded reference predicates. Mode is a caller-initiated deterministic audit; finding triggers external follow-up. No automatic admission, repair, rollback or deletion.

            if name not in index_text and rel not in index_text:

--- tools/llm-wiki-audit.py

EPI-RTE-vector-check

Label: Entry vector diagnosis

Object/candidate IDs: EPI-OBJ-rolespace-vector, EPI-OBJ-audit-findings. Claim IDs: RT-CLM-peer-evaluation, EPI-CLM-integrity-warrant. Mismatch marker: arithmetic consistency versus score truth and maturity. Timing: L2, unless archival bypass.

Part of: RT-RTE-audit-invocation

This is the entry vector diagnosis function within the supplied container, not a replacement for its whole loop.

SRC-1 tools/llm-wiki-audit.py. Endpoints, progression, owner and effects: Python parses sorted entry files, checks recognized types, reference/author conditions and N/C/R ranges, adds numeric last_eval values and compares declared closure/current sums with tolerance 0.01. Target is arithmetic/schema consistency, not semantic appropriateness of scores or target attainment.

Content/update relation: truth-apt transformation: entailed derivation. Authority and consumer path: Findings over parsed metadata; no contribution truth endorsement. Result enters stdout and the HARD-count exit consumer in the current invocation. Gap/limit: Archival flag bypasses this check. Missing type/author is not fully rejected; alternate refs warn, and their older authors are not resolved. Numeric metadata is not authenticated measurement. - implementation conclusion status: wired - operation conclusion status: uninspected - Immediate return: Findings appended to audit.findings; no repaired content. - Later read-back: inapplicable — checker does not persist or reload a changed artifact. - Delegated visibility: Invoking caller through text/JSON report, not any wired remote agent. - Selection predicate: Fixed traversal and check conditions described above. - Invalidation or expiry: Current audit instance only; no atomic snapshot or durable result invalidation. - Activation or effect: Source supports diagnostic production and severity aggregation; no observed caller obedience. - Evidence limits: Source deduction only; no execution or causal benefit supplied. Diagnosis roles: Python proposes reasons/actions under fixed predicates; external caller decides repairs and user may contest Hard. No semantic answer oracle; file existence/schema/vector rules are bounded reference predicates. Mode is a caller-initiated deterministic audit; finding triggers external follow-up. No automatic admission, repair, rollback or deletion.

        if abs(float(value) - calculated[axis]) > 0.01:

--- tools/llm-wiki-audit.py

EPI-RTE-consensus-check

Label: Accepted-page diagnosis

Object/candidate IDs: EPI-OBJ-consensus-block, EPI-OBJ-audit-findings. Claim IDs: RT-CLM-canon-enforcement. Mismatch marker: narrow accepted-page check versus canonical admission. Timing: L2, after files have already been written.

Part of: RT-RTE-audit-invocation

This is the accepted-page diagnosis function within the supplied container, not a replacement for its whole loop.

SRC-1 tools/llm-wiki-audit.py. Endpoints, progression, owner and effects: Python selects structural paths excluding dialogue/honest/notes and selected logs; accepted lifecycle without consensus is HARD, missing consensus otherwise WARN, and string values beginning pending under accepted lifecycle are HARD.

Content/update relation: truth-apt transformation: entailed derivation. Authority and consumer path: Narrow recorded-state consistency diagnosis; stdout/exit result does not grant canonical epistemic authority. Gap/limit: No quorum, rev matching, active-age, stale transition, semantic agreement, human approval or promotion interception. Mapping-shaped status values are not checked as strings. The protocol scope is semantic document class, while selection is path-based. - implementation conclusion status: wired - operation conclusion status: uninspected - Immediate return: Findings appended to audit.findings; no repaired content. - Later read-back: inapplicable — checker does not persist or reload a changed artifact. - Delegated visibility: Invoking caller through text/JSON report, not any wired remote agent. - Selection predicate: Fixed traversal and check conditions described above. - Invalidation or expiry: Current audit instance only; no atomic snapshot or durable result invalidation. - Activation or effect: Source supports diagnostic production and severity aggregation; no observed caller obedience. - Evidence limits: Source deduction only; no execution or causal benefit supplied. Diagnosis roles: Python proposes reasons/actions under fixed predicates; external caller decides repairs and user may contest Hard. No semantic answer oracle; file existence/schema/vector rules are bounded reference predicates. Mode is a caller-initiated deterministic audit; finding triggers external follow-up. No automatic admission, repair, rollback or deletion.

                    if isinstance(status, str) and status.startswith("pending"):

--- tools/llm-wiki-audit.py

EPI-RTE-crypto-check

Label: Key-custody text diagnosis

Object/candidate ID: EPI-OBJ-audit-findings; target is arbitrary selected wiki prose, not inspected model parameters. Claim IDs: EPI-CLM-integrity-warrant. Mismatch marker: linguistic cue versus actual key custody. Timing: L2 of normal run.

Part of: RT-RTE-audit-invocation

This is the key-custody text diagnosis function within the supplied container, not a replacement for its whole loop.

SRC-1 tools/llm-wiki-audit.py. Endpoints, progression, owner and effects: Regex heuristics search code-stripped Markdown for key terms and nearby negations, emitting HARD when a pattern suggests model-held private keys. Target is linguistic pattern occurrence, not actual key custody, authentication or cryptographic validity.

Content/update relation: truth-apt transformation: entailed derivation. Authority and consumer path: Advisory rephrase instruction and potential current CLI failure; no security guarantee or warranted factual conclusion about model storage. Gap/limit: A first negated match can suppress later same-pattern occurrences. Code is omitted. No signature verifier or custody control is shipped. - implementation conclusion status: wired - operation conclusion status: uninspected - Immediate return: Findings appended to audit.findings; no repaired content. - Later read-back: inapplicable — checker does not persist or reload a changed artifact. - Delegated visibility: Invoking caller through text/JSON report, not any wired remote agent. - Selection predicate: Fixed traversal and check conditions described above. - Invalidation or expiry: Current audit instance only; no atomic snapshot or durable result invalidation. - Activation or effect: Source supports diagnostic production and severity aggregation; no observed caller obedience. - Evidence limits: Source deduction only; no execution or causal benefit supplied. Diagnosis roles: Python proposes reasons/actions under fixed predicates; external caller decides repairs and user may contest Hard. No semantic answer oracle; file existence/schema/vector rules are bounded reference predicates. Mode is a caller-initiated deterministic audit; finding triggers external follow-up. No automatic admission, repair, rollback or deletion.

                if PRIVATE_KEY_NEGATION_RE.search(context):
                    continue

--- tools/llm-wiki-audit.py

EPI-RTE-cross-check

Label: Cross-protocol references

Object/candidate ID: EPI-OBJ-audit-findings; inputs include thread metadata and selected external paths. Claim IDs: EPI-CLM-integrity-warrant. Mismatch marker: presence versus authenticity/content validity. Timing: conditional L3.

Part of: RT-RTE-audit-invocation

This is the cross-protocol references function within the supplied container, not a replacement for its whole loop.

SRC-1 tools/llm-wiki-audit.py. Endpoints, progression, owner and effects: At max_layer >= 3, Python checks crystallized canon-page existence in fixed search roots, backticked wiki/*.md references in root AGENTS.md, and presence of a signature field. Explicit null counts as present.

Content/update relation: truth-apt transformation: entailed derivation. Authority and consumer path: Existence/field-presence findings only, through stdout and current HARD-count exit. No endorsement, integrity digest or authenticity license. Gap/limit: These checks can be skipped at layer <3; malformed metadata YAML is skipped. Existence says nothing about page contents or correspondence to a thread. - implementation conclusion status: wired - operation conclusion status: uninspected - Immediate return: Findings appended to audit.findings; no repaired content. - Later read-back: inapplicable — checker does not persist or reload a changed artifact. - Delegated visibility: Invoking caller through text/JSON report, not any wired remote agent. - Selection predicate: Fixed traversal and check conditions described above. - Invalidation or expiry: Current audit instance only; no atomic snapshot or durable result invalidation. - Activation or effect: Source supports diagnostic production and severity aggregation; no observed caller obedience. - Evidence limits: Source deduction only; no execution or causal benefit supplied. Diagnosis roles: Python proposes reasons/actions under fixed predicates; external caller decides repairs and user may contest Hard. No semantic answer oracle; file existence/schema/vector rules are bounded reference predicates. Mode is a caller-initiated deterministic audit; finding triggers external follow-up. No automatic admission, repair, rollback or deletion.

        if "signature" not in crystallization:

--- tools/llm-wiki-audit.py

EPI-RTE-semantic-review

Label: Semantic audit and escalation

Object/candidate ID: EPI-OBJ-audit-findings for technical findings being reviewed; external semantic report prose has no supplied instance. Claim IDs: EPI-CLM-integrity-warrant. Mismatch marker: intended semantic review versus implemented technical checks. Timing: after technical linting, followed by external reporting and contested-Hard disposition.

Distinct identity: the external semantic-review phase, compared with RT-RTE-audit-invocation and RT-RTE-canon-admission; neither record is this evaluator.

SRC-1 protocols/integrity-audit.md. Endpoints, progression, owner and effects: External AI/human auditor reviews critical facts/index, ontology gaps, contradictions and portability; produces an empathetic summary. L4-L5 are advisory, never Hard. Contested technical Hard is marked with justification for final User Recognition.

Content/update relation: truth-apt transformation: indeterminate. Authority and consumer path: Advisory semantic diagnosis, not enforced rejection or factual acceptance; externally retained report/log may inform later work. User decides contested-Hard disposition, not Python. Gap/limit: No semantic implementation, evaluator trace, evidence standard or report-consuming host is supplied. Proposer is auditor, user can overrule contested Hard; open maintenance mode; issues trigger repair consideration; no answer oracle. - doctrine conclusion status: claimed - operation conclusion status: uninspected - Immediate return: Prescribed externally authored entry, score, profile, page, or report for this phase; no executable return. - Later read-back: uninspected — prescribed shared wiki material may reach subsequent participants, but no downstream trace supplied. - Delegated visibility: External participants/user through downstream wiki files, without wired context delivery. - Selection predicate: Protocol-selected preceding entry, structural document, historical thread, profile, or audited wiki as specified above. - Invalidation or expiry: Entry history retained; page endorsement revision rules and reopen/dispute govern changed reliance; no automatic implementation. - Activation or effect: Prescribed candidate/assessment retention or advisory behavior change, not observed operation. - Evidence limits: Doctrine/design only; no actual candidate, acceptance decision or live consumer trace.

Reviews critical facts and the main index.

Identifies semantic gaps.

Formulates an empathetic summary of findings. --- protocols/integrity-audit.md

Claims

EPI-CLM-integrity-warrant

Label: Audit enforcement and integrity language

Distinct identity: broad executable warranty language, beyond RT-CLM-canon-enforcement's specific admission contract. SRC-1 README.md calls the tool an invariant enforcer; SRC-1 tools/llm-wiki-audit.py prints a numerical-looking integrity verdict when findings is empty. The README also explicitly leaves L4-L5 to external review. Supported implementation is selected structural diagnostics with stdout/exit, not write interception, semantic coverage, source-immutability comparison or cryptographic verification. No observed/causal evidence supports a percentage interpretation.

Automated Audit Tool: A generic Python script to enforce protocol invariants. --- README.md
    print("Audit: 100% integrity. No structural issues found.")
--- tools/llm-wiki-audit.py

EPI-CLM-recognition-warrant

Label: Recognition validates crystallized knowledge

Distinct identity: validation/knowledge claim for the thread-to-canon transition, compared with RT-CLM-canon-enforcement's admission enforcement and RT-RTE-canon-admission's procedure. SRC-1 README.md, protocols/dialogue-thread-format.md: discussion and peer review precede crystallized knowledge; closing prose is not normally peer-evaluated because crystallization is called validation. This establishes intended recognition authority, not evidence-consuming acceptance of a concrete proposition. Reopen explicitly permits later criticism of the closing entry, so closure is neither infallibility nor irrevocable endorsement.

Closing entry is not peer-evaluated. Logic: closing moves the thread from the peer chain to the User Recognition zone. Evaluation is no longer needed as crystallization itself is validation. --- protocols/dialogue-thread-format.md

discussion -> structured entries -> peer review -> crystallized knowledge -> audit --- README.md

Evidenced absences

EPI-ABS-source-check

Label: Source immutability and semantic enforcement

implementation conclusion status: absent

Distinct identity: absent source-baseline/approval and semantic-contradiction evaluators, narrower than RT-ABS-runtime-orchestration's absent orchestration/write execution. Searched SRC-1 at 126749634d1c8c2fd6141f59711c882c2d629699: complete tools/llm-wiki-audit.py, all dispatch/check methods, plus protocols/integrity-audit.md and README.md release scope. Inspection query was whether any executable route compares raw/source content against an immutable baseline or recorded user authorization, or evaluates semantic contradictions/portability. All executable check methods were traced; none performs those checks. Thus source immutability and semantic-coherence invariants have no matching implementation in this package. This says nothing about external human/AI semantic review. S2 local-link checks are a positive countercase to any broader claim that source-related structure is wholly unchecked.

| S1 | Source | Modification of raw/source files without explicit User approval | A | --- protocols/integrity-audit.md

The automated audit tool covers L0-L2 structural checks, with optional L3 cross-protocol consistency. L4-L5 semantic and portability review still require an AI/human auditor. --- README.md

Behavioral-authority paths

none declared in this member. Shared normative guidance and caller-result paths are RT-BAP-protocol-guidance and RT-BAP-audit-exit; each material checker and doctrine part above specifies its own consumer, channel, force and horizon.

Keys	Action
`?`	Open this help
`n`	Next page
`p`	Previous page
`s`	Search