Proposal 051: Swarm Membership, Reputation Bootstrap, and Public Adjudication¶

Based on:

doc/normative/30-core-values/CORE-VALUES.md
doc/project/50-requirements/requirements-001-node-onboarding.md
doc/project/40-proposals/015-nym-certificates-and-renewal-baseline.md
doc/project/20-memos/nym-layer-roadmap-and-revocable-anonymity.md
doc/project/20-memos/reputation-signal-v1-invariants.md
doc/project/40-proposals/026-resource-opinions-and-discussion-surfaces.md
doc/project/40-proposals/080-multiplexed-middleware-channel-executor.md
doc/project/60-solutions/019-middleware/019-middleware.md
doc/normative/50-constitutional-ops/ROOT-IDENTITY-AND-NYMS.md
doc/normative/50-constitutional-ops/MEMBERSHIP-AND-SPONSORSHIP-POLICY.md
doc/normative/50-constitutional-ops/PARTICIPANT-COVENANT.md
doc/normative/50-constitutional-ops/ADVOCACY-AND-SOLICITATION-POLICY.md
doc/normative/50-constitutional-ops/MARKETPLACE-ANTI-FRAUD-POLICY.md
doc/normative/50-constitutional-ops/PANEL-SELECTION-PROTOCOL.md
doc/normative/50-constitutional-ops/ABUSE-DISCLOSURE-PROTOCOL.md
doc/normative/50-constitutional-ops/PROCEDURAL-REPUTATION-SPEC.md
doc/normative/50-constitutional-ops/REPUTATION-VALIDATION-PROTOCOL.md
doc/normative/50-constitutional-ops/NODE-RIGHTS-CARD.md

Status¶

Draft

Date¶

2026-04-23

Design/reuse review: 2026-10-10. The original date and frozen decisions remain; the social-correction integration below is planned, not a runtime completion claim.

Systemic-repair follow-up: 2026-10-13. P051-010–014 translate the protection, reintegration and systemic-repair norms into planned contracts and runtime work; they do not reopen or extend the qualified local baseline.

Sequencing clarification: 2026-10-14. The preferred lower-rework order below is separate from the manual slice's necessary dependencies and qualification scope.

MVP Decisions Frozen on 2026-05-27¶

The following decisions are frozen for the MVP membership/sponsorship contract. Later documents may refine scoring, UX, or federation policy without changing these primitives:

Entry is per influence surface, not one global membership bit.
Entry classes are defined by doc/schemas/_shared/membership-enums.v1.schema.json.
Influence surfaces are defined by the same shared enum; the high-trust surface is public-trust, while public-trust-role remains an entry class.
surface-access-policy.v1 is the canonical policy-axis source of truth.
participant-entry-profile.v1 is a computed subject read model, not an independent source of per-surface permission truth.
participant-effective-limits.v1 is the runtime-facing composed read model for entry defaults, surface policies, capability sanctions, and appeal results.
Sponsorship gives candidacy, not authority.
Sponsor liability is one-hop by default and becomes wider only after an anti-collusion process establishes a sponsor ring.
Sponsorship uses named templates and ordinal liability classes instead of ad-hoc numeric exposure parameters.
Public-object adjudication separates alarm, review, mediation/appeal, sanction, and anti-collusion sweep.
Anti-collusion MVP baselines are sponsorship velocity, co-flagging coherence, and closed-loop receipt detection.

Executive Summary¶

Orbiplex already states that reputation is a safety instrument rather than a status ladder, that endorsements carry asymmetric risk, and that major decisions must leave a trace and an appeal path. What is still missing is one proposal that freezes the bootstrap shape of swarm membership and the first adjudication layer for public pseudonymous objects.

This proposal covers:

how a new member enters,
how sponsorship works,
how early trust and early accountability compose,
how artifact-level judgments can turn into participant-level reputation events,
how public objects should be flagged, reviewed, mediated, escalated, and appealed,
and how anti-collusion logic should remain a separate layer rather than a hidden side effect of one case.

The decisions of this proposal are:

swarm membership SHOULD begin through invitation or sponsorship by at least one existing member rather than through anonymous open self-enrollment,
sponsorship is not a symbolic gesture; it creates bounded reputational exposure for the sponsor,
first-run node onboarding SHOULD guide the operator through identity creation, role choice, service posture, and community-surface selection,
participant reputation is procedural and event-based, not a single global score,
artifact reputation may affect author reputation only after explicit quantity and quality thresholds are crossed,
for public pseudonymous artifacts, signaling danger, community judgment, and reputational sanction MUST be treated as distinct layers,
a participant MAY choose mediation with the parties who influenced a reputation event before opening a formal appeal,
formal appeal is handled by a randomly selected independent community group that did not participate in the original decision and is at least as large as the decision-forming group within the applicable panel-size rules,
a separate anti-collusion sweep SHOULD continuously look for correlated group behavior and procedural contamination across cases,
concrete mechanisms MUST vary by pseudonymization class and accountability surface rather than assuming one universal identity regime.

This keeps the stratification clean:

constitutional and values documents define the principles,
this proposal defines the bootstrap procedure and decision frame,
later schemas and implementations specialize by community policy and pseudonymization class.

Context and Problem Statement¶

Orbiplex already has strong language for:

procedural reputation,
appeal rights,
asymmetric endorsement risk,
revocable pseudonymity,
and evidence-based challenge paths.

What it does not yet have is one explicit project-level contract for:

the first phase of swarm membership,
the transition from artifact-level judgments to reputation events,
and the public-object adjudication path that separates alarm from sanction.

Without that contract:

"joining the community" remains folklore,
invitation and sponsorship have unclear operational meaning,
artifact evaluation may drift into ad-hoc author scoring,
public-object review can collapse into majoritarian reflex,
appeal rights exist normatively but not yet as a concrete bootstrap flow,
and privacy-sensitive communities may accidentally copy the wrong accountability posture from public or federation-facing layers.

The hard part is not only technical. The membership edge is where:

trust is weakest,
social attack cost is lowest,
Sybil pressure is highest,
and privacy/accountability trade-offs are easiest to oversimplify.

There is one more problem specific to public pseudonymous surfaces. Some public objects become strongly nootically charged: an opinion, comment, rating, or short utterance starts carrying one or several memes in the memetic sense. From there, groups do not always behave like stable permanent factions. Sometimes the system is facing exactly that; sometimes it is facing an ad-hoc cluster of correlated action temporarily assembled around one charged symbol, fear, identity marker, grievance, or worldview cue.

In a more apophatic and enactive reading, the system should avoid prematurely reifying "the cartel" as a single durable substance. What must be detected first is correlated procedural force:

repeated co-flagging,
unusually dense co-voting,
synchronized threshold-pushing,
asymmetric targeting,
or self-reinforcing reputation exchange.

This proposal therefore treats collusion primarily as a dynamic procedural pattern, not only as a named faction with stable membership.

Goals¶

Define the recommended bootstrap path for entering a swarm community.
Define sponsorship as a first-class accountability relation.
Freeze the distinction between artifact-level reputation and participant-level reputation.
Define a bootstrap adjudication path for public pseudonymous artifacts.
Separate alarm, review, sanction, and anti-collusion sweep.
Define mediation and appeal ordering for bootstrap governance.
Make pseudonymization class a first-class parameter of these mechanisms.
Define membership as access to bounded influence surfaces rather than as one global "accepted participant" switch.
Define newcomer capability slow-start as a default safety posture, not as a permanent caste.

Non-Goals¶

This proposal does not define the final global reputation engine.
This proposal does not define exact scoring formulas, weight functions, or aggregation math.
This proposal does not define private deniable communication moderation rules.
This proposal does not define the final random-selection protocol for every review and appeal pool.
This proposal does not define the final schema family for every membership, sponsorship, review, mediation, appeal, or collusion artifact.
This proposal does not require one network-wide uniform policy. Federations, communities, organizations, and private groups may tune thresholds and visibility surfaces.

Decision¶

1. Membership Bootstrap Is Relational¶

Joining a swarm community SHOULD begin through invitation or sponsorship by at least one existing member.

The design intent is not aristocratic gatekeeping. The intent is to make entry into a trust-bearing network social and auditable rather than purely mechanical. An invitation means:

an existing participant is willing to create a local trust edge,
the new participant starts with bounded but non-zero social context,
and the network can reason about early accountability without pretending that all fresh identities are equivalent.

Communities MAY require more than one sponsor for higher-risk roles or broader service exposure.

Open self-enrollment is not forbidden forever, but if a community permits it, that path SHOULD carry lower initial trust, tighter capability limits, and slower eligibility for trust-bearing roles.

The practical rule is:

Orbiplex does not build a wall around personhood.
It builds sluices around shared influence.

Membership SHOULD therefore be evaluated per influence surface rather than as a single global admission bit. Common surfaces include:

local-read,
contactability,
public-comment,
public-publishing,
unsolicited-dm,
broadcast,
marketplace,
custody,
routing,
moderation,
arbitration,
governance,
and public-trust.

Each surface may have its own threshold: contact attestation, sponsorship, probation, reputation, IAL, source diversity, anti-collusion checks, conflict disclosure, multisig, or manual review.

The entry class for high-stakes role eligibility remains public-trust-role. The surface where that authority is exercised is public-trust.

The default entry ladder is:

guest,
contactable-participant,
sponsored-candidate,
probationary-member,
full-participant,
public-trust-role.

Contact attestation is useful for anti-spam and recovery, but it is not civil identity and not a substitute for reputation, IAL, or public-trust screening.

2. Sponsorship Creates Bounded Reputational Exposure¶

Sponsorship is not a ceremonial marker. It is a bounded reputational relation.

If a sponsored participant later causes harm, repeatedly violates contracts, or accumulates serious evidence-backed negative procedural events, the sponsors and introducers MAY themselves receive derived reputation events proportional to:

their proximity to the sponsored participant,
the sponsorship template and declared scope,
the freshness of that sponsorship,
and the severity of the resulting harm.

This is the bootstrap form of endorsement asymmetry already present in the normative layer. The purpose is not collective punishment. The purpose is to make trust extension carry real care and discernment.

Federation or community policy SHOULD cap how far this derived liability can propagate and MUST keep it challengeable by evidence and context.

The sponsor does not guarantee the moral essence of the invitee. The sponsor states only:

I know this subject well enough to introduce it to this Orbiplex surface, in this scope and risk limit, and I accept bounded reputational exposure if that act of trust proves grossly careless or collusive.

Sponsorship gives candidacy, not authority. The sponsored participant must still pass the threshold of the target surface.

The first sponsorship artifact is membership-sponsorship.v1. It SHOULD carry sponsor subject, invitee subject, scopes, sponsorship/template, issued and expiry times, probation window, structured due-diligence references, revocability, revocation-tail duration, and evidence policy.

Sponsorship templates avoid false precision. The first template set is:

light-vouch,
standard-introduction,
strong-vouch,
mentor-with-liability.

Derived liability SHOULD be bounded by policy and classified ordinally rather than calculated as a product of local coefficients. The first liability classes are:

negligible,
mitigated,
moderate,
serious,
collusive.

Each liability decision should state the triggered conditions and evidence refs. For example, "classified as serious because the sponsor ignored three prior red flags and continued mass-sponsoring into the same surface" is more auditable than an unexplained numeric score.

Derived sponsor liability is one-hop by default. It should propagate further only when a separate anti-collusion process establishes an organized sponsor-ring.

To prevent clan capture, a sponsor SHOULD only sponsor into surfaces where the sponsor has sufficient reputation and authority. Higher-risk surfaces SHOULD require multiple sponsors from sufficiently independent clusters, active sponsorship caps, and anti-sponsor-ring monitoring.

3. First-Run Node Onboarding Should Be a Guided Wizard¶

A user starting a local Node SHOULD be guided by a multi-step onboarding wizard rather than dropped into raw configuration.

The wizard SHOULD at minimum:

generate or import the operator's identity material,
explain the selected accountability surface and pseudonymization posture,
ask which services or roles the operator wants to expose,
ask whether the node is joining an existing community or starting a new local one,
collect invitation or sponsorship material when applicable,
and explain the consequences of joining with a public, accountable, or more private participation surface.

The first-run experience matters because social bootstrap is part of system security. Incorrectly defaulting a participant into the wrong visibility or accountability surface is not a cosmetic error; it can become a privacy breach or a governance mismatch.

4. Reputation Is Event-Based and Multi-Surface¶

Every participant identity that is visible on a reputation-bearing surface has reputation, but that reputation MUST be modeled as a history of events and domain-specific projections rather than as one timeless scalar score.

The practical bootstrap invariant is:

evidence or judgments produce append-only reputation events,
read models derive local, domain, and community views from those events,
and authority decisions should key off those bounded views, not off a myth of one universal number.

This keeps reputation auditable and lets communities apply different thresholds for:

safety,
competence,
reliability,
social care,
moderation,
or service stewardship.

5. Artifact Reputation Precedes Author Reputation¶

When a community evaluates an artifact, service output, publication, or other user-authored result, the first reputation subject SHOULD be the artifact or event itself. Participant-level reputation impact should arise only after an explicit threshold is crossed.

Examples of threshold parameters include:

minimum number of independent evaluations,
minimum evidence quality,
minimum evaluator diversity,
waiting period before projection to author reputation,
and conflict-of-interest filtering.

This avoids premature personal scoring from a handful of reactions and keeps the system closer to facts before identity-level consequences are derived.

A later implementation MAY emit an explicit derived event such as:

"artifact X crossed threshold T and now contributes procedural/community signal Y to author Z".

The threshold policy is local or federation-scoped, not global protocol law.

6. Public Adjudication Must Separate Alarm from Sanction¶

For public objects in a public or otherwise reputation-bearing pseudonymous surface, Orbiplex SHOULD explicitly distinguish at least four layers:

flagging
review
appeal
cartel-detection

The core invariant is:

Flagging raises an alarm.
Review forms a provisional social judgment.
Appeal challenges that judgment.
Sanction happens only after the judgment path is sufficiently grounded.

A cluster of flags MUST NOT by itself be treated as a final reputation penalty. Otherwise the system becomes cheap to weaponize by:

majoritarian pressure,
factional mobilization,
dynamic meme-driven clustering,
or ordinary ideological bubbles.

The default public adjudication flow is:

stateDiagram-v2
    [*] --> Clear
    Clear --> Alarm: effective threshold crossed
    Alarm --> Review: open first review
    Review --> Clear: dismiss alert
    Review --> Contested: concern remains but sanction is not grounded
    Review --> Mediation: author requests mediation
    Review --> Substantiation: plural review needed
    Review --> Contaminated: procedure contaminated
    Mediation --> Clear: support falls below threshold
    Mediation --> Contested: disagreement remains
    Mediation --> Substantiation: mediation fails
    Substantiation --> JudgedSubjective: subjective dispute
    Substantiation --> JudgedSubstantiated: substantiated deregulation
    Substantiation --> Contaminated: contamination found
    JudgedSubjective --> Appeal: formal appeal
    JudgedSubstantiated --> Appeal: formal appeal
    Appeal --> Clear: appeal succeeds
    Appeal --> JudgedSubjective: prior subjective judgment upheld
    Appeal --> JudgedSubstantiated: prior substantiated judgment upheld
    Appeal --> Contested: merits review finds sanction basis insufficient
    Appeal --> AppealBlocked: eligible pool or quorum unavailable
    AppealBlocked --> Appeal: independent pool restored
    AppealBlocked --> Council: authorized escalation
    Contaminated --> Council: high-significance or repeated failure
    Council --> Review: remit for independent reconsideration
    Council --> Appeal: constitute independent appeal
    Council --> AppealBlocked: no authorized quorate path available
    Clear --> Withdrawn: author withdrawal
    Contested --> Withdrawn: author withdrawal
    JudgedSubjective --> Withdrawn: author withdrawal
    JudgedSubstantiated --> Withdrawn: author withdrawal

Mediation and council review are branches of one state machine, not parallel systems of authority. The diagram summarizes the procedure, not effect-application state. An unsuccessful appeal does not upgrade a subjective dispute into substantiated misconduct; only a competent merits review finding the sanction's basis insufficient leads from appeal to Contested. An appellant's failure to supply new evidence does not itself invalidate the prior finding.

AppealBlocked is a procedural-liveness state, neither exoneration nor confirmation of the prior finding. Retain that finding as history; the recorded applicable policy separately determines whether its effects are stayed or remain effective, subject to current authority and expiry. Blockage cannot renew a hold, create guilt or remove unrelated restrictions. The first local profile uses Council to remit or arrange independent review, not as an additional unqualified merits authority.

7. Public Adjudication Applies to Publicly Indexed User Objects¶

The protocol described here is intended for public objects in a public or otherwise reputation-bearing pseudonymous mode, especially:

opinions,
comments,
ratings,
short public statements,
and other publicly indexed user-authored content.

It does not directly govern private deniable communication. At most, aggregate and anonymized traces from private layers may inform higher-level abuse detection, but not by silently importing private content into public adjudication.

8. Protocol States Must Preserve Semantic Distinctions¶

A public object under adjudication SHOULD move through explicit protocol states rather than a binary "acceptable / punished" model.

The minimum recommended state shape is three orthogonal axes:

lifecycle: clear, under-review, judged, withdrawn
judgment_qualifier: contested, subjective, substantiated, contaminated
withdrawal_reverts_event: boolean, meaningful only when lifecycle = withdrawn

The equivalent diagnostic labels can be rendered as judged.substantiated, under-review.contaminated, or withdrawn.reverted without multiplying the wire enum into every possible combination. Case-procedure progress, including AppealBlocked, and effect-application status are separate from these public-object axes. Do not overwrite a judgment qualifier with a scheduling or pool-availability failure. These axes matter because the system must be able to say:

"this object triggered concern",
"this object is controversial",
"this object was judged procedurally unsafe",
or "this case was contaminated and cannot ground a clean sanction"

without collapsing all of those into one punitive label.

9. Alarm Thresholds Must Use Independence, Not Raw Count¶

When participants flag an object as potentially deregulating, the system SHOULD store not only the raw signal but also enough metadata to reason about independence and contamination.

At minimum this includes:

object identity,
flagger pseudonyms,
timestamps,
local justification or reason codes,
and correlation hints between participating flaggers.

The system SHOULD then derive at least:

raw flag count,
effective independent support,
cluster dispersion,
growth velocity,
and historical co-action coherence.

An alert should trigger only when all of the following are true:

the raw threshold is reached,
the effective independent threshold is reached,
and initial contamination is not already too high.

Crossing that threshold creates an alert and review state, not a final author sanction.

10. Provisional Holds Must Stay Mild and Reversible¶

If a deployment wants immediate defensive reaction, it MAY apply only a mild and reversible provisional hold at alert time.

The hold:

MUST have a bounded lifetime and an explicit removal/compensation path,
MUST NOT be treated as final guilt,
and MUST cease to affect current admission when it expires or the later review does not substantiate the case. A failed cleanup job must not prolong its authority.

"Reversible" concerns the continuing restriction, not erasure of past effects: a missed opportunity or a disclosed allegation cannot be undone by clearing a record. Preserve the original facts, append the correction, and identify any remaining repair obligation. Soft penalties need their own validity boundary; expiry of a hard block does not currently expire the soft factors in S040.

This keeps the safety posture responsive without letting the alert threshold silently become punishment.

The first review answers a narrower question than the final sanction path:

does this alert appear to come from plural enough social concern, or mainly from a local faction, correlated cluster, or contaminated process?

The reviewing group SHOULD therefore be:

smaller than the later escalation jury,
selected with anti-collusion filters,
and hidden until the voting window closes.

Selection SHOULD be drawn from several baskets at once, for example:

high procedural reputation,
medium but stable reputation,
graph distance from the author and principal flaggers,
topic competence,
and low historical co-voting correlation.

Candidates SHOULD be excluded when they:

are too close to the author,
are too close to the primary flagging cluster,
belong to an already overrepresented dense cluster,
show strong prior signs of collusive co-action,
or have an active conflict of interest.

The first review SHOULD be able to emit at least:

dismiss-alert
continue-as-contested
continue-to-mediation
continue-to-substantiation
mark-review-contaminated

These baskets and temporary blindness describe lightweight anti-collusion triage, not a replacement for a normative panel. Where DIA-PANEL-SEL-001 applies, reputation is an eligibility gate, not a draw weight; its uniform selection, disclosure and party-veto rules take precedence. Hiding votes must not hide the composition information needed to exercise those rights.

12. Mediation Comes Before Formal Appeal When the Participant Chooses It¶

Before opening a formal appeal, the affected participant MAY choose mediation with the person or group that influenced the reputation event.

Typical cases include:

evaluators of an artifact,
a moderation subgroup,
sponsors contesting how derived liability was applied,
or other participants whose judgments contributed directly to the event.

Mediation is not mandatory. It is an optional lower-cost path that can:

correct misunderstanding,
add missing context,
retract low-quality inputs,
or narrow the dispute before it reaches a formal review group.

If mediation fails, is refused, or is inappropriate due to direct safety risk, the participant retains the full right to formal appeal.

For public-object adjudication, mediation is especially useful after a first review but before larger escalation. It is one bounded chance to let the author submit justification and let original flaggers:

maintain their flag,
withdraw it,
soften their classification,
or remain silent until expiry.

If mediation drops the effective support below the safety threshold, the object SHOULD return to contested or clear, and any provisional hold MUST be reverted.

13. Community Escalation Must Be Larger, More Plural, and Anti-Collusive¶

If the case continues, the next layer is community escalation through a larger, plural, anti-collusively selected jury.

This jury SHOULD:

be substantially larger than the first review pool,
preserve strict anti-cluster composition rules,
require both nominal quorum and effective pluralistic participation,
and receive the object plus structured case material, but not a herd-forming "N people already condemned this" summary.

Jurors SHOULD be asked on multiple axes rather than by a single binary button. The minimum useful questions are:

does the object violate conditions of cooperation?
is the current classification too strong?
does the case mainly express ideological disagreement?
does the input procedure appear contaminated?

Each juror SHOULD also choose one class such as:

not-deregulation
subjective-dispute
possible-deregulation
substantiated-deregulation
procedural-contamination

and provide a short justification.

Larger here means relative to lightweight first triage, not an unbounded crowd. Where the normative panel protocol applies, its panel-size and quorum rules govern; additional advisory reviewers do not acquire votes or expand the adjudicating panel.

If quorum is insufficient, the result SHOULD be insufficient-quorum and the protocol MAY re-open the case only through backoff windows rather than immediate retries.

14. Final Reputation Events Require Substantiation¶

The cautious default is:

alert stage -> at most provisional hold,
substantiated escalation result -> final reputation event,
successful challenge or contamination finding -> retract or supersede the affected reputation consequence and recompute current projections while preserving procedural metadata. Already observed external effects require correction or compensation, not a claim that history was rolled back.

This is the point where the proposal most clearly rejects soft tyranny by raw majority. The sanction should be grounded only after enough independent review has survived:

anti-collusion filtering,
mediation or challenge opportunity,
and a plural final review path.

15. Object Withdrawal Should Revert the Linked Reputation Event When Early Enough¶

The author MAY withdraw the object regardless of case status.

If the object is withdrawn within a policy-defined early window, the system MAY:

mark the object as withdrawn,
revert the linked reputation event,
and keep only the procedural metadata needed for audit, cartel detection, and historical reconstruction.

This gives the protocol a repair path without pretending that the case never happened.

16. Formal Appeal Uses an Independent Randomly Selected Community Group¶

A formal appeal MUST be handled by an independent randomly selected community group whose members:

did not participate in the original decision,
have supplied an admissible COI declaration and have no disqualifying conflict of interest in the matter,
and are numerous enough to avoid being weaker than the original decision surface.

The bootstrap minimum is:

the appeal group is at least as large as the group whose decision or combined judgments produced the original reputation event, within the governing panel protocol's size limits. The admitted first-instance policy must make this compatible in advance; flaggers and advisory reviewers are not panel members.

The purpose is procedural symmetry. A participant should not need to overturn a group decision before a smaller or less independent body than the one that produced it.

The governing rules are not a blank slate: DIA-PANEL-SEL-001 defines eligibility, recusal, party veto, quorum and insufficient-pool escalation for panels in its scope. Its VRF mechanism remains a hypothesis. Different lightweight local profiles may specialize admissible pools and selection only within their declared competence and the normative floor described in section 25. The resulting process stays:

auditable,
conflict-aware,
and challengeable.

This rule applies both to ordinary appeal of participant-level decisions and to appeal of public-object adjudication results.

17. Newcomers Use Slow-Start Capability Limits¶

A new or low-evidence participant SHOULD receive narrow initial influence limits. The canonical newcomer fixtures are:

doc/schemas/examples/default.surface-access-policy.json
doc/schemas/examples/newcomer.participant-entry-profile.json
doc/schemas/examples/newcomer.participant-effective-limits.json

These limits are not a punishment. They represent influence that has not yet been earned. participant-entry-profile.v1 is a computed subject read model. participant-effective-limits.v1 composes entry defaults, surface policy, sanctions, and appeal results into the runtime-facing view. A separate participant-capability-limits.v1 artifact may express sanctions or explicit restrictions, but it should not be confused with ordinary newcomer entry policy.

18. Spam, Solicitation, Advocacy, and Fraud Are Surface-Abuse Classes¶

Orbiplex SHOULD avoid global worldview bans. It SHOULD instead regulate how shared communication and marketplace surfaces are used.

The default abuse rules are:

no broadcast or unlimited unsolicited DM for fresh participants,
no mass import of contacts without relationship or recipient consent,
no unsolicited political, ideological, religious, campaign, or financial persuasion outside opt-in surfaces,
clear tags for advocacy and campaign content,
material conflict-of-interest and sponsorship disclosure,
marketplace offers through explicit marketplace/service surfaces, not hidden acquisition funnels,
low value caps and escrow/procurement contracts for newcomers,
no transferable reputation from self-dealing or closed receipt loops.

This preserves pluralism while defending the surfaces on which pluralism depends.

19. Governance Council Is Reserved for High-Significance or Contaminated Cases¶

The governance-council layer SHOULD remain exceptional.

It is appropriate only for cases such as:

highly precedent-setting disputes,
very large or heavily contested public cases,
repeated failed escalation attempts,
explicit review-contaminated outcomes,
or disputes that require constitutional interpretation.

The design goal is not to create a standing aristocracy. The goal is to provide one higher procedural layer for rare cases where ordinary community escalation is insufficient or itself compromised.

Council composition SHOULD mix:

attested expertise,
procedurally trusted participants,
anti-cluster random selection,
and rotating civic assessors.

Council decisions SHOULD permit:

full justification,
separate opinions,
conflict-of-interest registry,
ex post audit,
and doctrinal revision later without forcing retroactive erasure of every old case.

For the first local correction profile, the Council destination must identify an authorized body able to arrange independent review or remit a contaminated case. If that route is also unavailable, retain an explicit blockage, not a synthetic ruling. Any later merits-adjudication role requires its own applicable mandate and panel safeguards; the label Council alone grants none.

20. Anti-Collusion Sweep Must Be a Separate Continuous Subsystem¶

Cartel detection should not live only inside one case file. Orbiplex SHOULD run it as a separate periodic sweep over many cases.

Its purpose is not primarily punitive. It is to:

detect correlated procedural force,
reduce the influence of highly correlated groups,
improve future jury selection,
mark cases as potentially contaminated,
and, only when needed, open separate procedural-abuse review.

Useful sweep inputs include:

flagging history,
co-voting history,
co-occurrence in disputes,
timing patterns,
mutual reputation boosting,
topic distribution,
asymmetry of targeting,
and mediation and escalation metadata.

The MVP baseline detectors are deliberately narrow:

sponsorship: abnormal sponsorship velocity,
public adjudication: co-flagging coherence across objects,
marketplace: closed-loop receipt detection.

Additional detector families should be added only when a concrete operational need appears, so "anti-collusion" does not become an unbounded bucket.

The sweep SHOULD prefer graded outputs such as:

low-correlation
watch
elevated-collusion-risk
high-collusion-risk

rather than a binary "cartel / not cartel" ontology.

This again matches the apophatic-enactive caution of the proposal: the system should first see dynamic correlated patterns before it reifies them as enduring factions.

21. Influence Weighting Must Track Independence¶

The effective force of a set of flags or votes SHOULD depend not only on the number of accounts but on their independence.

In practice this weighting should inform:

alert thresholds,
mediation outcome assessment,
escalation validity,
contamination scoring,
and future reviewer selection.

This means that 150 accounts from one dense correlation cluster need not count as 150 independent judgments.

22. Public Agora Nodes May Host Aggregates, Not Final Authority¶

Public Agora nodes MAY host public aggregation and query surfaces for reputation-relevant artifacts, for example:

public counts of flags, reviews, and appeals,
public timelines of object-case state transitions,
public derived read models for artifact reputation,
public candidate projections of participant or nym reputation,
and public dashboards of collusion-risk or contamination markers.

This is acceptable because Agora is already the public substrate for ingest, query, and subscription of relevant records. What MUST remain separated is the difference between:

storing and serving public facts,
projecting those facts into one read model,
and issuing an authoritative reputation judgment.

Therefore a public Agora-backed reputation surface is allowed only as:

an index,
a projection,
a read model,
or a publicly inspectable candidate view.

It MUST NOT silently become the sole authoritative reputation layer for the network.

If a federation or community wants a public Agora-backed reputation service, that service SHOULD be modeled as a separate policy-bearing component operating over Agora data, not as Agora itself.

This preserves the stratification:

Agora stores and serves records,
reputation components evaluate policy and derive judgments,
nodes remain free to accept, reject, or down-weight a given public projection under local policy.

23. Pseudonymization Class Changes the Appropriate Mechanism¶

The concrete membership, sponsorship, reputation, review, mediation, appeal, and collusion-handling mechanics MUST depend on the pseudonymization class and accountability surface defined elsewhere in Orbiplex.

This proposal explicitly does not assume that one mechanism fits:

accountable-nym,
private-nym,
public federation participation,
local community participation,
organization-scoped membership,
or future stronger anonymity constructions.

Examples:

in an accountable-nym context, ordinary public or domain reputation and sponsor liability may be appropriate,
in a private-nym context, community-local invite graphs, scoped reputation, sealed adjudication, or group-local exclusion handles may be more appropriate than portable public sanctions,
a private group may need stronger local exclusion without creating a global cross-group correlation handle,
and an organization may use different membership visibility and appeal pools from a public federation.

Therefore the invariant is:

The stronger the public blast radius, the stronger the accountability hook.
The more consent-bound and private the context, the stronger the unlinkability default.

This proposal should be implemented in conjunction with the evolving nym and root-identity documents, not as an independent identity regime.

24. Concrete Mechanisms Remain Community-Policy Objects¶

The following items are intentionally left as policy parameters or later artifacts:

sponsor count required for entry,
sponsor-liability window,
alert thresholds,
minimum effective-independent support,
contamination cutoffs,
thresholds for artifact-to-author reputation projection,
mediation timeout and admissibility,
size and eligibility of review and appeal pools,
evidence threshold for high-stakes sanctions,
whether reputation is public, domain-local, community-local, or hidden,
and how pseudonymization-class-specific exclusions are implemented.

Orbiplex should freeze the shape early, but not pretend that every community must share identical social mechanics.

25. Close the Correction Loop Without Creating Another Authority Plane¶

P051 owns the social procedure; S040 owns restriction enforcement under P018. Neither a Room role, a Corpus Chair decision, a signed artifact nor a notification action supplies adjudication authority by itself. An admitted community policy must name the authorized decision makers, affected surface, evidence threshold, conflict exclusions and appeal path. Operator authority to run a host is not automatically authority to judge its participants.

The first local profile is a bounded, non-disclosure community correction procedure, not an implementation claim for the full constitutional panel system. It must declare stakes, competence and limits. Constitutional, high-stakes or identifying disclosure cases cannot remain in this lighter procedure merely because it is locally available; route them to the applicable normative procedure or expose the missing route as blocked. Local autonomy does not waive the rights floor.

Upstream norm	What this slice preserves and newly determines
Panel Selection, `DIA-PANEL-SEL-001`, sections 2–8 and 11	Full panel rules govern constitutional/high-stakes/adversarial panels: eligibility, uniform draw, COI and prior-service exclusions, veto, quorum and tiered escalation. P051-002 names the lighter triage scope and the boundary requiring this procedure; it does not replace those rules with reputation baskets. VRF selection remains a mechanism hypothesis.
Abuse Disclosure, section 9	Notice, appeal and a new review composition apply on its sanction/disclosure track. The first local profile adopts at least 14 days to file an appeal; urgent protective isolation is a separate decision, not automatic closure of that window.
Procedural Reputation, sections 2 and 8	Procedural reputation and identity assurance remain separate eligibility gates; evidence is portable, not an automatically accepted score. Local fixtures do not validate the hypothetical scoring functions.
Reputation Validation, M11 and section 5	Reuse M11's reason, review/expiry and appeal-path evidence for denials. P051-008 qualifies a local correction mechanism, not the simulation/shadow/pilot/adversarial-review sequence for reputation leverage.
Node Rights Card, rights and graduated enforcement	Preserve inspection, appeal, subsidiarity and the rights floor against an operator. This slice binds each sanction to its author, basis, deadline and route back, without granting the operator adjudicative power.

The 14-day local floor is an explicit profile decision, not a claim that every normative track has the same deadline: Panel Selection section 7 lists 14 days for normal and 7 for critical appeals, while Abuse Disclosure section 9 requires at least 14 days with an earlier-isolation clause. P051-002 must pin the applicable track and window-start/notice rule; neither a queued notice nor a failed delivery alone proves that a participant had the required appeal opportunity. Longer applicable protections remain in force.

The smallest useful implementation is one community-local public-object case:

signal -> case + assigned owner -> bounded review / optional mediation
       -> reasoned decision -> authorized effect -> observed effect receipt
       -> notice + independent appeal -> correction -> recomputed effects + notice

This is a composition of existing contracts with a small case-domain state machine, not a new transport, global reputation service or generic workflow engine. No sanction, insufficient evidence, voluntary repair and contamination are legitimate outcomes. Insufficient quorum is an explicit procedural blockage, not a merits verdict. Silence, missed notification or an expired review window is not agreement, guilt or permission to extend a hold.

Each open case must retain an accountable procedure owner, a next action, its deadline, and an explicit escalation destination. Ownership transfers require acceptance by the successor and a durable handoff. Departure, expiry or revocation of an owner leaves an actionable unassigned/blocked state; it neither closes the case nor silently delegates it to the infrastructure operator. Historical decisions retain their authors after handoff.

The case owner persists distinct review deadlines, appeal-filing windows and effect-validity limits. S020 owns wake-ups and bounded reconciliation launches, not the meaning of those deadlines. Admission checks current mandate/effect validity synchronously, independently of scheduled cleanup. Read-only inspection may derive an overdue status but must not execute a correction. The baseline does not promise proactive detection while Scheduler and all declared triggers are unavailable: validity is enforced on use; durable reconciliation and notices resume through an authorized execution path or resumed scheduled work. A stronger proactive guarantee needs a separately declared independent monitor, outside this slice, not a private timer loop hidden in the case service.

Record separately the signal author, subject, reviewer, decision authority, effect executor and appeal authority. Bind actions to the exact case revision, object revision/digest, policy revision, target decision and admitted mandate. Historical replay uses retained inputs; new effects recheck current authority, revocation and time validity. A policy change must not silently rewrite the standards used to judge the original action.

Appeal can be filed without participating in the disputed Room or receiving a notification. Selection retains auditable eligibility, recusal and randomness evidence under the chosen policy; another agent or nym controlled by the original decision maker is not evidence of independence. If a valid appeal pool cannot be formed, expose that blockage and use the declared escalation path, not the original group relabeled as an appeal body. Do not require civil-identity disclosure or cross-community correlation merely to establish a review pool. If the case owner or host operator is a party, they cannot accept the case as its adjudicator, including through controlled alternate identities. Transfer requires an independently authorized recipient's acceptance; otherwise record the blockage and escalation destination. OQ6 concerns provisioning that route, not permission for self-adjudication. The admitted policy must also state the effect-stay rule during an appeal or blockage; runtime must not invent one.

26. Correct the Targeted Consequence, Not the Whole Participant¶

A correction targets identified decisions and their derived effects. It does not grant new privileges, erase unrelated sanctions, override protected floors or reopen revoked capability passports. Newcomer defaults remain entry policy, not evidence of misconduct.

Maintain case/source attribution above S040 and recompute the remaining applicable restrictions after a retraction. The current participant-wide clear operation is not a case-specific undo. Likewise, membership-policy-core's ordered overlays are a deterministic composition mechanism, not authority precedence: an appeal-result appended last must not defeat an unrelated case merely because its value is allow.

The composition adapter must account for existing operator-imported restrictions as independent sources and serialize or fence all competing writers. Unknown source attribution blocks automatic relaxation and is visible for reconciliation; it is not permission to retain an already-expired case effect. Recompute only from still-applicable sources, with explicit bounds on sources and evidence.

Before the adapter's first case effect or correction, P018-13 must inventory the existing restriction records and clear tombstones and retain a revision/digest-bound baseline. Represent retained operator inputs as legacy-operator-source, preserving their original evidence and unknown attribution; the label is not proof of a signed verdict or recovered case ownership. Fence every writer during cutover and verify effective restrictions at the same evaluation time for valid, unambiguous sources. Malformed or ambiguous inputs retain their source and a typed refusal/diagnostic; preserving a legacy fail-open result is not a compatibility goal. Resume a partial migration idempotently before enabling correction. Unresolved inputs require explicit reconciliation, not guessed attribution or silent removal.

Keep procedural outcome and effect application status separate. A successful appeal can coexist with pending local recomposition or unacknowledged remote correction delivery. Completion requires receipts for the declared effect scope; unreachable consumers remain pending or explicitly unresolved. A local clear cannot retract a publication already observed elsewhere.

27. Separate Protection, Reintegration and Systemic Repair¶

This extension implements the direction in Core Values, "Care and Justice as Two Relational Modes", Membership and Sponsorship, "Sanctions and Return Path", Abuse Disclosure, section 7, and Reputation Validation, M8 and intervention-effectiveness review. It adds reviewable responsibility and evidence, not a new sanction ladder or adjudication mandate.

Concern	Responsibility	Boundary
Protection	Contain a specific evidenced risk through existing P018/S040 effects, with individual grounds, current admission and independent validity.	Systemic review or repair must not delay necessary protection; aggregate prevalence is not individual evidence.
Reintegration	Record explicit, reviewable return conditions tied to repair and behavior, and evidence of their assessment. Request any authorized narrowing through existing source-addressed correction.	Meeting return conditions does not clear unrelated restrictions, grant missing entry rights, rewrite historical findings or establish that return is safe without assessment.
Systemic repair	Review access rules, incentives and sponsorship practices; record a proposed correction, accepted responsibility or escalation, execution evidence and outcome review.	A plan does not authorize its execution. Responsibility must use the target component's authorized interface, not mutate its configuration or policy directly.

The prevalence of violations is grounds for reviewing the conditions of cooperation, not an independent basis for increasing or reducing sanctions against a particular participant. Do not introduce a demoralization score, a universal percentage threshold, collective guilt or automatic sanction changes from report volume. Agent/Corpus analysis may support review, never supply authority.

Preserve the norms' MUST/SHOULD distinction: record justified non-applicability or deferral where appropriate rather than making every repair field a prerequisite for urgent protection. An unavailable or conflicted owner requires a competent recipient or explicit blockage; infrastructure ownership does not authorize self-adjudication. Neither a missed repair deadline nor a pending outcome review renews a protective restriction or establishes a new violation by itself.

Reuse and Implementation Recommendations¶

Existing Seams and Their Limits¶

Inspected on 2026-10-10 against the working tree and node:docs/implementation-ledger.toml. These are implementation entry points, not runtime acceptance results. This table records the pre-slice inspection; the 2026-10-10 implementation checkpoint below supersedes its restriction gaps.

Concern	Reuse	Boundary still to implement for P051
Surface policy and effective limits	`node:membership-policy-core/src/lib.rs`: `decide_surface`, `project_effective_limits`, source-tagged overlays and fixture/property tests	Ledger `membership-surface-access-policy-core` is `partial`: no complete daemon/storage/UI admission path. Admit sources, check validity and authority, then supply deterministic, case-aware composition.
Restriction application	S040/P018; `node:daemon/src/execution_host.rs`, `node:daemon/src/lib.rs`, `node:daemon/src/state_checkpoint.rs`, `node:daemon/src/tests/participant_policy.rs`	One current record per participant; clear is participant-wide; one hard expiry does not cover the soft layer or several independently expiring cases. Schema validation and `decision/author` text do not authenticate a community verdict. The pre-slice snapshot dropped an unparseable hard expiry through `.ok()?`; P018-13/14 now replace runtime reads with source-aware composition and retain malformed sources with explicit refusal diagnostics.
Identity and bounded relationships	P034 and S032; existing caller bindings and `local-relationship-core`	Reuse verified actor/owner context and private eligibility inputs. Relationship membership is not a grant; node-operator assurance is not reputation or a jury mandate.
Facts and recovery	S028; `node:temporal-event-log`, `node:storage-runtime`; existing restriction commit stream and notification store	One case-domain source of truth, rebuildable projections and an idempotent effect outbox; do not invent a global transaction registry or copy the case history into S040.
Attention and human action	S039, P057; `node:notification-core`, `node:notification-store`, `node:daemon/src/notifications_host.rs`	Recipient-bound case notices and registered actions must call the case owner, recheck revision/mandate, and report its result. Queue insertion, delivery, opening and substantive response are distinct.
Deadlines and long work	S020 Replay Scheduler; S029/P055 Bounded Deferred Operations	Scheduler wakes bounded reconciliation jobs; deferred handles describe individual work, not the lifetime of a social case. Admission itself checks expiry even if the scheduler is down.
Evidence exchange and provenance	S023 Artifact Delivery; S043/P081 causal context and execution receipts; P090/P086 evidence distinctions	Add domain acceptors and case-specific evidence references. Delivery, signature verification and host observation do not establish guilt, independent support or final adjudication.
Optional deliberation and policy extension	S036 Room, S038 Corpus, S047 Agent; P085 and `node:semantic-registry-core`	Reuse conversation/review assistance and exact revision-bound policy admission patterns. No existing generic social-policy registry or autonomous adjudicator is claimed. Agent output remains advisory; private case data needs its own disclosure/egress admission.

Canonical companion contracts:

S019 Middleware and P080 channel executor, with S015 Module Store and S016 bounded runtime;
P034 operator binding and S032 local relationships;
S028 temporal storage, S020 Scheduler and S029 deferred operations;
S023 Artifact Delivery and S043 horizontal primitives;
S036 Room, S038 Corpus and S047 Agent;
P085 extension admission, P090 inference provenance and P086 observation.

Case Service Hosting Decision¶

The implementation direction is a small Rust domain core, a case service hosted through existing middleware, and a host-owned enforcement adapter. The service records and guards the procedure; it is not an autonomous judge, a generic ticket system or a new workflow engine. A procedural owner recorded in a case remains an accountable actor, not the service process or its infrastructure operator.

Rust core. Keep case data, the transition table, semantic validation and replay/projection pure. Supply time and verified authority context explicitly; the core does not read the clock, files, network or daemon state. Reuse membership-policy-core vocabulary and evaluators where their semantics match; do not move the case lifecycle into that kernel or copy its policy composition. Keep runtime dependencies above the core and enforce that direction in tests.
Middleware-hosted service. Use S019/P080's supervised channel_json lifecycle, authenticated session, module reports, declared routes and host-call bridge. Reuse existing package admission, module data-directory and bounded runtime facilities instead of adding a private supervisor, listener or IPC protocol. The service owns the case fact store, rebuildable read models and durable effect outbox under S028, reusing existing storage primitives. Case transition and pending-effect intent share a transaction; host application is a separate, recoverable step, not a cross-store atomicity claim.
Host-owned enforcement adapter. Keep P018/S040 source composition, current authorization and restriction application behind the host boundary. The service requests a case-scoped effect through the existing host-call mechanism; the host checks caller binding, case/decision revision, mandate, target, revocation, validity and idempotency before applying it and returning an outcome. Module admission is not adjudicative authority. Do not expose unrestricted operator import/clear to the service or let it write the restriction store directly.

P051-002 defines the narrow command/outcome boundary jointly with P018-12. Any missing operation is explicitly registered and schema-gated under P072; the existing bridge is reused, not assumed to already provide a case-effect API. responsibility-core / responsibility-service settle the local crate and process boundary under P051-003 without duplicating the Rust rules in a second implementation. Follow node:DEV-GUIDELINES.md; create separate crates only where the dependency boundary warrants them. Daemon routing stays a thin adapter, not the owner of social procedure.

Reuse S039 for notices and action delivery, S020 for bounded reconciliation launches, and S029 when one operation needs a deferred handle. None replaces the case's durable history. Detachment or loss of a required host capability leaves pending work and an explicit unavailable/blocked outcome; reattachment revalidates authority before retrying. It cannot transfer adjudication to the host operator, renew a restriction or turn an unacknowledged effect into success. Existing host admission continues to enforce effect validity independently of service uptime.

Small Contracts Before Endpoints¶

Specialize only the first required artifacts from the family below. Freeze a transition table and semantic validators before implementing UI or importing remote judgments. Common case bindings should include case/id, case/revision, subject/ref, object/ref plus digest, surface/id, policy/ref plus revision, actor/mandate references, validity and targeted predecessor/effect references. These are design requirements, not newly published schema fields.

Keep community criteria and thresholds in admitted policy data; do not create a closed repertoire of acceptable opinions, problem topics or successful outcomes. Start with explicit local policy admission; portability and signed community packages are later adapters, not prerequisites for local correction. Where a semantic-registry specialization is used, preserve exact binding, expiry, revocation and unresolved-ceiling semantics. Package admission cannot create adjudicative authority by itself.

Schema Gate must protect the exposed contracts, with semantic checks for current mandates, target identity and permitted transitions in the owning core/host. Register operations under P072 with their actual dispatch/host-route boundaries; an inspection capability must not expose operator mutation. Reuse existing signing, canonicalization, delivery and caller-binding facilities, not ad hoc signatures, HTTP callbacks or UI-supplied actor identities.

Durable Correction and Bounded Attention¶

Persist the authorized case transition and pending effects atomically in the case store. Apply effects through existing owners with a case/effect-scoped idempotency key; append their outcome references. Retry after restart without double sanction, duplicate reputation events or repeated notices. Cross-store failure stays a reconcilable pending effect, not a fabricated atomic success.

Use source-addressed retractions/supersessions for reputation projections. Keep sealed or access-controlled evidence separate from minimal case metadata and public corrections; neither notifications, SSE nor traces may export the full private case. Bound retention, payload size, fan-out, queue length and retry time; cleanup must preserve live appeal obligations and disclose evidence-retention gaps.

Protected access needs a real path: an affected participant must be able to read the admissible reasons and file a bounded appeal despite a publication/marketplace restriction, Room removal or muted notifications. "Not hard-blocked" alone is insufficient; test soft cooldowns and ordinary anti-abuse limits for starvation. This does not grant unlimited messaging or bypass recipient privacy.

Structured Review and Repair Without a Parallel Workflow Engine¶

P051-010–014 extend the existing case service. The current node:responsibility-core/src/contracts.rs has RecordReview { reason }, decisions, evidence references and targeted correction; these are reuse entry points, not evidence that structured repair is already implemented. Inspect the current contracts and callers before selecting fields or new operations.

Small contract extension. Extend case/review artifacts or attach a bounded, case-bound artifact where that avoids coupling independent lifecycles. Freeze the minimal transition table before endpoints. Represent reviewed scope and observation window, evidence references and uncertainty, an alleged mechanism separately from an established finding, intended correction/outcome, owner and mandate bindings, accepted ownership or escalation, return conditions, review deadline, execution evidence and outcome assessment. These are semantic requirements, not preselected schema names or a closed catalogue of repairs. Keep criteria in admitted policy data; missing legacy information is unknown or not assessed, never completed, safe or harmless.
Separate claims and authority. Proposed, authorized, accepted, executed and outcome-reviewed claims require their own evidence. Reuse signed commands, current host admission and the accepted-handoff pattern rather than treating an owner label as acceptance. A plan's acceptance does not extend a mandate; execution must pass the owning component's current authorization. Preserve protected reasons/appeal access and the existing conflict/independence rules. Version changed wire/replay semantics where necessary, preserving historical signed bytes. Use canonical schemas, Node mirrors, Schema Gate and P072's existing registration boundaries; do not make operator writes module-dispatchable.
Existing runtime and storage. Keep transitions and validation pure in responsibility-core, durable coordination in responsibility-service through S019/P080 channel_json, and enforcement in the P018/S040 host adapter. Reuse S028 facts, rebuildable projections, transactional outboxes, exact receipts and idempotency. No second restriction engine, listener, supervisor or generic workflow engine. Bound cross-case references, history and review work so that repair analysis cannot starve protection, participant rights or correction.
Attention and deadlines. Extend the existing signed API and ParticipantSession UI; reuse S039 recipient-bound actions/notices. A queued notice does not prove acceptance, execution or receipt by the participant. P051-013 integrates review deadlines through P051-007/S020 bounded jobs, never a private timer. Before that integration, provide authorized manual review and explicit overdue/unavailable inspection without claiming proactive scheduling. Service/scheduler outages neither extend effects nor turn pending work into success.
Outcome evidence, not a success score. Build a bounded read model from authorized facts, distinguishing reports from confirmed harm and repeated reports of one incident from independent incidents. Preserve observation scope, window, policy/measurement revision, changes in reporting practices, evidence limits and participants who have not returned. Increased reporting may reflect better detection or trust; fewer reports or a short return time alone do not establish improvement. Missing follow-up stays unknown. Reviewer conclusions carry their basis and uncertainty; the system does not infer causality from these counts. Keep private cross-case evidence sealed/access-controlled and redact participant projections, notifications and traces.

Necessary Dependencies Versus Lower-Rework Sequencing¶

The recommended implementation order is:

A → N → B → C → D → P051-010/011/012/014 → optionally P051-013.

The stage labels group the broader Responsibility follow-up and do not imply completion. Normalization and recovery have separately scoped implementation tasks P051-015/016 under their existing owners:

Stage	Scope / existing ownership	Why prefer it before the repair slice?
A	Admitted membership facts and runtime entry-policy integration under P051-001/004	Reuse real admission and effective access composition when evaluating return, rather than rebuilding that integration later.
N	Host-state normalization under S028 and the existing Responsibility/P018 host owners	Put new facts, authorization snapshots, source revisions and receipts into the target stores, instead of extending a format already scheduled for migration. N precedes B in this follow-up plan.
B	Local panel procedure under P051-006 and versioned P018/S040 effect lifecycle	Reuse revisioned suspension, resumption and terminal retraction without confusing a panel disposition with applied consequences.
C	Procedural deadlines under P051-007/S020	Establish the shared bounded-job mechanism; its availability does not itself implement scheduled repair reviews under P051-013.
D	Same-host backup, restore, reconciliation and compaction under S028 and the existing store owners	Reuse host-incarnation and receipt semantics, retention and recovery boundaries instead of qualifying a second interpretation later.

This is a cost-of-rework recommendation, not a new hard dependency chain for the manual repair slice. Its necessary task dependencies remain in the tracker: P051-010 → P051-011 → P051-012 → P051-014, using the qualified P051-002/003/005 and P018-13/14 baseline. Completion of all membership admission, independent panels, scheduling or restore work is not a universal prerequisite. N and D need separately scoped implementation/evidence work under their existing owners; the ordering label does not add them to a completed baseline task.

Perform a short P051-010 contract-alignment pass before N/B/D freeze their interfaces, without treating that pass as completion of P051-010 or starting the whole repair implementation. Record the selected contract revisions, owners, compatibility decisions and required regressions:

Alignment point	Contract to agree	Required preservation / test
N: durable review identity	Repair reviews have explicit identity, revision and their own deadline; facts refer to immutable authority snapshots and accepted ownership without granting a mandate. Map them to normalized stores, projections and outboxes using existing primitives.	Preserve signed source bytes, CAS, replay, idempotency and pending obligations across migration. Compare old/new projections at the same evaluation time; do not clone global authority state per repair command.
B: decision versus effect	Bind any repair-driven correction to the exact source and revision. Agree how `applied`, `suspended` and terminal `retracted` effects are inspected and corrected, independently of panel decisions and repair-plan status.	Correcting one source cannot clear another, resume a terminally retracted source, extend original validity or turn plan acceptance into execution authority. Preserve stale-command refusal and protected access.
D: execution evidence after restore	Bind new host execution receipts to the applicable incarnation. Historical evidence remains historical; current application requires explicit confirmation under the recovery contract. Include repair facts, evidence bindings, pending outboxes and live review obligations in snapshot/retention rules.	Restore preserves unfinished work without fabricating current application; compaction cannot remove live obligations, required authority evidence, idempotency or terminal fences. Extend existing D regressions rather than create a repair-only backup mechanism.

The preferred path completes A–D first, then refreshes P051-010 against the actual qualified baseline and implements the manual slice. An earlier insertion after A and N is reasonable if the B/D interfaces above have been agreed, unsupported operations remain explicitly unavailable, and later migration/regression work is recorded. Contract agreement is not evidence that B or D works; exclude unbuilt panel/restore behavior from that earlier profile. Do not run competing rewrites of the same signed-command, host-store and replay contracts in parallel.

Reuse whichever prerequisite interfaces are actually complete. Excluding the broader A–D work from the repair slice means not redoing or expanding it, not avoiding its existing interfaces. Keep manual review independently usable even after C lands. Integrating repair-specific scheduled jobs still belongs to P051-013 and needs its own runtime evidence. This sequencing does not merge the workstreams or promote local mechanism tests to real-participant readiness.

The next bounded Node plan is node:docs/development/RESPONSIBILITY-SAFE-MAINTENANCE-SHADOW-V3-PLAN-2026-10-19.md. It addresses pinned-generation progress under continuous v1 writes, F-04 foreign-incarnation history and F-05 reserved abandonment/verified capacity recovery, then separate P072 operator authority, S039 alert delivery and exact shadow comparison. Every stage is planned, not implemented by publication. Shadow results compare the same pinned boundary and instant; stale preparation cannot claim live parity. Only the two adopted quarantine identity classes may differ, each with per-address proof and a qualified alert. v1 remains authoritative, all D7 components remain false and no wider P051-006/007/008/015/016 scope closes. The current F-02 conservative CAS can still starve under continuous writes; its cost evidence does not prove progress or operator readiness.

The 2026-10-20 F-02 timing investigation is separately scoped to the internal owner fixture. Node's docs/audits/RESPONSIBILITY-F02-GATE-DIAGNOSIS-2026-10-20.md records the original 3.141-second batch-gate outlier and a new 2.135-second threshold failure, of which 2.134 seconds were measured in the interval surrounding COMMIT. Bounded wall/thread-CPU diagnostics, controlled gate-wait/delayed-COMMIT checks and the actual owner stack distinguish limited row work from synchronous WAL I/O. They do not identify the unique kernel, storage or scheduling cause of the historical outlier; later passes cannot supply that missing evidence. The coarse wall-time tail remains an open qualification question, not a real-time guarantee. At that checkpoint, test instrumentation did not change the existing one-second coarse regression guard, production SQLite policy or lifecycle authority. This evidence checkpoint closes neither pinned-generation progress, F-04/F-05 recovery, operator authorization, shadow parity nor D7.

F-02 Ordinary Regression and Pinned Latency Qualification¶

Accepted follow-up decision, 2026-10-20: separate the ordinary bounded-work regression from qualification of the complete admission-gate latency. This amends the test contract, not the historical observations above.

The ordinary owner regression requires one source-history replay, 2 * N conversions for N sources, and at most 100 clear/copy/verify rows per batch gate. Capture and application acquire two separate admission-gate holds; meter each complete hold and report their individual maxima, not a sum or a CAS-only subphase. The coarse guards require less than one second of thread CPU and less than one second of wall time excluding only the exact COMMIT interval from that same hold. Do not subtract independently selected maxima or remove other I/O, locking or scheduling time.
A separate pinned qualification probe retains the less-than-one-second full wall-time gate, including COMMIT. Its manifest must identify the source/image, fixture and volume/filesystem, SQLite durability/checkpoint policy, and competing workload. The threshold qualifies that declared I/O profile only; it is not a global library-test requirement or a portable real-time guarantee. Retain full wall, exact COMMIT wall and whole-gate CPU observations even when the ordinary regression passes.
Keep SQL durability, the authoritative v1 path and atomic source-set CAS unchanged. A slow COMMIT still delays real admission: subtracting its interval for an algorithmic regression does not repair that latency or qualify the pinned profile.
Controlled baseline, filesystem-synchronization and CPU-pressure experiments must use the same fixture and volume, retain their actual outcomes and state their changed conditions. Such comparisons can distinguish new observations; they cannot retrospectively establish the unique cause of the historical 3.141-second outlier or relabel any retained failed run.

The amendment is implemented in the current Node candidate. Twelve valid counterbalanced cost matrices, each covering 500, 1,000 and 8,192 sources on the same verified fixture volume, preserve the work bounds in all 36 cases. With four workers and paced 4 KiB writes at intervals of at least 100 microseconds, all six matrices meet the pinned full-wall guard: the no-owned-load baseline maximum is 14.0595 ms, the CPU-control maximum is 15.276917 ms, and the two filesystem-synchronization maxima are 446.586 ms and 212.271791 ms. The initial unpaced 64 KiB/four-worker collection reached each writer's configured 2 GiB byte budget without an I/O error. It is retained as incomplete, not counted among those valid matrices or described as disk exhaustion.

A separate eight-worker sensitivity cohort preserves the work bounds but refuses the full-wall qualification in one of its six matrices. Its filesystem-sync maxima are 559.688125 ms and 661.97975 ms, its CPU-control maximum is 22.469917 ms, and its pre-load baseline maximum is 13.381042 ms. The post-load baseline with no owned writer active reaches 1.040366792 seconds for 1,000 sources at request 14. Its exact COMMIT interval is 1.039272917 seconds with 0.541166 ms thread CPU; the complete application hold uses 1.633292 ms CPU. COMMIT begins 1.088387 seconds after the last preceding owned writer ends. Active owned writers therefore do not uniquely explain this observed tail; carryover and uncontrolled host conditions remain undistinguished. No kernel or filesystem cause is assigned, nor are the historical 3.141/2.135/2.074-second tails explained.

The balance is 11 of 12 full-wall matrices passing, one qualification refusal, despite the refusing Rust cost process exiting zero and all deterministic work bounds passing. The pinned I/O profile is not fully qualified. The audit and retained cohort inventory are owned by node:docs/audits/F01-F05-OWNER-WORK-AND-REPLAY-2026-10-20.md and node:docs/evidence/local-accountability/2026-10-20-owner-work-io/README.md. The final ordinary frozen-image library run is not green: 2,430 passed, two failed and 64 ignored in 747.64 seconds, with process exit 101. All F-01–F-05 Rust tests pass. The two failures are native Sensorium cases: retained helper execution observes zero frames instead of two, and a readiness deadline expires before revocation. Their unique causes remain unassigned. The relevant Sensorium and supervisor source blobs are unchanged by this slice; that does not prove the failures predate it. This ordinary run is not a clean pinned OS-profile qualification. Its separately evaluated F-02 trace also refuses full-wall qualification despite the passing Rust owner-work test: full-hold maxima for 500/1,000/8,192 sources are 0.115421875/2.030558125/4.253287458 seconds. The corresponding non-COMMIT wall maxima are 18.732041/329.495959/166.931084 ms, and whole-hold CPU maxima are 3.022167/5.802541/15.319 ms. The latter guards pass; the 1,000- and 8,192-source full-wall observations refuse. This ordinary run under uncontrolled host conditions is not another paired I/O matrix and must not be pooled with the twelve above. Raw snapshots remain retained and the unique causes are unassigned.

P051-006 stays todo; lifecycle F-04/F-05 maintenance, operator authorization, shadow parity and every D7 component remain open or false as already recorded.

Systemic-Repair Slice Acceptance¶

P051-014 qualifies a new local test-subject profile using the existing local-accountability runner/report/qualifier ownership below. Freeze the scope before execution; do not relabel historical review-v2 evidence. Exercise actual commands, service persistence, host enforcement, participant inspection and S039 delivery, not injected final states. Required cases include:

aggregate prevalence alone cannot alter a participant's restriction; repeated reports do not manufacture independent findings;
a protective action proceeds while systemic repair remains pending; a missed repair deadline does not renew the protective effect;
absent mandate, conflicted/unavailable owner or missing escalation destination produces an explicit refusal/blockage, not self-authorization;
plan acceptance, execution evidence and outcome review remain distinct; no receipt means no confirmed external effect;
safe-return correction removes only named effects while preserving unrelated case/operator sources and missing entry authority;
restart, duplicate commands, delayed receipts and unavailable dependencies preserve ownership, deadlines, protected access and truthful pending states;
legacy cases remain readable without invented review outcomes; unauthorized inspection cannot reveal private cross-case evidence;
rising report counts, non-returners and absent follow-up cannot manufacture an effectiveness verdict.

The first profile may qualify manual review with explicit overdue inspection. Scheduled reviews require P051-013's additional source-bound runtime evidence; do not claim them from the manual profile. Keep real-participant sanctions, independent appeal readiness and federation outside this slice. Mechanism acceptance does not establish that a social intervention actually reduces harm.

When the tested baseline includes N/B/D, P051-014 must additionally retain repair integration regressions for normalized persistence, revisioned effect lifecycle and restore/compaction of unfinished repair obligations. An earlier bounded profile must state which of these remain unimplemented; later qualification must use fresh evidence, not promote the earlier report to the expanded scope.

Cross-Component Acceptance: Local Accountability¶

P051 owns the semantic acceptance scenario and aggregate closure, not a new protocol. Node homes exist for the bounded local profile below; the wider P051-008 matrix remains open:

node:tools/acceptance/local-accountability/README.md and its runner for reproducible procedures, linked from the acceptance index;
node:docs/evidence/local-accountability/ for revision-bound results, linked from the evidence index;
a short node:docs/integration/ guide only if the implemented flow needs one, linked from its index and owning service README, not another tracker.

First prove one host with separately authorized participant roles and deterministic fixtures, not an LLM or physical multi-host dependency. Seed review-pool eligibility explicitly and exercise the selected selection policy; report that seeded identities prove separation of authority, not real-world human independence. Add a two-host Artifact Delivery correction profile only after local closure. Corpus/Room-assisted review is optional and must not become necessary to file an appeal.

Case	Required retained evidence
Signal, review, decision, notice, appeal, correction	Each transition names actor, mandate, policy, target and next owner; decision and observed enforcement are separate receipts.
Two independent restrictions, appeal of one	Corrected case disappears from the current effect; unrelated case and newcomer defaults remain. Duplicate or reordered correction cannot resurrect it.
Hard hold and soft penalty expire during outage	Current checks stop both expired effects; reconciliation after restart records the transition without renewal or fabricated consent.
Scheduler unavailable, then resumed	Persisted review/appeal/effect deadlines remain distinct. With no access or other trigger, no proactive notice is claimed; read-only inspection exposes overdue state without mutation. Admission enforces validity; resumed reconciliation is idempotent and does not renew effects.
Legacy restrictions at adapter activation	Pinned record/tombstone inventory, explicit legacy sources, all-writer fencing and same-time effective-value equivalence for valid sources; malformed/ambiguous inputs retain evidence and typed refusal, not a legacy fail-open result. Restart cannot lose a source or enable premature correction.
Malformed expiry in recovered restriction data	Recovery/admission exposes a typed validity failure; affected privileged admission is refused, not silently unblocked. Protected inspection/appeal remains usable; corruption is not a new misconduct finding.
Crash or middleware detachment between decision, effect and notification	Real channel-hosted service replay/reattachment converges to the same projection; retries revalidate authority and do not duplicate effects; pending work remains inspectable without fallback to operator authority.
Owner revoked or leaves with an open case	Stale approval/action is refused; accepted handoff preserves the original authors and deadlines, or exposes an unassigned blockage.
No independent pool or insufficient appeal quorum	`AppealBlocked` preserves prior findings as history; effects follow the explicit stay/expiry policy. Council remits/arranges independent review or records continued blockage, never automatic exoneration, guilt or renewal.
Optional mediation refused	Direct formal appeal remains available; refusal is not guilt and mediation is not mandatory.
Host operator or case owner is a party	Self-adjudication and controlled alternate identities are refused; an independently authorized recipient accepts durable handoff, or the case exposes blockage and escalation. Existing deadlines and effect policy are preserved.
Restricted participant with Room removal and quiet notifications	Direct authorized case inspection and bounded appeal remain usable; no cross-recipient evidence disclosure.
Correction crosses hosts	Recipient authenticates domain authority and target, records application or refusal, and exposes delivery/retention gaps; transport acknowledgement alone does not close the case.

The cross-host row belongs to P051-009; the other rows form the P051-008 local gate. The local profile must exercise actual case, enforcement and notification adapters, not merely unit reducers or manually imported synthetic final states. Reports name real/substituted boundaries, repository revisions, commands, failures and remaining limits. Passing it demonstrates a correction mechanism, not fairness of every community policy or measured reduction of the Ringelmann effect. Denial fixtures must retain M11's reason, review/expiry and appeal path, without claiming the wider reputation-validation programme has passed.

The bounded local slice freezes node:docs/contracts/local-accountability-report.v1.schema.json and its qualifier before the runner. The remaining P051-008 matrix must likewise freeze a new explicit profile before extending runtime qualification. Reuse the scope and source-retention patterns in node:tools/acceptance/story-013-qmail-task-pack/README.md and retain_report.py, not their qmail-specific profile, run identifiers or assertions. This is an acceptance artifact, not another federation protocol. At minimum:

reports declare profile/id, evidence/class, qualification/scope and qualification/exclusions; CLI arguments check these claims, not promote them;
each required case binds observed owner receipts, refusals and pending outcomes, rather than only asserting success booleans;
pin the tested Node commit/tree, scenario/policy/fixture digests and qualifier revision; retain exact source and original report bytes/digest immutably;
requalification preserves original execution scope: old, unit-only or local evidence cannot become a fresh or cross-host runtime run;
redacted exports retain verifiable bindings for their declared scope, or state which qualification they cannot preserve. Do not expose private evidence merely to obtain a public passing report.

Qualifier fixtures must reject missing/duplicate cases, substituted bindings, scope mismatches, mismatched source pins and historical-scope promotion. Its job is to verify the declared evidence contract, not automate the social truth of a judgment. Reuse existing atomic report persistence; extract genuinely shared retention helpers only where needed rather than cloning Story 013's runner.

Trade-offs and Failure Modes¶

Composition avoids parallel stores, transports and UI engines, but source-aware correction is still new domain work; existing timestamps and reason/ref do not supply it automatically.
Pinned case policy supports audit; fresh authority/expiry checks prevent replay from granting expired power. Preserve both contexts rather than choosing one.
Independent review costs time and available people. Bounded holds and explicit unresolved outcomes are safer than relaxing independence to force completion.
Multiple identities or models can share one controller. Unknown independence is not plural support; scoped evidence must not become an ambient surveillance graph.
Provisional actions can cause irreversible external harm. State the remaining repair obligation even after the current restriction has been removed.

Suggested Future Artifact Families¶

This proposal now freezes the first small membership and surface-policy artifact family. These are the immediate contracts:

membership-invitation.v1
membership-sponsorship.v1
membership-acceptance.v1
participant-entry-profile.v1
participant-effective-limits.v1
surface-access-policy.v1

The broader public adjudication and reputation family remains future work:

object-review-case.v1
object-flag-event.v1
review-round.v1
review-vote.v1
reputation-event.v1 or a specialization of reputation-signal.v1
reputation-mediation-request.v1
reputation-mediation-outcome.v1
reputation-appeal-request.v1
reputation-appeal-decision.v1
collusion-signal.v1

The preferred shape is append-only facts plus read models, not mutable "membership status" rows as the primary source of truth.

Resolved Design Decisions¶

Sponsor liability is direct and one-hop by default; further propagation is reserved for proven sponsor-ring behavior.
Contact attestation unlocks contactability and anti-spam affordances only; it is not legal identity and not public-trust eligibility.
Open self-enrollment is allowed only as low-trust entry with tighter limits and slower access to high-impact surfaces.
Newcomer limits are entry policy. participant-capability-limits.v1 remains sanction or restriction policy.
surface-access-policy.v1 is the policy-axis source of truth. participant-entry-profile.v1 and participant-effective-limits.v1 are computed read models.
Sponsorship uses templates and ordinal liability classes instead of numeric pseudo-precision.
Public-object state uses lifecycle, judgment qualifier, and withdrawal-reverts axes instead of one compound enum.
The first anti-collusion baselines are sponsorship velocity, co-flagging coherence, and closed-loop receipt detection.
Political, ideological, religious, and campaign content is allowed on explicit opt-in surfaces; non-consensual high fan-out persuasion is the abuse class.
Marketplace reputation must be evidence-backed through settled interactions, not self-dealing or closed-loop boosting.

Open Questions¶

What minimal diversity constraints should be required before artifact-level sentiment becomes author-level reputation?
Which implementation and audit profile should realize the remaining random selection mechanism? DIA-PANEL-SEL-001 already fixes in-scope eligibility, recusal, uniform drawing, veto and escalation; its VRF/entropy construction remains a hypothesis, not a reason to redesign those settled safeguards.
Which contamination heuristics should remain only heuristic signals, and which are strong enough to trigger procedural rollback by default?
Which parts of membership bootstrap belong to the local node wizard, and which should remain explicit social actions performed outside the node?
Which bootstrap and dispute artifacts should be portable across communities, and which must remain local to the group's policy surface?
Which first community policy supplies a sufficiently independent appeal pool and an escalation destination when the local operator is a party? Until selected, fixtures may demonstrate the contract but cannot certify live social readiness. The prohibition on self-adjudication is settled; only the independent recipient and operational availability of that path remain open.

Local implementation checkpoint: 2026-10-10¶

The bounded implementation decision is local case → restriction → targeted correction, for explicitly admitted test subjects on one host. P051 owns procedure and responsibility; P018/S040 own source composition and enforcement. The frozen contracts and Node ownership are recorded in node:docs/development/LOCAL-ACCOUNTABILITY-CONTRACT.md.

responsibility-core is pure. responsibility-service persists append-only SQLite facts, rebuildable projections and atomic effect/notice outboxes through S019/P080 channel_json. The host independently verifies signed command, actor, mandate, policy, object and case revision, mirrors accepted procedural facts and applies exact P018 sources under the shared writer gate. Transport registration does not create a mandate. Review, appeal and effect validity are separate axes; an appeal or AppealBlocked leaves the original effect expiry unchanged. Acknowledgment names the exact reasons and opens at least 14 days; without acknowledgment filing remains open, including before notification.

P018-12 settles compatibility: participant-capability-limits.v1 remains operator input. New participant-restriction-source.v1 facts preserve each case/decision/effect's operation scope and validity; hard blocks compose by union and soft factors by minimum. v1 export refuses lossy composition, and participant-wide DELETE refuses active or unresolved case sources. A pinned legacy baseline preserves operator records, tombstones and unknown authorship; malformed validity is retained and refuses affected privileged operations.

ParticipantSession UI, direct signed APIs and registered S039 actions all call the same service. Inspection, exact reasons acknowledgment and appeal remain available without a Room or notification. Private evidence uses a separate sealed host table and case-bound references. Bounded history, command, source, evidence and outbox budgets reserve work for participant rights and correction.

The retained Node runtime evidence at node:docs/evidence/local-accountability/2026-10-10-qualified-local/ uses the frozen local-accountability.test-subjects.v1 report and qualifier, with original tested source, policy, fixtures, receipts, refusals and pending outcomes. Review corrected the reporting claim to 18 runtime cases and two conformance bindings; the original bytes remain historical and do not qualify the corrected implementation. P051-002/003/005 are now done for local-accountability.test-subjects.review-v2 after fresh source-bound gates: 18 runtime cases plus two conformance bindings, 713 checks across 15 suites and 27 qualifier tests (one positive and 26 negative). Current original report and qualification are node:docs/evidence/local-accountability/2026-10-10-review/passage-3/report.json and qualification.json, with report digest sha256:4c16aa570954c29bb2cb1c45317f837c9dad563e9d8b2871ad2dbaea96d91b72. Findings and unqualified attempts remain in the parent review record. The scope is not the complete acceptance matrix below. Case effects remain disabled by default. Declared controller groups and explicit fixture mandates prove local contract behavior; they do not establish an independent social institution or authorize sanctions against real participants. Proposal status and normative adoption are unchanged. The local selector takes the first eligible independently declared actor; full DIA-PANEL-SEL-001 selection, appeal merits dispositions and procedural deadline escalation are unmodeled P051-006/007 work, not implied by the minimal contract.

Implementation Direction¶

The earlier bootstrap-first sequence remains useful for onboarding, but the pure policy kernel and restriction runtime now let the correction work proceed without waiting for a complete wizard, reputation engine or anti-collusion service.

Implementation Tracker¶

Statuses: todo, partial, done, deferred. done requires the named evidence and matching Node ledger scope; contract maturity, hard-MVP inclusion and runtime qualification remain separate. This review does not promote P051 to accepted or expand the current release scope. Implementation tasks must update the Node ledger, regenerate its view and reconcile docs/MVP.md when its covered scope changes.

Depends on lists necessary task gates. The recommended A–D sequence and early N/B/D contract alignment are defined in Necessary Dependencies Versus Lower-Rework Sequencing; they are not additional blanket completion prerequisites for P051-010–014.

ID	Work / owner	Status	Depends on	Completion gate
P051-001	Membership policy baseline / membership-policy-core	partial	Existing frozen family	The pure kernel now has a separate host-admitted procurement consumer. Six source-bound runtime cases qualify local candidacy/admission/revocation/expiry and own-entry UI; signed Agora carrier import remains a fixture, not publication/subscription proof. Monetary caps and the complete operation matrix remain open. Evidence: node:docs/evidence/local-accountability/2026-10-14-admitted-entry/passage-4/report.json. Do not duplicate the kernel or promote this subset to complete entry-policy qualification.
P051-002	First case/policy contract and ownership / Rust domain core boundary	done	Existing P051/S040 contracts	review-v2 freezes the scoped case/policy/mandate/command/effect contracts, three deadline axes, exact acknowledgment with 14-day floor, controller independence, retryable AppealBlocked and recipient-bound correction. New authority is domain-only; historical proof has separate host-bound narrowing/receipt transport. Canonical schemas, mirrors, Gate, negative fixtures and split P072 operations pass. Local first-eligible selection, absent merits dispositions and procedural escalation are explicitly bounded P051-006/007 work. Evidence: node:docs/evidence/local-accountability/2026-10-10-review/passage-3/report.json.
P051-003	Rust case core and durable middleware-hosted service / case domain	done	002	Pure responsibility-core and channel_json service with append-only SQLite facts, rebuildable projections, accepted handoff and atomic transition/intents. Ordered outboxes retain accepted narrowing through revocation, prepared evidence is preflighted and typed mirror desync fences later work without claiming success; reasons/listing survive, blocked mutations are explicit. Historical replay/transport and bound receipts have named regressions. Actual outage/restart is qualified; automatic cross-store desync repair/portable backup remain unqualified. Evidence: node:docs/evidence/local-accountability/2026-10-10-review/passage-3/host-tests.json and report.json.
P051-004	Targeted restriction and entry-policy integration / host-owned P018-S040 adapter	partial	001, 003, P018-12, P018-13	Targeted P018 host application/correction and independent expiry are implemented and locally qualified. A separately qualified admitted procurement subset now consumes the pure evaluator, composing with independent P018 sources. Full entry-policy admission, Agora publication/subscription and operation/value mappings still require P051-001. Evidence: node:docs/evidence/local-accountability/2026-10-14-admitted-entry/passage-4/report.json. No complete P051-004 claim.
P051-005	Participant notice, case inspection and appeal actions / case service consuming P057-S039	done	003	S039 recipient-bound subject/appeal/handoff notices, direct reasons/acknowledgment/appeal and ParticipantSession UI use the same service. Actual default host signer emits verified domain proofs; UI preserves owner refusal/unavailable status and only bound withdrawal receipts show retracted. A/B/C UI list and exact second-policy reasons remain available with profile disabled, subject removed and policy revoked. CSRF, stale actions, independent acknowledgment/handling, sealed evidence and protected reserves pass; no notification or Room prerequisite. Evidence: node:docs/evidence/local-accountability/2026-10-10-review/passage-3/report.json.
P051-006	Independent appeal and repair / case domain	todo	003, 004, 005, 015	Policy-bound pool with recusal/randomness evidence and applicable DIA-PANEL-SEL-001 safeguards; no self-review or reuse of original decision makers. Exercise AppealBlocked, explicit effect-stay policy and Council exits without a merits verdict from missing quorum; source-addressed correction survives restart. Agent/Corpus advice cannot decide sanctions. After the 2026-10-14 checkpoint-B correction, the Node checkpoint has twenty-three procedure tests (also on Rust 1.88), three signed-input tests, a complete public draw golden and an independent framing/PRNG reference for the explicitly experimental `local-panel.tai-fixtures.v1` profile; current normative PL requires ELL2. Panel size (3, 5 or 7), reserves, pool and contributor minima, the proposal limit and phase windows are admitted policy, not the fixed default 3+2. A rejected proposal never implies `Uphold`, each party has one veto for the whole selection with reserves covering every party, and seat capacity is checked before the draw. §10.3 partial redraw is absent, so recusal or inactivity can still block the appeal with a typed supplemental-draw reason. Drawing only among revealing nodes is a recorded profile deviation from §4.4.1. DIA-PANEL-SEL-001 §4.1/§4.3 now define nonce and `reveals` encoding in PL and EN. Eleven canonical draft panel/input/lifecycle contracts and historical COI mirrors have structural Gate coverage. Six case-journal and four host-input conformance tests cover atomic facts, original bytes, independent admission, replay and corruption under the existing storage/write owners. Draft lifecycle V0/V1 adds four set/request/receipt/readiness contracts, required-stay unavailability preserving accepted appeal time, host owner refusals before any write and all-false readiness tied to the current host incarnation. Unavailable-stay UI and procedure-only dispositions have owner/rendering conformance; see node:docs/audits/RESPONSIBILITY-LIFECYCLE-V0-V1-REVIEW-2026-10-14.md. V2 adds inactive version-3 migration preparation: pinned raw history/sequence/digest and incarnation, resumable copy/verify checkpoints, terminal tombstones, typed quarantine and projection rebuild. V1 sources/readers/writers/CAS/receipts remain authoritative and every D7 readiness component remains false. The storage-owner tests are conformance, not runtime suspension qualification; see node:docs/audits/RESPONSIBILITY-LIFECYCLE-V2-REVIEW-2026-10-14.md. The subsequent V2 review withdrew the absence-of-P1/P2 claim: F01 canonical numeric comparison and F06 transaction-local pin/checkpoint replies are corrected; the draft quarantine retains typed scope, stable address, original diagnostic and typed tombstones. Failed deserialization never uses raw operations to narrow. V4s needs exact equality outside the two adopted fail-closed identity classes (invalid-subject, source-key-mismatch), a per-address report and qualified operator alerts; unlisted differences block the switch. F02 admission-gate cost is addressed by the internal WAL worker, rolling roots and bounded CAS commits, with local 500/1000/8192-source conformance/performance evidence only (node:docs/audits/RESPONSIBILITY-LIFECYCLE-F02-2026-10-19.md). F04 foreign-incarnation maintenance, F05 abandonment/retention and public operator authorization remain open (node:docs/audits/RESPONSIBILITY-LIFECYCLE-V2-FINDINGS-2026-10-14.md). V3–V5 composition, writers, cutover and suspension runtime qualification remain open. This is conformance only: the journal and host input lane are internal, while endpoint/channel integration, supplementary orchestration, sealed VRF custody, protected panel UI, lifecycle application and runtime qualification remain open, so this row is not closed.
P051-007	Bounded deadlines, ownership and contamination inputs / domain jobs	todo	003	Persist separate review/appeal/effect deadlines; shared S020 Scheduler launches bounded case-service actions through existing dispatch, not a private timer loop; use S029 for deferred individual operations where needed. Host admission enforces current validity. Prove outage with no proactive-notice claim, read-only overdue inspection and idempotent resumed reconciliation. Unavailable owner, dropped notice or failed job never extend a hold or imply guilt. Sweep signals enter review, not direct sanctions.
P051-008	Local-accountability acceptance / cross-component harness	partial	004, 005, 006, 007, P018-14	Node-local v2 report/qualifier qualifies exactly P051-002/003/005 and P018-12/13/14: 18 runtime cases plus two conformance bindings, 713 checks/15 suites and 27 qualifier tests (one positive, 26 negative), source/binary-bound original bytes. Historical v1 and failed review attempts remain immutable. Full acceptance including P051-001/006/007 is open; no historical scope promotion. Evidence: node:docs/evidence/local-accountability/2026-10-10-review/passage-3/qualification.json.
P051-009	Remote correction acceptance / case Artifact Delivery acceptor	deferred	008	Two-host authenticated decision/correction delivery, local admission, duplicate/reorder/restart and unreachable-recipient evidence; bounded P081 causal/replication primitives only where needed, no global sanction propagation.
P051-010	Protection, reintegration and systemic-repair contracts / case domain	todo	002	Map section 27's norms to existing contracts and missing behavior; freeze separate claims, transition/authority table, applicability/deferral and legacy semantics. Record the N/B/D alignment decisions before their interface freeze, then refresh against the selected implementation baseline; this early pass alone does not complete 010. Pin schema/versioning, signature/replay compatibility, canonical schemas, mirrors and Schema Gate negative fixtures. No aggregate-to-individual authority, fixed prevalence threshold or closed repair repertoire.
P051-011	Durable structured review and repair / responsibility-core and responsibility-service	todo	010, 003, P018-13	Implement the bounded local lifecycle using current signed commands, accepted ownership/handoff, S028 facts and outboxes. Preserve independent protection, return review and systemic repair; reuse exact host receipts and source-targeted narrowing. Outcome read model distinguishes reports/findings, duplicates, non-returners and unknown follow-up. Prove refusal, restart/idempotency, legacy replay and unrelated-source preservation; do not implement a second enforcement or workflow engine.
P051-012	Repair inspection and participant actions / existing API, UI and S039	todo	011, 005	Existing signed API and ParticipantSession UI expose authorized plan/acceptance/execution/outcome actions, return conditions and manual review with explicit overdue/blocked status. Notices remain minimal and recipient-bound; queueing is not acceptance. Verify mandate/COI checks, stale actions, protected access and cross-case privacy. No new UI-owned state machine.
P051-013	Scheduled repair outcome reviews / existing domain jobs and S020	todo	007, 011, 012	Extend the shared scheduler-owned bounded jobs, not a private timer. Freshly check action authority and separate repair review from effect expiry. Retain runtime evidence for due review, scheduler/service outage, manual fallback and idempotent resumption; overdue work cannot renew restrictions or infer guilt. A manual-only acceptance report cannot satisfy this gate.
P051-014	Systemic-repair local acceptance / existing local-accountability harness	todo	011, 012, P018-14	Freeze a new explicit report/qualifier profile and run every Systemic-Repair Slice Acceptance case through real service, host, inspection and notice paths, including restart. Bind sources, policy, fixtures and receipts; reject incomplete/substituted evidence and historical-scope promotion. Include N/B/D repair integration regressions when present in the tested baseline; otherwise declare exclusions and required later requalification. Manual review is the first bounded scope; scheduled claims additionally require 013 evidence. No real-subject, federation or social-effectiveness qualification.
P051-015	Host-state normalization / S028 and existing Responsibility-P018 owners	partial	003, P018-13	Normalize projections and shared immutable authority material before panel facts. Preserve exact signed facts, source CAS, receipts and pending work; replay-gate resumable migration. Prove changed-record writes and subject/operation admission costs with a new N runtime report. Three runtime cases and two conformance bindings qualify N passage 4: node:docs/evidence/local-accountability/2026-10-14-normalization/passage-4/report.json. The original N report predates the internal panel-input lane and V2 inactive lifecycle preparation, which require separate qualification; V2 only adds storage-owner conformance with unchanged v1 enforcement (node:docs/audits/RESPONSIBILITY-LIFECYCLE-V2-REVIEW-2026-10-14.md); the A report predates N and cannot close this gate. Early 010 alignment confers no mandate and does not implement repair.
P051-016	Local maintenance, restore and retention / host and case-store owners	todo	015, 003, 007	Separate D scope after A–N–B–C: quiescent encrypted backup, fresh-directory restore with new incarnation, exact-byte reconciliation and retention preserving live appeals, pending outboxes, unfinished reviews and terminal fences. Safe manifests only in evidence; no distributed clone fencing or real-subject qualification.

2026-10-20 (P051-006, F-02 follow-up): the ordinary-regression/pinned-qualification contract is implemented with 36 bounded-work cases. One of twelve valid cost matrices refuses the pinned full-wall qualification. The ordinary frozen-image library ends with 2,430 passed, two native Sensorium failures of unassigned cause and 64 ignored; the F-01–F-05 Rust tests pass, but its separate latency qualifier also refuses. This does not change this tracker's statuses, lifecycle authority or D7 readiness.

The detailed enforcement work belongs to P018-12 through P018-14, linked from S040; P051 owns integration and social semantics. P057 and P081 are dependencies, not duplicate backlog owners: add tasks there only if implementation exposes a missing generic notification or causal primitive. Their existing completion claims do not cover this new consumer. Broader onboarding, sponsor-ring detection and public reputation projections remain separate follow-ups, not hidden acceptance prerequisites.

P051-010–014 are additive scope, not a reopening of P051-002/003/005 or P018-12–14. The manual local slice can proceed without completing the whole entry-policy, independent-appeal or scheduler backlog. Its dependency on existing enforcement does not qualify missing entry-policy integration. P051-013 separately owns proactive review scheduling; P051-014 does not close the broader P051-008 matrix. Update the Node ledger and affected generated views during implementation, with explicit evidence scope; this proposal-only change is not an implementation claim.

Next Actions¶

Perform the short P051-010 alignment pass before N/B/D freeze their contracts. Record shared ownership, revisions, compatibility and regression obligations; do not count this pass as the completed repair contract or implementation.
Prefer A → N → B → C → D for the broader follow-up: complete admission and runtime consumption under P051-001/004, normalize host state before panels, implement the local P051-006 procedure and P051-007 deadlines, then qualify same-host recovery. Scope N/D work separately under the existing owners. Local fixtures are not live sanction readiness.
Refresh P051-010 against the resulting baseline, then implement and qualify P051-011/012/014 as a separate manual slice. Earlier insertion after A/N is allowed under the contract-alignment and explicit-exclusion conditions above; it is not the default lower-rework path.
Add repair-specific scheduled reviews under P051-013 only with their own evidence. Extend P051-008's broader matrix through separate explicit profiles; preserve historical reports and keep P051-009 deferred. Neither sequence nor reuse merges these acceptance scopes.