Consequential Settlement

Consequential Settlement

This live case retrospectively records the distributed Agent-led development, KFD-4 recognition, and KFD-5 qualification that produced the KFD-11 candidate procedure:

Claim -> Assessment -> Decision -> Admission

The important genesis fact is not that an agent drafted wording after a human supplied the model. On 2026-07-11, agents first decomposed trust into Claim, purpose-bound Assessment, and TrustReport; then composed those objects with Mission/Go and generated Decision as part of the work lifecycle. Subsequent Project Cut, independent review, and Action Loop dogfood made Decision and Admission independently consequential. The maintainer supplied the reality pressure and authorized the work, but did not specify these intermediate objects and reports only learning that they existed during the later KFD discussion.

The 2026-07-21 transformation was therefore not the one-time invention of all four nouns. The agent recognized an already functioning but distributed structure, connected Fact Admission to the trust and authority lifecycle, and compressed it into one cross-domain settlement procedure.

The maintainer then authorized the pre-stable Foundation Revision and the Field Responsibility Matrix. That authorization is the qualification and governance decision; it is not rewritten as candidate generation.

Early KFD-6 evidence

This development line is also an early retained natural pre-enactment of part of KFD-6. Agents generated ontology under real development pressure and later recognized it across retained causal evidence before separate human and repository admission. The case therefore supports KFD-6’s feasibility, but it does not pass KFD-6: human prompts still triggered and corrected the replay, the evidence cut was retrospective, and no bounded multi-method comparison, baseline, held-out transfer, or systematic false-candidate rejection ran.

The exact loop mapping and failed obligations are recorded in the development lineage.

Current settlement

Case surfaces

• Genesis
• Development lineage
• KFD method trace
• Distinguishability and qualification
• Responsibility split
• Propagation hypothesis
• Current KFD-5 cut
• Superseded first cut
• Qualification reviews
• Live case registry

Evidence boundary

A retained maintainer-held task transcript and maintainer testimony support the conversational attribution. Public Git history independently preserves the semantic sequence from Fact Admission and claim-triggered Assessment through Mission Control, independent review, recoverable settlement, KFD formalization, and pull-request review. Git can audit the public development sequence; it cannot independently reconstruct every utterance in a transcript that is not distributed with KFD.

This case does not claim historical novelty, universal minimality, KFD-11 activation, or that four physical records are required.

The first cut remains immutable as evidence of the initial, too-late genesis reconstruction. The second cut corrects the lineage without rewriting that earlier public fact.

Genesis: From Development Pressure to Consequential Settlement

This record separates the distributed development genesis from later KFD recognition and public qualification. It must not be rewritten to make maintainer pressure or authorization look like maintainer generation of the intermediate ontology.

The discovery began before KFD-11

On 2026-07-11, the maintainer asked how Kungfu should determine whether Agent work was trustworthy and later asked the agent to connect KFD, Fact Manager, ADR-0048, and Atlas Mission/Go into one product. Those prompts supplied the real problem and desired outcome. They did not specify Claim, Assessment, or Decision as product objects.

The agent first generated a trust decomposition:

Claim + purpose + pinned cut + policy -> Assessment -> TrustReport

It then generated the Mission Control chain and identity graph:

Mission -> Go -> Episode -> Fact query -> Assessment -> Decision

mission -> go -> run -> episode -> claim -> query -> assessment -> decision

Kungfu PRs #542, #545, and #585 preserve the public architecture and implementation sequence. Claim, Assessment, and Decision therefore predate the later KFD numbering question.

The later maintainer question

On 2026-07-21, the maintainer asked whether Claim, Assessment, and Decision needed to become KFDs. The maintainer reports that this was also the first time they became aware those named objects already existed. The question did not propose the three objects, Admission, a four-stage cross-domain procedure, or the conclusion that the visible nouns should not become three separate decisions.

A maintainer-held local task transcript retains the earlier exchanges, but it is not distributed in the public repository. Public Git proves the subsequent semantic and implementation sequence, not every conversational utterance.

The Agent’s retrospective transformation

By then, Project Cut settlement, independent review, and Action Loop recovery had made the responsibilities operational. Admission also already existed in the Fact architecture as the difference between a recorded observation and admitted state. The agent stopped treating the visible nouns as a numbering problem and recognized that these separately generated lines shared one procedure.

The compression was:

• Claim and Assessment were not missing KFDs; KFD-2 already owned them.
• Assessment could not authorize action; a Decision required Warrant.
• A valid Decision still could become stale, conflict, be denied, or fail to change the owning Fact authority.
• Therefore existing Fact Admission had to remain distinct from Decision in a cross-domain settlement model.

The candidate was compressed into one procedure:

Claim -> Assessment -> Decision -> Admission

This was Agent Primitive recognition and structural compression over an Agent-generated development ontology, not Agent verification of a model already supplied by the maintainer.

Later human and repository roles

The maintainer accepted the structural correction, authorized renumbering before stable, and requested the Field Responsibility Matrix. KFD PR #230 then made the candidate public through a formalizing commit that declares Agent: Codex, a machine schema, and a Foundation Revision. kungfu-origin independently reviewed the responsibility split and approved the change. PR #231 closed the public evidence coordinates.

These later actions qualify, authorize, review, and preserve the candidate. They do not transfer generation of Claim, Assessment, Decision, or their composition from the agents to the maintainer or reviewer.

The full dated evidence chain is preserved in the development lineage.

Counterfactual alternatives retained at genesis

• allocate separate KFDs to Claim, Assessment, and Decision;
• leave all three inside the software Domain Profile;
• treat an authorized Decision as proof of successful state change;
• append a generic settlement rule after the software-specific decisions;
• conclude that existing transaction or approval vocabulary already preserves every responsibility without a new KFD procedure.

KFD Method Trace

KFD-1: preserve genesis and qualification separately

The record keeps the retained private task-transcript evidence, maintainer testimony, dated public implementation coordinates, the formalizing commit’s Agent: Codex declaration, human authorization, and independent review as different facts. Later KFD recognition does not rewrite the earlier development sequence.

KFD-2: bound every claim to its evidence

Conversational attribution rests on a retained maintainer-held task transcript plus maintainer testimony. Public semantic chronology rests on Kungfu PRs 542, 545, 585, 958, 969, 980, and 1120. The cross-domain claim rests on the prior KFD-2/KFD-7/KFD-10 contracts, the KFD-11 schema, deletion and fuse witnesses, and repository review. None of these evidence classes silently proves the others.

KFD-3: preserve distinct participant value

The human supplied reality pressure, desired outcomes, authorization, dogfood, and final value judgment. Agents generated the Claim/Assessment/Decision intermediate model, composed it with Mission/Go, implemented it, and later connected it to Admission as a cross-domain procedure. Independent review supplied a separate challenge and verification boundary. Treating the maintainer as the source of the intermediate ontology would erase the collaboration that made the result possible.

KFD-4: change the consequence-bearing observer

KFD-4 operated more than once. The first transformations moved from raw Agent reports to the assessor, from Mission/Go coordinator to independent reviewer, and from process completion to successor-state authority. The final transformation moved from allocating KFD numbers to asking what must remain distinct before a successor Fact may exist.

These views made Claim, Assessment, Decision, and Admission visible at different times. The final cross-perspective replay exposed them as one procedure rather than four unrelated implementation details.

KFD-5: qualify the generated procedure

Qualification did not rely on origin prestige. It tested whether deleting or fusing a role changes consequential decisions, compared alternative allocations, published the contract and schema, retained falsifiers, and required independent review.

The current outcome is provisional. The procedure is strong enough to hold a numbered draft and guide implementations, but KFD has not yet established independent cross-domain adoption, universal minimality, acceptable cost, or the activation gate.

KFD-6: preserve the partial autonomous loop without certifying it

The lineage demonstrates real Agent ontology generation, causal development pressure, later replay, KFD-5 handoff, and separated promotion authority. It therefore supplies early feasibility evidence for KFD-6. It does not supply a predeclared experiment cut, plural method comparison, required baselines, held-out transfer, systematic false-candidate rejection, or an autonomous trigger. The case remains a natural pre-enactment, not a KFD-6 witness.

Method boundary

This case shows that agents can generate an intermediate ontology during real implementation and later recognize a Primitive by replaying and compressing that distributed work. It does not show that perspective transformation is always the best discovery method, that every Agent proposal should be promoted, or that human judgment becomes optional.

Responsibility Split

KFD-2 trust world
  Claim -> Assessment

KFD-10 authority geometry
  Warrant -> Decision authority

KFD-7 Fact-Episode Ontology
  requested effect -> Admission -> successor Fact

KFD-11 does not create a second Claim, Assessment, Warrant, Fact, or Episode authority. It names the procedure that carries one consequential proposition from purpose-bound trust judgment through authorized disposition to the independently recorded result of applying that disposition.

Domain Profiles own proposition kinds, assessment methods, disposition vocabulary, transaction technology, retry policy, and participant projection. They may not redefine a passing Assessment as authority or a valid Decision as successful Admission.

Distinguishability and Qualification

The candidate is one procedure with four independently variable responsibilities, not a claim that all four are newly discovered nouns.

Deletion witnesses

If Admission is removed, two histories can contain the same Claim, Assessment, Decision, and requested effect while only one has a valid successor Fact. The procedure therefore cannot stop at Decision.

Fuse tests

• Fusing Claim with evidence makes evidence state its own proposition.
• Fusing Assessment with Claim makes a statement certify itself.
• Fusing Assessment with Decision turns trust judgment into authority.
• Fusing Decision with Admission makes authorization self-executing and hides stale basis, conflicts, denial, write failure, and partial application.

Equivalent implementations may use one transaction, object, API, or screen if the distinctions remain independently inspectable and retry-safe.

Alternatives

KFD-2 already owns generic Claim and Assessment, so separate numbered KFDs for those nouns would duplicate authority. Leaving Decision and Admission inside software work would make the boundary appear domain-specific. Treating a database transaction, approval, merge, release, or mandate as sufficient is retained as an implementation possibility only when it preserves equivalent responsibility, evidence, and failure semantics.

Qualification result

The deletion and fuse tests pass at the contract level. The KFD repository publishes a schema and verifier checks, and PR #230 received independent review. First-party settlement and release workflows provide pressure but do not yet establish independent cross-domain adoption. The KFD-5 result therefore remains provisional.

Falsifiers

• Claim or Assessment cannot vary independently in a relevant adopter domain.
• Decision authority and Admission outcome can always be fused without losing stale, conflict, denial, failure, retry, or successor-state information.
• Existing portable transaction or authorization standards preserve the same responsibilities with lower total burden.
• Implementations cannot expose the distinctions without unacceptable cost or user friction.
• Cross-domain examples require incompatible meanings rather than one portable procedure.

Propagation Hypothesis

Agent work crosses repositories, tools, organizations, and authority systems. At each boundary, persuasive output is cheap but admitted consequence is not. The KFD-11 procedure may propagate because it lets downstream participants ask four stable questions without reconstructing the producer’s context:

What is claimed?
For what purpose was it assessed?
Who authorized which disposition?
What effect was actually admitted?

Release passports, review decisions, project cuts, mandates, regulated submissions, and other proof-carrying work objects may implement this procedure. This is a hypothesis, not evidence of adoption or superiority.

Qualification Reviews

• Kungfu PRs #542, #545, and #585: public evidence for Fact Admission, claim-triggered Assessment, and Mission Control Claim/Assessment implementation.
• Kungfu PRs #958, #969, #980, and #1120: public evidence for Project Cut authority, agent-first settlement, independent Decision, and recoverable successor-state settlement.
• KFD PR #230: public formalization whose commit declares Agent: Codex, Foundation Revision, KFD-11 schema, and independent kungfu-origin approval.
• KFD PR #231: evidence coordinate and review-closure repair.
• KFD PR #232: first live-case reconstruction, retained as a superseded cut rather than silently rewritten.

The PR reviews qualify public semantics and implementation. They do not independently prove conversational attribution in the retained but non-publicly-distributed task transcript.

Development Lineage

The settlement procedure did not originate in one KFD numbering conversation. It emerged across several Agent-led design and implementation episodes, then was recognized and compressed into KFD-11. This record separates the reality pressure supplied by the maintainer from the intermediate objects generated by agents.

1. Trust decomposition: 2026-07-11

The maintainer asked how Kungfu should assess continuously reported Agent work, when assessment should run, and whether it should remain off the recording hot path. The question did not supply Claim, purpose-bound Assessment, or a settlement lifecycle.

In a retained maintainer-held task transcript, the agent decomposed the problem as:

Claim + purpose + pinned cut + policy -> Assessment -> TrustReport

It then proposed claim-triggered assessment jobs, independent Assessment Episodes, and process/thread execution profiles. Public repository history preserves the resulting architecture:

• Kungfu PR #542 introduced domain Fact admission and historical interpretation.
• Kungfu PR #545 explicitly introduced claim-triggered KFD-2 assessment jobs.

The maintainer subsequently used the new term claim while asking how a workspace master should dispatch an assessor. That later use is adoption of the Agent-generated decomposition, not evidence that the maintainer supplied it.

2. Mission Control composition: 2026-07-11

Later that day, the maintainer asked the agent to connect KFD, Fact Manager, ADR-0048, and Atlas Mission/Go into one product. The requested outcome was clear: preserve Mission/Go over time, inspect drift and execution, and judge whether work was reasonably advancing. The prompt did not specify Completion Claim, Assessment, Decision, or their lifecycle.

The agent generated the operating chain:

Mission -> Go -> Episode -> admitted Facts -> query
  -> Assessment -> human or Agent Decision -> successor work

It also generated the first explicit work lifecycle containing CompletionClaimed, CompletionAssessed, and DecisionRequested, plus the identity graph:

mission -> go -> run -> episode -> claim -> query -> assessment -> decision

Kungfu PR #585 then made that model executable through Mission/Go Fact admission, proof-backed Mission progress assessment, and evidence-bound completion claims. This is the first public implementation coordinate showing that Claim and Assessment were not later KFD editorial inventions.

3. Independent Decision: 2026-07-15/2026-07-16

Project Cut and successor-agent dogfood exposed that an assessment result still did not determine what action should follow. The implementation therefore made review and continuation independent:

• Kungfu PR #958 defined the Project Cut authority boundary.
• Kungfu PR #969 added agent-first Project Cut settlement.
• Kungfu PR #980 added distinct-reviewer Completion Claim assessment, a purpose-bound TrustReport, bounded continuation, and human authority gates.

This development pressure turned Decision from a diagram label into a durable, independently responsible step. The claimant, assessor, and decision authority could now disagree without rewriting one another.

4. Recoverable effect and Admission: 2026-07-19

Admission had already existed separately in the Fact architecture: a recorded observation did not become admitted state merely because it was written or assessed. Action-loop dogfood then exposed the same boundary after a Decision.

Kungfu PR #1120 implemented a recoverable settlement bridge that seals the work Episode, refreshes successor Atlas and Fact authority, publishes and independently reviews a completion claim, records continuation, and closes Pursuit and Warrant transitions without treating process exit as completion.

This made the remaining distinction operationally visible:

authorized or recommended next action
  is not yet
successfully admitted successor state

5. KFD compression: 2026-07-21

Only after the software had accumulated these objects did the maintainer ask whether Claim, Assessment, and Decision required separate KFDs. The maintainer reports that this was also the first time they became aware that those named objects already existed in the implementation model.

The agent did not invent all four nouns at that moment. It recognized that three Agent-generated development lines had converged:

claim-triggered trust assessment
+ independent continuation decision
+ independently admitted successor state
= Claim -> Assessment -> Decision -> Admission

The KFD contribution was therefore retrospective Primitive recognition and cross-domain compression. KFD PR #230 formalized the procedure; KFD PR #232 opened this live qualification case.

6. KFD-6 early feasibility case

The complete lineage is a natural, human-assisted pre-enactment of part of the KFD-6 loop:

O_n: Fact Admission + KFD-2 trust execution + Mission/Go semantics
  -> Agent-generated Claim / Assessment / Decision
  -> implementation, review, failure, and dogfood evidence
  -> later Agent replay and compression with Admission
  -> provisional KFD-5 qualification
  -> separate maintainer authorization and repository admission
  -> O_(n+1): draft KFD-11

This is stronger than a hypothetical example because the intermediate ontology and implementation coordinates predate the later KFD interpretation. It shows that agents can generate and later recognize a candidate ontology from causal development pressure without the maintainer first naming its objects.

It is not a complete KFD-6 run. The maintainer initiated the original work and later demanded the deeper historical replay; the causal-experience cut was reconstructed rather than frozen before discovery; no plural methods shared a declared budget; fixed-ontology and no-new-Primitive baselines were absent; repository review was not held-out evaluation of the discovery method; and no autonomous trigger decided when ontology review should begin. These omissions remain experimental requirements rather than being repaired through retrospective narration.

Responsibility attribution

The result is collaborative, but the intermediate ontology was not specified by the maintainer. Human authorization and later understanding must not be rewritten as human candidate generation.

Evidence boundary

Public Git proves dates, public semantics, implementation sequence, tests, review, and merge. A retained maintainer-held task transcript plus maintainer testimony support conversational attribution. The transcript is not part of the public KFD package, so an independent public reader can audit the development sequence but cannot independently verify every private utterance.

This lineage does not prove historical novelty, universal minimality, or that the same agent identity generated every stage.