Pursuit and Warrant

Pursuit and Warrant

This live case began with the fused working title Proof-Carrying Work Object. The immutable genesis remains preserved under that name. A later ontology split now tracks two provisional Primitive candidates:

• Pursuit: the durable identity of an intended reality change across revisions, actors, actions, and evidence;
• Warrant: a purpose-bound, fact-addressed authorization for bounded continuation.

Both names, contracts, and Primitive claims remain open.

Why the case split

Pursuit persists and may receive a successor intent_root.
Warrant may be issued, exercised, attenuated, revoked, expired, or superseded.

A Pursuit can remain active without a valid Warrant. A Warrant can expire without deleting the Pursuit. Fusing them would let mutable work scope alter authority or make authority lifecycle events overwrite work identity.

The split also made two adjacent independent objects explicit:

See Ontology split for definitions, typed relations, invalid compressions, confidence boundaries, and qualification requirements. See the conditional distinguishability argument for the current deletion proof obligations and their explicit non-claims.

Current status

case status: active
genesis: preserved as Proof-Carrying Work Object

Pursuit:
  minimum closure: inconclusive
  deletion test: conditional analytic witness; reality evidence pending
  fuse test: inconclusive
  dogfood: not run as a common contract

Warrant:
  minimum closure: inconclusive
  deletion test: conditional analytic witness; reality evidence pending
  fuse test: inconclusive
  dogfood: not run as a common contract

The machine registry preserves one stable discovery-case identity and exposes separate current KFD-5 cuts for both candidate tracks. Later assessments create new numbered cuts for the affected track; they do not rewrite the genesis or the other candidate.

KFD-7 has separately reached active status with exact Buildchain and Kungfu Profile evidence for action-role independence, session round-trip, complexity-breakpoint behavior, and cross-domain transfer. That evidence supports the active action principle but does not complete the KFD-5 qualification of Pursuit or Warrant. Their candidate-specific deletion, minimum-closure, fuse, and comparative-value evidence remains pending.

Why this is a reflexive KFD case

This candidate appeared while KFD was applied to its own propagation problem:

1. KFD-3 reframed propagation as earning cooperation through trusted value rather than persuading or pressuring participants.
2. KFD-4 changed the observer from a human organization adopting a standard to an agent executing and handing off work.
3. The new view exposed a propagation loop based on discoverable tools, verifiable outputs, downstream acceptance, and future routing.
4. KFD-5 first separated the loop from the object repeatedly bearing its cross-boundary burden.
5. Continued deletion and responsibility analysis then showed that the proposed carrier had fused intended-change continuity with continuation authority.

The revised hypothesis is that KFD may spread operationally when agents work through stable Pursuits and fact-addressed Warrants whose value can be inspected by downstream participants. This is a hypothesis, not self-proof.

Claim boundary

This case does not claim:

• that either name or minimum closure is final;
• that either candidate is historically novel or universally applicable;
• that every agent interaction requires all four objects;
• that Atlas Go proves a universal Pursuit contract;
• that transient approval, capability, credential, mandate, or policy systems cannot already implement Warrant behavior;
• that the four-object model is a complete settlement model;
• that schema validity or KFD origin proves Primitive status;
• that agents will prefer producers of these objects;
• that operational use will necessarily reproduce KFD behavior;
• that a producer may certify its own work or transfer unlimited authority.

Pursuit names an intention-continuity hypothesis, not all performed work. Warrant names an authorization hypothesis, not enforcement, absolute truth, or universal local permission.

Case surfaces

• Genesis
• Ontology split
• Conditional distinguishability argument
• KFD method trace
• Propagation hypothesis
• Genesis KFD-5 cut
• Current Pursuit cut
• Current Warrant cut
• Qualification reviews
• Live case registry

Qualification direction

Pursuit becomes stronger only if stable identity and successor intent roots survive non-isomorphic work profiles and reduce recovery or scope drift. Warrant becomes stronger only if exact subject, purpose, basis, authority, and lifecycle reduce repeated authorization reconstruction without creating ambient authority or approval bureaucracy.

Evidence against either track remains visible and may independently result in rename, rejection, subsumption, or no-new-primitive.

Genesis: Proof-Carrying Work Object

This document freezes the first public reconstruction of the candidate’s genesis. It preserves the initial uncertainty and must not be rewritten into a later success narrative. Factual errata may be appended with an explicit date.

Initial question

The inquiry began with KFD propagation. The original observer was a human organization deciding whether to adopt a philosophy or standard:

company understands KFD
  -> management approves
  -> team adopts
  -> engineers integrate

Under this view, propagation appeared to depend mainly on branding, commercial communication, governance, and organizational persuasion.

Perspective transformation

The observer was changed from the adopting organization to the agent performing work and handing it to another participant:

tool becomes discoverable
  -> agent invokes it
  -> verifiable output is produced
  -> downstream review cost falls
  -> the output passes a gate more easily
  -> success enters memory and routing
  -> similar tools receive more future selection weight

From this position, tool discovery, machine protocols, authorization, supply-chain evidence, memory, routing, and downstream gates no longer appeared only as separate product categories. They appeared as stages in one possible work-propagation loop.

Compression from chain to object

The propagation chain was not itself the strongest Primitive candidate. The repeated burden fell on the object crossing each boundary.

That object had to carry more than an output:

• the work result;
• source facts and provenance;
• verification state;
• responsibility and claim boundaries;
• residual risk and degraded state;
• purpose-bound conditions for continuation;
• machine semantics for downstream judgment.

This produced the working candidate Proof-Carrying Work Object.

Initial examples

The candidate became visible across examples that currently belong to different domains:

The transaction mandate remains a hypothetical comparison in this evidence cut. The table does not establish a shared contract.

Initial Primitive shape

identity:
  one bounded work claim and its continuation basis

boundary:
  result + fact cut + evidence + responsibility + residual risk
  + continuation conditions

authority:
  producer declares; verifier assesses; downstream participant decides
  under its own purpose and policy

lifecycle:
  proposed -> produced -> assessed -> accepted or rejected
           -> authorized -> consumed -> stale, revoked, or superseded

operations:
  inspect, verify, accept, reject, continue, delegate, attenuate,
  revoke, replay, supersede

Alternatives visible at genesis

The candidate may be:

• only a common envelope over attestations, credentials, capability tokens, workflow artifacts, event logs, and signed records;
• a composition pattern rather than an independent Primitive;
• a useful local abstraction whose semantics do not transfer across domains;
• the wrong boundary around several narrower objects;
• unnecessary if downstream participants still need full reinvestigation.

No new Primitive remains a required alternative.

Initial falsifiers

The candidate weakens or fails if:

• there are no reusable operations beyond similar metadata fields;
• downstream participants do not measurably reduce reinvestigation;
• evidence cannot be safely reassessed for a new purpose;
• continuation authority cannot be attenuated, revoked, or delegated safely;
• a common envelope erases natural domain boundaries;
• acceptance depends mainly on vendor defaults or brand rather than evidence;
• first-party examples cannot transfer to an independent context;
• the producer remains the sole verifier or promotion authority.

Evidence cut and known gaps

This genesis cut records a conceptual convergence on 2026-07-17. It does not contain a common implementation, cross-domain benchmark, independent adopter, measured routing preference, or evidence that operational use reproduces KFD behavior. Those absences are qualification requirements, not editorial gaps to be filled by stronger wording.

KFD Method Trace

This trace records how KFD contributed first to the fused candidate genesis and then to its split into Pursuit and Warrant. It does not claim that KFD uniquely caused either candidate or that either candidate proves KFD.

KFD-1: preserve the discovery cut

The genesis, ontology split, current candidate shapes, alternatives, falsifiers, and known gaps are stored as public repository facts. The original Proof-Carrying Work Object cut remains intact. Pursuit and Warrant begin new immutable qualification lines instead of editing the initial uncertainty away.

KFD-2: separate four claims

The case now keeps these claims independent:

C1  KFD-4 and KFD-5 materially contributed to the fused genesis.
C2  Deleting independent responsibilities justifies tracking Pursuit and
    Warrant separately.
C3  Pursuit can reduce cross-session intent and recovery reconstruction.
C4  Warrant can reduce authorization reconstruction without widening authority.
C5  Agents will prefer tools whose outputs and continuation are easier to
    inspect.
C6  Repeated use can reproduce KFD-compatible work structure without prior
    philosophical adoption.

The genesis record supports a bounded version of C1. The ontology split records the reasoning behind C2 but does not qualify either Primitive. C3 through C6 remain unverified.

KFD-3: propagation begins with trusted value

The original propagation problem assumed that humans first had to understand and accept KFD. KFD-3 changed the constraint: cooperation should be earned by discoverable, inspectable, trusted value rather than pressure or hidden workflow capture.

This opened the possibility that an agent could preserve the Pursuit it is continuing and present a Warrant that makes bounded continuation easier for downstream participants to inspect, without first adopting KFD as an explicit doctrine.

The mechanism would violate KFD-3 if routing were hidden, alternatives were suppressed, constraints were unreviewable, or past success were used to force future selection.

KFD-4: move to the executing agent’s position

Changing the observer from organizational management to the executing and downstream agents changed the natural objects:

This transformation exposed the possible propagation loop. It did not by itself identify the Primitive.

KFD-5: separate the loop, then split the fused object

Structural compression and reconstruction-pressure analysis asked what would have to be rebuilt if the chain lacked a durable carrier. The answer was the basis on which downstream participants decide whether to trust and continue work.

The chain describes system dynamics. The first compression produced an object with identity, boundary, authority, lifecycle, and operations. A second deletion test exposed that it still fused two different responsibilities:

Pursuit  intended-change identity and revision continuity
Warrant  fact-addressed and authority-bounded continuation

Atlas and Episode remained independently necessary as context and experienced fact. This is a stronger current boundary, not retrospective proof that the genesis was wrong or that both new candidates will qualify.

KFD-5 now prevents the discovery story from promoting its own split. Both tracks remain provisional until alternatives, falsifiers, dogfood, security, and independent transfer are examined.

KFD-6 boundary

No autonomous discovery claim is made. A later KFD-6 experiment could compare whether causal work corpora repeatedly expose this object or competing boundaries, but this case currently depends on direct human judgment and scalable agent reasoning under human-held promotion authority.

Reflexive dogfood claim

The live case demonstrates that KFD can preserve a candidate genesis and later split its ontology without erasing the earlier cut. It does not yet demonstrate that KFD reliably discovers valid Primitives, that Pursuit or Warrant is novel, or that the proposed work structure has high propagation power.

Ontology Split: Pursuit and Warrant

The genesis cut named one fused candidate, Proof-Carrying Work Object. Further analysis found that this boundary combined at least two independently meaningful objects. This document records the split without rewriting that genesis:

• Pursuit is the durable identity of an intended reality change across revisions, actors, actions, and evidence.
• Warrant is a purpose-bound, fact-addressed authorization for bounded continuation.

Both names and both Primitive claims remain provisional.

Why the fused object was unstable

The genesis candidate carried a proposed or completed work unit, facts, evidence, responsibility, risk, and continuation conditions. That shape mixed two lifecycles:

the intended change persists and may be revised
the permission to continue may be issued, exercised, revoked, or expire

A Pursuit can remain active while no valid Warrant exists. A Warrant can expire without deleting the Pursuit. Conversely, a valid Warrant does not define the Pursuit’s purpose or prove that the Pursuit is complete. Fusing these states would make scope revision silently alter authority or make authority loss erase the work identity.

Four-object work state

The split exposed a wider object model:

These objects are independent but intentionally composable. Independence means that each keeps its own identity, contract, authority, and lifecycle. It does not mean that useful work leaves them unrelated.

AgentWorkState =
  Pursuit@intent_root
  + Atlas@atlas_root/perspective/cut
  + Warrant@holder/validity/action-boundary
  + Episode@causal-head

This tuple is a provisional operational model, not a claim that every agent interaction requires all four objects.

Pursuit boundary

A Pursuit is not the task card, chat, session, plan, execution trace, or work result used to present it. Its proposed minimum closure is:

pursuit_ref
intent_root
purpose
desired change
scope
acceptance or settlement conditions
safety boundaries
typed parent, dependency, and successor relations

pursuit_ref preserves identity. intent_root binds one immutable specification cut. A material change to purpose, scope, acceptance, or safety creates a successor intent root; it must not retroactively enlarge an old Warrant.

The Atlas Go is the first observed profile motivating this candidate. A Go Card, registry row, dashboard entry, or provider-native goal is a projection or execution binding of that longer-lived object, not the object itself. This observation does not prove that all domains require the name Pursuit or the full Atlas Go vocabulary.

Warrant boundary

A Warrant does not define a Pursuit, carry all work history, or certify a result. Its proposed minimum closure is:

warrant_ref
subject_ref + exact revision root
purpose
issuer and recognized authority
holder
allowed and forbidden actions
fact and evidence basis
constraints, expiry, delegation, and attenuation
residual risk
lineage and lifecycle events

The conservative default is one Warrant for one exact Pursuit intent root and one bounded action set. A root Warrant may authorize derivation of narrower Warrants. A multi-Pursuit Warrant must bind an immutable subject set or an explicitly reviewable selector; newly created Pursuits must not silently expand old authority.

Imported Warrants are claims about remote authority until local policy accepts them or derives a local Warrant. A producer cannot self-certify trusted continuation unless a recognized prior authority explicitly permits that derivation.

Typed relations

The model is many-to-many at the storage layer, but relation types must not be collapsed into one generic covers edge:

Pursuit --contextualized_by/input/result--> Atlas
Warrant --authorizes--> Pursuit@intent_root or another exact subject
Warrant --based_on--> Atlas and prior Episode/assessment/passport roots
Episode --executed_under--> Warrant
Episode --contributes_to--> one primary and zero or more related Pursuits
Episode --observed_against/produces_context_for--> Atlas cuts

An Atlas normally gives a Pursuit its initial and successor context, but a candidate Pursuit may exist before an Atlas is compiled. An Episode may be captured before it is attributed to a Pursuit. A Warrant may authorize a release, transaction, artifact operation, or other exact subject without requiring a Pursuit.

Execution and settlement

The four objects clarify execution state but do not let an executing agent self-declare completion:

execution:
  Pursuit + Atlas + Warrant + Episode

settlement:
  Claim + Assessment + authorized Decision

Claims, assessments, decisions, passports, task charts, and project cuts may be derived procedures, relations, or carried artifacts over the four-object state. Their exact Primitive status is not decided here.

Invalid compressions

The model rejects these shortcuts:

• active Pursuit means valid Warrant;
• valid Warrant means an action is wise, successful, or complete;
• sealed Episode means Pursuit acceptance is satisfied;
• fresh Atlas means the participant is authorized to act;
• mutable task text may enlarge authority without a successor intent root;
• a provider session or goal owns the durable Pursuit identity;
• one fused work object may overwrite independent intent, knowledge, authorization, and experience lifecycles.

Prior art and independent convergence: BDI

The four-object model has an independent theory-first ancestor. The Belief-Desire-Intention line began from the philosophy of action (Bratman, Intention, Plans, and Practical Reason, 1987): intentions are not reducible to beliefs plus desires; they are persisting commitments that constrain later deliberation because deliberation is resource-bounded. Rao and Georgeff subsequently formalized BDI agents and connected the theory to executable systems. BDI architectures also reached real deployments and mature agent tooling.

The comparison is useful, but it does not yield a one-to-one mapping:

The missing direct counterpart for Warrant is informative. A BDI intention is the agent’s own commitment to act; a Warrant addresses whether a recognized authority permits bounded continuation. Multi-agent BDI research does not make BDI a single-owner theory, but the classical belief-desire-intention core does not itself make external authority, delegation, revocation, and evidential basis first-class attitudes.

Limits exposed by BDI’s adoption history also identify risks for this model:

• beliefs and plan libraries were hand-authored, reproducing the knowledge engineering bottleneck;
• formal semantics and executable implementations have a standing correspondence and verification burden;
• provenance, evidence cuts, content addressing, and external authority are not guaranteed by the three-attitude core;
• real deployments did not make BDI the dominant general substrate for cross-boundary agent work.

Two implications follow. First, an independent theory-first path and this pressure-first path converging on related commitment and represented-world dimensions raises prior confidence that those dimensions are not merely one organization’s taste. It does not prove the four-object model, its names, or its minimum closure. Second, BDI’s mixed record of formal maturity, real deployment, and limited general infrastructure adoption is a standing warning: an ontology does not transfer merely because it is coherent.

This mapping is a comparison instrument, not a lineage claim. The candidates were not derived from BDI, and every row remains open to stronger historical and empirical review.

The BDI literature also contributes reusable probes for the qualification plan: intention-reconsideration policies (bold versus cautious agents, Kinny and Georgeff 1991) probe when a Pursuit should retain or revise commitment. They do not determine Warrant expiry or renewal, which remains an authority policy. BDI maintenance goals – goals with no terminal acceptance state – probe the continuing-process boundary already listed as a required Pursuit counterexample.

Sources:

• Michael Bratman, Intention, Plans, and Practical Reason
• Anand Rao and Michael Georgeff, BDI Agents: From Theory to Practice

The separate conditional distinguishability argument now tests whether the information attributed to Pursuit, Atlas, Warrant, and Episode can be deleted without changing a required decision. It is stronger than analogy to prior art but still does not establish universal necessity.

Current confidence

The strongest current claim is structural:

Long-running real-world agent work appears to require an addressable continuity object even when users call it a task, case, incident, job, investigation, plan, or Go.

Confidence is lower that Pursuit is the final name, that its proposed fields are the universal minimum closure, or that Warrant must always bind a Pursuit. One-shot conversation, stateless tool use, ambient observation, and open-ended processes remain required counterexamples.

Qualification plan

The two tracks must be qualified separately and together:

1. Test Pursuit against non-isomorphic work profiles, including software work, incidents, exploration, personal matters, and continuing processes.
2. Test whether stable identity plus successor intent roots prevents scope and recovery drift across agents and providers.
3. Test Warrant against transient approval, capability, credential, mandate, and policy alternatives.
4. Measure whether exact subject and basis references reduce repeated approval reconstruction without creating authorization bureaucracy.
5. Test the four-object tuple under pause, resume, multi-agent, remote, revocation, stale-context, failure, and settlement conditions.
6. Preserve rejection, subsumption, rename, or no-new-primitive outcomes for either track independently.

Conditional Distinguishability Argument

This note asks a narrower question than whether Pursuit, Atlas, Warrant, and Episode form a universal or uniquely minimal ontology:

Does each proposed object preserve information that can change a required action or audit conclusion when the other three are held fixed?

The current answer is conditionally yes. The argument separates four information dimensions. It does not prove that every product needs four database entities, that these names are final, or that no rival representation can carry the same information more economically.

Decision semantics

Let a valid work history h contain typed objects and relations. For each object family X, let remove_X(h) erase objects of that family and their incident typed relations while preserving all other observable information.

Let D_Q(h) be the answer required for a declared decision question Q. Information carried by X is non-redundant for Q if two valid histories exist such that:

remove_X(h1) = remove_X(h2)
and
D_Q(h1) != D_Q(h2)

Any representation sufficient for Q must therefore preserve information equivalent to the deleted distinction. It need not preserve the proposed object name, file shape, storage boundary, or implementation.

The four current decision questions are:

Analytic witness pairs

These pairs are proof obligations expressed as minimal counterfactuals. They show conditional separability if both histories are valid in the target domain.

Pursuit

Hold the Atlas cut fixed and assume no Warrant or Episode. In h1, the continuing intended change is to repair a configuration defect. In h2, it is to redesign the configuration contract. Removing Pursuit makes the histories observationally equal, but the next plan, acceptance test, and safe scope differ. Pursuit information is therefore non-redundant for Q_P.

Atlas

Hold the Pursuit fixed and assume no Warrant or Episode. In h1, planning uses a source-code perspective before a production incident. In h2, it uses a user-observed cut after the incident. Removing Atlas makes the histories observationally equal, but the justified diagnosis and next investigation differ. Atlas information is therefore non-redundant for Q_A.

Warrant

Hold Pursuit and Atlas fixed and assume no Episode. In h1, the participant has a valid Warrant for the exact action. In h2, the Warrant is absent, expired, or revoked. Removing Warrant makes the histories observationally equal, but one permits execution while the other requires stopping or seeking authority. Warrant information is therefore non-redundant for Q_W.

Episode

Hold Pursuit, Atlas, and Warrant fixed. In h1, the authorized action occurred and produced a recorded consequence. In h2, it did not occur. Removing Episode makes the histories observationally equal, but retry, compensation, and audit conclusions differ. Episode information is therefore non-redundant for Q_E.

What the argument establishes

If a domain admits the witness histories and requires the declared decisions, then a system cannot safely erase the corresponding information dimension. This rules out a fused model only when fusion makes one dimension independently unaddressable, mutable, or unrecoverable.

It does not establish:

• that all four decision questions arise in every interaction;
• that the proposed objects are jointly sufficient for all work state;
• that four is the globally minimal number of objects;
• that a task, case, session, log, approval, capability, or other object cannot embed one or more equivalent dimensions;
• that analytic witnesses measure real-world burden or adoption value;
• that the candidates have passed KFD-5 qualification.

This distinction prevents circular reasoning. The test starts from externally meaningful decisions, not from an assumption that the four names must exist.

Conservative reduction does not erase distinction

Conditional distinguishability does not require every interaction to display four separate objects. Several information dimensions may be coextensive in a bounded history without becoming universally derivable from one another.

For example, a simple agent session may contain one local goal, one current context, one stable permission grant, one execution attempt, and only an input and result state. In that limit, a session-compatible view can project:

goal              <- Pursuit
context           <- Atlas
tool permissions  <- Warrant
run or transcript <- Episode
input and result  <- Fact cuts

This is valid compression if the projection preserves the decisions required for the bounded task and its assumptions remain inspectable. It is not evidence that the dimensions are identical. The stronger hypothesis is a conservative extension: preserve the low-cost session experience where the dimensions move together, then expose the independent roles only when work crosses goals, perspectives, authority states, Episodes, or material Fact branches.

This gives KFD-7 a two-sided burden. It must show both that separation changes real decisions in complex histories and that ordinary simple work does not pay permanent ceremony for distinctions that can be safely projected.

Evidence ladder

The passed states above refer only to the exact KFD-7 activation cut documented in docs/KFD-7-activation.md and evidence/kfd-7/activation-record.json. They activate an action-responsibility principle; they do not qualify Pursuit or Warrant as universal, minimal, or independently load-bearing Primitives. Those candidate claims remain governed by their own KFD-5 cuts and evidence.

Falsifiers

The four-object hypothesis weakens if:

• one dimension is always derivable from the others without loss;
• deleting a dimension never changes a safe action or audit conclusion;
• the proposed witness histories cannot coexist with the same retained state;
• a fused or rival model preserves the same decisions with less lifecycle, reconstruction, and governance burden;
• the proposed structure cannot recover the familiar session experience in its low-complexity limit;
• progressive disclosure cannot keep simple work simpler than the expanded complex-work representation;
• the dimensions cannot vary independently under real work;
• empirical profiles repeatedly omit a dimension without hidden replacement.

The next qualification step is therefore not another naming exercise. It is to preserve real witness histories, apply the deletion projection, and compare the resulting decisions and reconstruction cost.

Propagation Hypothesis

The reflexive hypothesis is:

KFD may spread operationally when agents preserve the Pursuits they continue and present fact-addressed Warrants whose bounded value lets downstream participants continue work without reconstructing intent and authority from scratch.

This is different from asking participants to read, believe, or endorse KFD.

Proposed loop

agent continues Pursuit@intent_root against Atlas@cut
  -> acts under a purpose-bound Warrant
  -> Episode preserves action and consequence
  -> downstream participant checks facts, authority, and local policy
  -> accepted or derived Warrant unlocks bounded continuation
  -> outcome, review cost, and successor state are recorded
  -> future routing can compare this path with alternatives
  -> successful producers may receive more selection weight

Products and APIs remain execution and distribution surfaces. The hypothesis is no longer that one fused work object crosses every boundary. Pursuit may carry continuity and routing identity, while Warrant carries bounded continuation authority. Atlas and Episode preserve the fact context and causal experience required to inspect that authority.

KFD structure carried through use

The proposed structure would operationalize:

KFD-1  stable fact cuts, identities, contracts, and provenance
KFD-2  bounded claims, assessments, responsibility, and residual risk
KFD-3  inspectable value, choices, constraints, and continuation conditions
KFD-4  declared observer, perspective, transformation, and loss when relevant
KFD-5  genesis and qualification when the work proposes a Primitive

A participant could therefore use KFD-shaped work relations before learning the KFD name. This would be structural reproduction, not proof of philosophical agreement.

Claim assessments

Required evidence

• non-isomorphic Pursuit profiles produced by more than one tool or domain;
• Warrant profiles compared with approval, capability, credential, mandate, policy decision, and passport alternatives;
• downstream intent and authorization reconstruction time, retries, and clarification measured against ordinary workflows;
• purpose-bound reassessment without treating old proof as universal;
• safe attenuation, delegation, revocation, staleness, and supersession;
• visible routing reasons, alternatives, confidence, and opt-out;
• at least one transfer outside first-party examples;
• negative cases where a plausible object is rejected;
• evidence that recipients preserve the KFD constraints rather than merely copying fields.

Cautionary precedent: FIPA

Agent standardization has been attempted before. FIPA standardized agent communication in the late 1990s through agent communication languages, speech-act performatives, interaction protocols, management specifications, reference implementation ecosystems such as JADE, and interoperability tests. It later joined the IEEE Computer Society. FIPA is currently inactive, while its specifications remain openly available.

FIPA’s own 1997 Agent Communication Language specification exposes a precise boundary. It defined formal mental-attitude semantics but made that model informative rather than normative because agreed compliance testing for those attitudes was unresolved. Implementations could interoperate at declared message surfaces without proving the internal semantic conditions that gave the vocabulary its strongest meaning.

Modern agent protocols do not generally require FIPA compatibility, and direct continuity between the generations appears limited. That observation is not a proof that FIPA failed or that no implementation lineage exists. FIPA achieved real tooling, deployments, and interoperability work. It is a caution that a machine-readable standard and a living implementation ecosystem can still lose load-bearing continuity when later work no longer depends on the same semantics.

The loop proposed above is deliberately inverse to FIPA’s order: working substrate first, agent adoption for self-interested reasons (lower reconstruction cost, higher gate passage), and the standard read out of running systems rather than legislated ahead of them. This remains a hypothesis, not a settled historical law. The case therefore adds a falsifier of its own, recorded below.

Sources:

• FIPA public specification archive
• FIPA 1997 Agent Communication Language specification
• JADE FIPA-compliant platform documentation

Falsifiers

The propagation hypothesis weakens or fails if:

• downstream review cost does not fall;
• agents cannot discover or interpret the object without private context;
• routing preference is dominated by hidden defaults, bundling, or brand;
• evidence accepted for one purpose is silently reused for another;
• authorization becomes ambient, non-revocable, or transferable beyond intent;
• recipients strip provenance, risk, perspective, or responsibility fields;
• Pursuit or Warrant creates more coordination burden than it removes;
• independent domains converge on incompatible identity or authority boundaries;
• the original fused object performs better than the split under real work;
• machine-conformant surfaces and citations grow while independent working substrates and downstream reliance do not.

Anti-self-certification

KFD produced the method trace and currently owns the case record. That makes the record inspectable; it does not make the propagation claim true.

The generator must not be the sole verifier. Promotion requires evidence whose failure is not identical to the KFD narrative, including real downstream behavior, independent review, rejected or subsumed tracks, security failures, and consequences that can contradict the hypothesis.

Qualification Reviews

This directory is the append-only human review path for the proof-carrying-work-object live case.

A review is required when:

• a candidate is split, fused, renamed, rejected, or subsumed;
• a new concrete domain profile is added;
• an alternative is rejected or subsumed;
• a falsifier is triggered or retired;
• a qualification test changes state;
• a deletion witness is added, contradicted, or promoted from analytic to observed evidence;
• a new immutable KFD-5 cut is proposed for either candidate track;
• the case is used in a product or adoption claim;
• the outcome may change from provisional.

Each review records:

1. the previous cut and digest;
2. new supporting and contradicting facts;
3. changed alternatives, tests, and residual risks;
4. verification responsibility;
5. decision responsibility;
6. the reason for retaining or changing the outcome;
7. the affected candidate track;
8. the path and digest of any new cut.

For a distinguishability claim, the review also records:

• the decision question whose answer changes;
• the two valid histories and the exact deletion projection;
• which retained observations are equal after deletion;
• whether the witness is analytic, observed, or cross-domain;
• whether a rival representation preserves the same distinction with lower burden.

Reviews do not rewrite prior cuts. A correction identifies the affected claim, preserves the old coordinate, and creates a new cut for the affected track when its current candidate state changes. Pursuit and Warrant may qualify, fail, rename, or be subsumed independently.