Federated Work Continuity

This live case tracks a provisional question exposed by multi-workspace agent coordination:

Is Work an independently necessary higher-layer object, a perspective-bound view over existing objects, or only ordinary product vocabulary?

The candidate appeared when the observer changed from the developer managing repositories and Assignment records to the person responsible for one real outcome. From that position, execution location became secondary. The primary question became whether the whole body of work is advancing, where judgment is pending, and what may happen next.

Current result

candidate: Work
status: provisional
minimum closure: inconclusive
deletion test: analytic hypothesis only
fuse test: inconclusive
dogfood: founding Work Control implementation exists; comparative KFD-5 dogfood remains inconclusive

Since the first cut, Kungfu has implemented Initiative, Assignment, WorkRef, Portfolio, independent review, continuation, and Project Cut settlement, and its derived Primitive Catalog exposes Work with experimental admission. That is new implementation pressure, not qualification. The strongest counterevidence remains present in KFD: Initiative is an accepted software-domain Primitive for one continuing coordinated body of work. Work must therefore demonstrate a decision-relevant responsibility that cannot be represented by one Initiative, several related Initiatives, or an ordinary cut-bound query.

Candidate shape

The initial engineering hypothesis is:

workspace-qualified WorkRef
  + typed Assignment Graph
  + component cuts and proofs
  + declared observer and traversal policy
  + current action frontier
  = perspective-bound Work view

This shape preserves autonomous fact worlds. It does not create a central Work database, global clock, or global atomic Cut.

Case surfaces

Claim boundary

This case does not establish Work as a Primitive, allocate a KFD number, qualify the founding implementation, prove cross-domain transfer, or show that an independent Work entity is preferable to Initiative plus a derived graph view. Those are explicit open tests.

Genesis: From Cards to the Work That Matters

Initial observer

The initial observer was implementing local-first agent work management. The natural objects were repositories, workspaces, Initiative and Assignment records, parent-child cards, runtime Episodes, and Project Cuts.

From that position, the problem appeared to be better routing and a richer cross-workspace Assignment graph.

Perspective transformation

The observer changed to the person who initiated a group of related cards. That person did not primarily care which repository contained each card. The consequence-bearing question was:

What is the state of the one thing these cards are jointly advancing?

The relation graph remained necessary, but became implementation structure beneath a possible user-level object: Work.

Why genesis is not qualification

KFD already contains a strong adjacent object. Initiative preserves one continuing coordinated software-work context across Pursuits, Assignments, participants, lineage, and settlement. The new observation may therefore be:

a newly visible higher-layer Primitive;
a useful federated projection over one or more Initiatives;
a product-level name for Initiative;
or no new object at all.

The live case freezes the question before implementation and dogfood can make the outcome look inevitable.

Successor evidence boundary

The first cut consisted of situated maintainer judgment, existing public KFD Initiative/Assignment/Project Cut cuts, and public Kungfu workspace and WorkRef contracts. The 2026-08-03 successor adds Kungfu’s current Work Control Profile and experimental Work catalog entry at exact commit a2967f22d0d23766eb887168e8fffc41251bad54. The implementation does not supply a fixed-root comparison against Initiative, derived view, Portfolio, Profile vocabulary, and no-new-Primitive outcomes; it also supplies no independent review or non-software transfer case.

KFD Method Trace

KFD-1: preserve the unresolved cut

The candidate, alternatives, current evidence, and missing qualification are published before implementation can rewrite the story into a success narrative.

KFD-2: keep confidence bounded

The current claim is only that Work deserves explicit investigation. It is not an accepted Primitive, a numbered KFD, or a qualified product capability.

KFD-3: preserve participant value

The product motivation is to let humans and agents manage the same work without forcing either to learn repository topology or hidden workflow rules. Exact authority and lower objects remain available rather than being concealed by a simple user view.

KFD-4: change the consequence-bearing observer

The developer saw repositories, workspaces, cards, and graph edges. The person responsible for the outcome saw one body of work and asked for its state and next action. That transformation generated the candidate.

KFD-5: retain no-new-Primitive

The same transformation may merely reveal Initiative or a derived query from a new angle. KFD-5 therefore requires a shared evidence cut comparing an independent Work object, a derived Work view, Initiative, and no new Primitive. The 2026-08-03 successor adds Work Control and Primitive Catalog implementation evidence while preserving every alternative and the provisional result.

KFD-6 boundary

No autonomous discovery claim is made. Human situated judgment supplied the perspective change; agents helped reconstruct adjacent contracts and explicit falsifiers. Future Episodes may support later replay and qualification.

Ontology Alternatives

The current question is not whether users benefit from a Work view. It is which object, if any, must own its semantics.

Existing layers

Fact and Episode
  -> Pursuit, Atlas, Warrant
  -> Initiative and Assignment
  -> Claim, Assessment, Decision, Admission
  -> Project Cut

Candidate boundary

Work would answer:

Which coordinated whole is this observer advancing across independently owned
workspaces, and what is its current action frontier?

It would not answer the narrower questions already owned by the objects above.

Three live hypotheses

Hypothesis	Identity	Authority	Main risk
Independent Work	Stable Work identity and versions	Own declaration only; component authority remains external	Duplicates Initiative or becomes a global database
Derived Work view	Root WorkRef plus observer, relation policy, and component cuts	No new semantic writer	Too unstable to deserve Primitive status
No new Primitive	Initiative and Assignment Graph	Existing owners only	Leaves material reconstruction burden hidden

Forbidden fusion

Even if Work qualifies, it may not absorb:

Pursuit direction;
Atlas perspective and freshness;
Warrant authority;
Episode occurrence;
Assignment responsibility;
Assessment or Decision;
Project Cut settlement;
workspace-local Fact authority.

The candidate is valuable only if it coordinates these boundaries while keeping them independently inspectable.

Conditional Distinguishability

Deletion experiment

Hold fixed every component workspace cut, Initiative, Assignment, typed edge, Episode, Claim, Assessment, Decision, and Project Cut. Remove only the proposed Work identity or view.

The deletion matters only if a participant must materially reconstruct one or more of:

which component objects jointly advance one declared whole;
which pending judgments block that whole;
which degraded or unavailable components change safe continuation;
which exact next actions remain available across authority boundaries.

An attractive dashboard is not enough. The deletion must change a bounded decision or impose measurable reconstruction cost.

Initiative comparison

The primary falsifier is:

one Initiative + typed Assignment relations + ordinary query
  == proposed Work semantics

If this equivalence holds for consequential cases at lower total cost, Work is subsumed or rejected. A Work view spanning several Initiatives does not by itself defeat the alternative; an Initiative-of-Initiatives or Portfolio query may still be equivalent.

Fuse experiment

Potential evidence for separation would require cases where:

one Initiative participates in more than one Work view without changing its own identity;
one Work view spans several Initiatives whose lifecycles remain independent;
changing Work membership changes coordination or settlement scope while all Initiative and Assignment roots remain fixed;
two observers derive different valid Work views from the same component facts and preserve the difference rather than merging it.

Kungfu’s current Work Control and Portfolio implementation makes this test executable, but no retained fixed-root result currently satisfies it.

Current verdict

The candidate is analytically distinguishable enough to incubate and now has founding implementation pressure, but it is not qualified. Minimum closure, fuse, deletion, independent implementation value, and cross-domain transfer remain inconclusive.

Propagation Boundary

If Work qualifies, it may lower adoption friction by letting a participant start from one understandable question while agents traverse the exact lower contracts:

show this Work
  -> inspect component authority and current frontier
  -> act on one exact owning Assignment
  -> admit evidence and settlement locally
  -> refresh the federated view

This could let products expose KFD-shaped responsibility without requiring a new user to learn every lower Primitive first.

The hypothesis remains unverified. A simpler Initiative or Portfolio view may provide the same value. Hidden routing, forced centralization, unexplained aggregation, or completion inferred from child activity would contradict KFD-3 even if the interface appeared convenient.