Purposeful Systems — why people do what they do on the floor.

In 1972 Russell Ackoff and Fred Emery published On Purposeful Systems. Their central distinction: a purposeful system can pursue different goals under constant conditions — it selects ends as well as means, not merely means to a fixed end. That is stronger than goal-seeking, which varies means toward a goal fixed by design. The distinction maps directly onto the floor.

A conveyor belt is goal-seeking: the means adapt, the goal is fixed. A loader is purposeful — they choose how they scan and whether to pursue the system's fidelity goal at all. The coordinator is ideal-seeking: zero missorts, perfect fidelity, optimal staffing — never fully attained, continuously shaping decisions. The table classifies behavioral capability, not job performance; "compliant / engaged / expert" are observations within that capability, not falsifiable system-type assignments. The taxonomy is an organizing lens, not a measurement instrument — stated plainly so unobservable typing is never presented as an empirical claim.

The central operational thesis of this branch: an individual loader's PPH is, in part, a measurement of the system acting through that person, so hub throughput is not simply the sum of isolated individual performances. The Emergence Gap is defined as the difference between actual throughput and the reductionist (sum-of-individuals) estimate:

It is essential to be honest about status. Because $\mathcal{E}$ is defined as actual minus reductionist, the statement "$\mathcal{E}\neq 0$" merely says the reductionist model has error — true of any imperfect model, not a derivable theorem. So the proposition that the gap is non-zero and operationally significant (large, sign-predictable from social conditions, inestimable from individual data alone) is stated as Hypothesis 6.1, requiring zone-level data the project does not yet have. Prior versions presented this as a theorem with a proof that cited a now-retracted compliance equilibrium and appealed to authority; that proof is withdrawn.

The one piece of this part that carries a valid proof is Proposition 9.1 (the reductionist fallacy): by the law of total variance, a model that also observes zone-level conditions weakly reduces RMSE versus one using individual data alone. Its honest scope is modest — it is the generic statement that any informative regressor weakly reduces error; it says nothing specific to emergence and holds only if the social covariate actually carries information about $\mathcal{E}$. It is not a deep result; it is the rigorous floor under the Emergence Principle, and the rest of the part is explicitly hypothesis.

Each loader's scanning-compliance choice is modelled as a single-agent decision under a fixed, exogenous supervisory monitoring rate $\bar{p}$. The binary effort has a precise contractual referent: $e_j=1$ means following OJS Loader Methods #30, #33, and #7 — scan the bay-door ULD before entering, scan every package ("scan one, load one"), and log on with one's own correct employee ID. A continuous-effort relaxation (effort read as a long-run compliance rate) yields a descriptive cost-ratio relation:

This is not a Nash equilibrium and not a game. The supervisor has no payoff function and no strategy space, and $\bar{p}$ is a fixed parameter — there is no second player to best-respond to. Prior versions (including v4.0-ds) labelled the closed form a "Nash equilibrium" / "mixed-strategy equilibrium"; that claim is retracted. The relation is offered only as a descriptive heuristic for an aggregated compliance rate: compliance rises with monitoring $\bar{p}$ and with the intervention-cost-to-friction ratio $\kappa_{\text{int}}/(s+d)$. The parameters are illustrative; no equilibrium is computed or verified, and the reading is directional, not a calibrated prediction.

The useful corollary is a design one, stated as a structural conjecture: OJS supervision plausibly acts on this ratio by lowering the discomfort term $d$ (its methods are engineered to minimize physical strain), so when methods are internalized, compliant scanning is less effortful and the heuristic expects higher compliance without raising monitoring. When OJS practice degrades, $d$ rises, and a FidelityScore drop then reflects worsening method internalization rather than worsening intent — indicating OJS refresh over more monitoring. This is consistent with OJS design intent, not a theorem; measuring $d$ directly is what would test it.

If the gap is real, what governs it? The framework proposes a scalar Zone Social Potential $\Phi_z(t)$ — a convex combination of norm level, visibility, feedback, skill-match, interaction rate, and (negatively) belt pressure, with literature-grounded prior weights summing to one:

$\Phi_z$ is high when conditions favor collective flow, low or negative when they suppress it. Hypothesis 8.1 proposes the emergence gap is a monotone-increasing function of $\Phi_z$ — positive premia above a flow threshold, negative cascades below a cascade threshold. This is explicitly a hypothesis: $g(\cdot)$ carries free parameters. Prior versions showed a "within 1% of prediction" worked example (predicted swing ≈ 600 vs observed 606), but the two premium knobs were chosen to hit the observed number — fitting two free parameters to one data point is curve-fitting, not validation, and that example is retained only as an illustration of the functional form.

The deployable part is humbler and real: $\Phi_z$'s six inputs map onto existing tracker fields, and three event types already in iGate/SOR history (a PPH drop without an induction decrease; a fidelity type-error cluster; a cross-zone correlation event) serve as implicit backtrack annotations with no new data collection. That implicit-$D_{back}$ construction — not the Φz field itself — is the paper's most directly shippable contribution. The operational takeaway survives as a strategy hypothesis: the coordinator's highest-leverage object is the zone condition, not the individual conversation.

v4.0 reconciles three prior documents (v3.26, the κ-conformance revision, and the DeepSeek v4.0-ds audit) into one authoritative Branch C head and finishes the demotions the earlier audit left undone. What changed: the compliance "Nash equilibrium" (Thm 2.3) is demoted to a descriptive single-agent cost-ratio heuristic, the "Nash"/"game" language retracted, and its four downstream dependents — the Emergence-Gap proof, the OJS interpretation, the OJS-recertification corollary, and the visibility prior-weight rationale — corrected to no longer claim equilibrium. The Brouwer fixed-point theorem (left in v4.0-ds) is removed, the category-theory section confirmed deleted, and the series-independence theorem kept only as a neutral note. Tautology-"theorems" are demoted to definitions / hypotheses / conjectures (Thm 2.1 → Definition; 6.1, 8.1, 10.1 → Hypotheses; 13.1 → Conjecture; Remark 14.1 → a conditional observation under an explicit accurate-scorer assumption). Proposition 9.1 is the one valid formal result, kept with honest scope.

Constants are pinned to metric_contract.json: γ = 0.982 ± 0.010 (full 254-sort corpus; supersedes the stale 0.958 / 0.938); κ_Z3 is the DOW vector on the SEAS basis (Mon 0.413 → Fri 0.322, full-year mean ≈ 0.368 ± 0.014), with the iGate basis ≈ 0.31–0.33 sitting lower at every phase by the CCHIL→PD-09 attribution rule (not a contradiction); ρ_PD/Hub = 0.509 ± 0.025; and Π_SOR ≈ 120.3 with cost field c = 1/Π. Statistical hygiene followed: the Hackman 35–45% figure is down-qualified to a directional prior (the precise value could not be verified), the norm-anchor "35% core" is flagged n=1 / untested with surnames removed, and the SOR-signal table is flagged n=3 directional.

Honestly disclosed open items: the Φz field, the coordinator MDP, the compliance model, and the norm-anchor predictions are all untested — coordinator intervention logs do not exist, so the action variable is unlogged and the borrowed convergence result's coverage precondition is unmet. A cross-artifact naming collision remains flagged for the constants owner (the retired 188.6 / 155 avgPPH series still appears as engine cold-start seeds under a different object name). The deployable content is the implicit-$D_{back}$ construction and Proposition 9.1; the rest is operational hypothesis pending the Tailscale instrumentation.