QUESTION-268: What record of an organization's or an animal's conduct shows that one ordering prevails in the negotiation among its perspectives, which conflicts it wins, for how long, and whether it returns after interruption, and does potential measured on conflicts actually won give the same verdicts as potential read as a disposition?

Why it is a question

TERM-076 makes valence a regime of potential and reads potential through Frijda’s marks of control precedence (THEORY-055). Every consequence CLAIM-440 draws, and every defeater it names, needs potential to be measurable: which ranking wins the pairs on which the perspectives disagree, whether it keeps winning, and whether it returns after a perturbation.

Three candidate measures are in reach.

• Conflict record. For an organization, its decisions on the pairs where its perspectives disagree, with the perspective each decision followed (for Wells Fargo, CASE-151, where the sales model’s ranking repeatedly overrode risk and HR). For an animal, choices in conflict tasks.
• Out-prediction. A ranking prevails when it predicts the system’s next moves better than its rivals, the evidential grade CLAIM-427 asks for.
• The coupled meaning–influence model. Past the instability threshold of CLAIM-330, one direction grows and mixes meaning with authority: an ordering coming to prevail, in the model’s own terms.

Three established tools are nearer.

• Resumption: attractor strength. Kuppens, Oravecz and Tuerlinckx’s DynAffect (LIT-1469, unread beyond the abstract) fits an “affective home base, a baseline attractor state around which affect fluctuates”, with fluctuations “tied back to the home base by the attractor strength”. Fitted to a different series, the share of conflict decisions that follow a candidate ranking, it would estimate that ranking’s resumption. Whether the model carries over is unchecked.
• Persistence: hysteresis. Lange and colleagues (LIT-1436, NOTE-1098) give a network of emotion components in which strong connections mean that “it becomes more difficult to switch back to a deactivated state”, and “In less strongly connected networks, hysteresis does not occur” (p. 460). Their network ranks nothing, so it has no conflicts to win and no analogue of priority.
• Which ranking: devaluation. Moors (LIT-1434, NOTE-1117) supplies the interventional test of goal-directed control: devalue an outcome, or degrade its contingency, and see whether conduct still follows the ranking that values it. A ranking that keeps winning after its outcome is devalued is not the one prevailing.

Two measures from emotion research are near but not the thing. Scherer’s synchronization of response components (LIT-1451, NOTE-1113) and Lewis’s predicted theta phase synchrony (LIT-1455, NOTE-1119) are correlates that potential would produce. Neither is signed, neither says which ranking is winning, and neither is given as a measure per episode. The component process model proposes two coherence measures, neither signed: an EEG gamma and beta contrast between felt and unfelt episodes (Grandjean, Sander and Scherer, LIT-1452, p. 489), and a function-on-function regression of EEG against EMG (Scherer and Moors, LIT-1457, p. 738). For organisms, the physiological split between intrinsic valence (facial EMG) and goal congruence (skin conductance and skin temperature; LIT-1457, p. 730) separates the two levels of valence CLAIM-433 distinguishes. The componential measurement literature since does not close the gap. Mohammadi and Vuilleumier (LIT-1443), Meuleman and Rudrauf (LIT-1440) and Meuleman et al. (LIT-1463) measure multi-componentiality as structure across a sample of ratings, not as coordination within an episode. Sander, Grandjean and Scherer (LIT-1471, NOTE-1103) say what synchronization should be in the brain and call outlining it “a major challenge” (p. 228); they also keep control precedence, “some action tendencies with control precedence take priority over other potential action tendencies” (p. 223), apart from synchronization, as this question does. The nearest to a per-episode measure is Mohammadi, Van De Ville and Vuilleumier’s latent processes (LIT-1444, NOTE-1107), defined for each 12-second event, but their scores rest on reported tendencies and are bipolar, so they cannot show “continue” and “stop” both high, the ambivalent case this question must separate from neutrality. Meuleman and Rudrauf’s virtual-reality paradigm, whose headset tracking the paper does not use, is the kind of setting in which a conflict record of enacted tendencies could be kept.

The second half of the question is the eraser. Priority, persistence and resumption are dispositional. If potential is read from conflicts actually won over an interval, a clamp that never acts cannot change it; if it is read as a disposition, THEORY-252’s argument applies to it.

When to read it. Kahneman, Wakker and Sarin (LIT-1448, NOTE-1105) separate decision utility from experienced utility and show later choices following remembered utility, which neglects duration and weights peaks and ends: “In many other situations the sovereign masters that determine what people do are not pleasure and pain, but memories of pleasure and pain” (p. 385). Potential during an episode must therefore be read from the conflicts won during it, not from later choices about it, which measure a different ordering, the remembered one.

A procedure: ordinal tests first, the Hodge decomposition for margins

With the rename to “potential” (TERM-076) the question has a candidate procedure. It was first drafted as a Hodge decomposition of the netted record; a review of the pull request found that this confused ordinal and cardinal structure and erased alternation, and the procedure below replaces it. Each step has a synthetic case in scratch/formal/valence.py.

1. Keep the record in time. For each contested pair, record which side each decision took, when, and in what context, the set of options then available. Cut the history into windows. Never net a pair’s contests over the whole history: a pair won alternately by each side nets to no conflict (CLAIM-434, part 5). Keep apart, for each pair and window, the direction of its wins, the number of contests and their variability.
2. Require conflict. A window with no contested pair on which the candidate orderings disagree has every ordering as its potential (part 8), and returns no verdict on valence.
3. Test existence ordinally. In each window, whether the wins are acyclic, and if not, the least fraction whose reversal makes them so, with the best-fitting orderings (TERM-077). Circular triads are a direct count of the conflicts no ordering absorbs.
4. Measure alignment ordinally. For each candidate ordering, each perspective’s and the maintenance ordering’s, the share of contested wins it agrees with, on the pairs where it and its rivals disagree. This, not coherence, is how far an ordering prevails (part 3). An ordering fixed in advance is scored on the contests at hand; one selected from the record, as a best-fitting ordering, is scored on contests held out from its selection, since in-sample agreement is inflated (part 11).
5. Report the class. The orderings that fit the record equally form a class (part 7). Narrow it by intervention: devaluation or perturbation that contests the pairs on which its members differ. Until then the verdict is about the class: the sign is the set of signs its members give the state against the reference, determinate only when that set is a singleton (part 10), and intensity is the range of its members’ agreements. Options never contested are not compared by the record at all; that the wins are acyclic says only that some complete ordering fits them (TERM-077).
6. Read persistence and ambivalence across windows. Persistence is one ordering staying in each window’s class; resumption, its return after a perturbation (TERM-078). Resumption is read off the record, as observed returns, unless the potential is exact: an approximate one is no Lyapunov function (part 9), and inferring return from it needs a drift condition on a cardinal potential under a stated model of the moves. Ambivalence is cycles within windows or a pair’s outcome reversing across them, more often than a null model allows that keeps the record’s graph of contested pairs and draws its orientations from orderings, and conditioned on context, so that a change in the options available is not read as ambivalence.
7. Only then, and only where margins are intensities, the Hodge decomposition. HodgeRank (Jiang, Lim, Yao and Ye, LIT-1439, NOTE-1099; THEORY-894) splits a flow of margins into a gradient, the least-squares global ranking, and a cyclic part. It measures cardinal inconsistency. Its gradient is not an ordinal potential and can reverse a comparison the wins settle (part 4), and its cyclic share is positive on transitive records with uneven margins. Its baseline for a transitive binary record, (n − 2)/3n, holds only for a complete tournament contesting every pair once (parts 2 and 6).

Three cautions from the readings remain. The Hodge shares depend on the weights given to pairs, on how repeated contests are aggregated, and on the norm (NOTE-1099). Their paper’s Table 1 reports a cyclicity ratio above 1, which orthogonality forbids, so a reported figure is not taken on trust. And no step separates a lexicographic override from a compensatory potential on a finite record (CLAIM-436).

What would count as an answer

A procedure that, applied to a record of conduct, returns which ranking prevails and how strongly, with a case where it separates ambivalence (no potential ranking, or split potential) from neutrality (potential with no movement), and a statement of whether it measures actual or dispositional potential.

What would not answer it

A measure of how strongly a system pursues one goal, without the conflicts it wins against the others.