Why Selection Cannot See It

Insolvency, positive deviation, and a case (Nokia) in which market and evolutionary predictions pointed confidently in the wrong direction — the case that motivates the claim that the two instruments are complementary, not rival. Why selection does not refute this theory; it nests inside it.

0. Scope, and the rival to beat

Papers 1 and 2 gave the object and the agent-level mechanism. This paper gives the dynamics — how a divergent system moves over time — and meets the objection that could actually topple the program: that systems evolve toward greater efficiency, that inefficient systems are competed away, and that a theory predicting decay therefore contradicts the most established pattern we have. That rival is named here and answered on its own ground. The answer is not that selection is wrong: external selection can register realized outcomes at the system boundary but does not, by itself, recover the internal map–flow divergence that produced them, while internal managerial selection operates on declared proxies and can actively mismeasure that cause. Neither reaches into the gap — and a witnessed case shows how far the boundary reading and the internal reading can pull apart.

A note on register, because it governs how strong each claim is meant to be read. The defensible core of this paper is the gradual one: a system drifts toward inefficiency by its own ordinary, lawful operation — measurably, incrementally, comparably across more-mature and less-mature estates. That slow drift, not dramatic collapse, is the headline and the thing easiest to defend. Terminal collapse is the rare limiting case, fenced as conjecture throughout; it is where the drift can end, not what the theory chiefly asserts.

1. Where capability lives, and the two signs of deviation

[definition] Value — capability — lives in two places: in the architecture itself (the on-map structure, and the agents who follow the map and genuinely contribute through it) and in the positive deviants (off the map). In a healthy, low-gap system, most capability sits in the architecture. As the process loses solvency, capability migrates: the more the map ceases to produce, the more of the real production passes to the positive deviants — until, at the insolvency threshold (§2, the verifiable point where following the map produces nothing or harms), essentially all of it is off-map, and the share of successful output causally dependent on positive deviation approaches one.

[tendency — convergence in the limit] So the rough identity “capability = positive deviation” is a limit reached at insolvency, not a global fact — and even then it is a co-location, not a true equation. These are different ontological categories: the capability is the capacity to produce an outcome; the carrier is the agent, application, or routine embodying it; the positive deviation is the non-conformant behaviour. They increasingly come to rest in the same carrier as the system diverges, which is why the precise statement is the measurable one above — the share of output dependent on positive deviation approaches one — not that a capacity and a behaviour become literally identical. In a healthy system they are clearly distinct, and capability is mostly on-map. Everything Paper 2 established about the coherent agent — its invisibility, its head-down posture, its destruction by declaration — transfers to the positive deviant without re-proof, because in the divergent regime they are co-located in the same carrier.

[definition] Two signs of deviation. A deviant is any agent acting off the map, and both signs are maximizers (Paper 2). The positive deviant maximizes contribution — aligned to the unstated upstream principle — and is the migrating capability above. The negative deviant also acts off the map, but maximizes something on neither the map nor the principle: pure noise. Not every departure from the map is value-producing, so capability must never be read as “any off-map behaviour.”

[mechanism] How negative deviation arises — from contradiction. When the three layers disagree — principle, law, and regulation pointing different ways (Paper 2, §4) — the agent faces an openness, a superposition of admissible targets (Paper 2, §9). It may collapse that openness onto the map: maximize the principle, or the law, or the regulation, or an implicit rule — in every such case still following the map. Or it may, by the sheer collision of the contradiction, collapse onto something unrelated entirely — “the rules are not working, so I will do what I want.” That is the negative deviant: an off-map maximizer of an irrelevant target, operating precisely because the contradiction makes compliance impossible. This theory works on positive deviation — the contributor — and treats negative deviation only far enough to mark that it exists, that it is also a maximizer, and that it is a promising separate study; it is named here so the taxonomy is honest, not because the program rests on it.

2. The insolvency threshold

The dynamics are not only “collapse someday.” There is a nameable instant. As the gap widens the map stops being functional as an instruction set: an agent following it faithfully — explicit and implicit rules, end to end — produces nothing, or produces damage. The system is insolvent by compliance: failing not by accident but because the rules were obeyed. The surgical image is the intuition — if following the protocol with the scissors kills the patient every time, the protocol is not imperfect, it is insolvent, and only the surgeon’s deviation keeps anyone alive.

[falsifiable] Formally, and stated precisely so the feasible set is unambiguous: compliance insolvency occurs when no execution path satisfying all mandatory constraints can preserve the independently established upstream outcome. Optional rules, and locally contradictory ones, are excluded from the mandatory set explicitly; what remains is the set of constraints that genuinely must hold. This is mappable: take the mandatory constraints as sets and search for a path that satisfies them all and still preserves the outcome. If no such path exists, the system is insolvent — exactly, not impressionistically. (This is the precise form of “the emptiness of the coherent, producing subset” from Paper 2’s two modes.)

3. Two measurables, today

The threshold and the dependence it produces are reachable now, on estates that already exist — but with a caution: process mining is the discovery layer, not the whole method. It can classify conformant and non-conformant paths; establishing that a non-conformant path is positive deviation (rather than negative deviation, an approved-but-undocumented variant, or a data-quality artefact) takes two further layers: outcome comparison, and causal or counterfactual evidence that the non-conformant path actually preserved the outcome.

[falsifiable] The conformance counterfactual. Reconstruct the as-designed path and ask what it yields if followed faithfully. The prediction at a wide gap is zero output or harm. Falsified if compliant execution reliably produces normal output even where the gap is wide.

[falsifiable] Deviance-dependence. Measure the fraction of actual successful output flowing through non-conformant paths. As the gap widens this fraction climbs toward one. A system whose real production is almost entirely off-map is surviving purely on deviance — one step from the threshold, since deviance is unsupported and resident in specific agents who can leave. Falsified if output stays on conformant paths as estates age and tangle.

4. The maturity inversion (control-density without integration)

[tendency — falsifiable] Maturity is supposed to mean less dependence on heroics: more process, more control, more first-, second-, and third-party audits, multi-norm certification. The theory predicts the opposite — but the precise driver is not maturity in general, which bundles age, legacy density, audit intensity, certification and complexity, and which a well-integrated mature process could satisfy with less deviation. The driving variable is control-density without cross-level integration: as independently designed, insufficiently integrated controls accumulate, each adds a constraint to the declared map, the conjunction becomes impossible to satisfy by compliance, and successful output becomes increasingly dependent on non-conformant reconciliation. So the sharp claim is: the more a system accretes unintegrated controls, the higher its dependence on positive deviation. Falsified if systems with high control-density but strong cross-level integration show lower deviance-dependence than fragmented ones — which is the right test, and is no longer refuted by pointing to a genuinely well-integrated mature process. (“Maturity inversion” is kept as the name of the phenomenon for its counterintuitive force; the measured variable is control-density-without-integration.)

[Illustration — composite, anonymized] Two cases of one structure, blurred across industries so neither identifies a firm. A reconstituted-product line builds a sellable carton from a library of formulas because its input is bought on the spot market at the lowest price daily — only the residual volume and quality others reject — so it can never lock supply; each auditor checks a slice (a few components; sourcing across a deliberately complex supplier web), and none ever places side by side the only two numbers that would expose it — raw input bought against finished product sold. Follow every rule and no compliant unit can be produced sustainably. A short-measure dispense sells marginally less of a metered product than it bills (fuel is the recognizable form): in a firm small enough this is a known implicit rule that travels and is governable; in a larger firm the same rule does not travel upward, and whether it is benign in one frame or fraudulent in another, the firm above a certain size cannot even see it is operating on it. In both cases the conformant path yields zero; the output is entirely off-map. An auditor confirming each slice conforms certifies a system that is, in aggregate, insolvent and surviving on the very deviance the audits are blind to. More control can mean more insolvency, better hidden.

5. Capability is relative to a boundary — semantic windows

[definition] Capability is not absolute; it is the capacity to produce within a stated boundary. Inside the firm’s boundary the short-measure deviance is a capability; at society’s boundary it is fraud, and there clarity helps — revealing that the product is not what the label says serves the larger frame even as it destroys the smaller. The semantic window is the boundary at which a phenomenon is read — the scope within which “what is the real structure, and where are the capabilities” has a determinate answer. Map-vs-reality, capability-vs-deviance are each defined only relative to a chosen window; get the window wrong and the same deviation flips between capability and pathology. Choosing the window, and producing clarity of that window, is itself the structural-awareness act.

6. Inefficiency rarely kills the irreducible act — feasibility versus viability

[mechanism] A system can accrete inefficiency for a very long time and still perform its irreducible act, because that act underneath is usually trivial. Picture the real process at bottom as “write the name and the date on a list.” You can do that with twenty agents or twenty thousand, each re-entering the name on a different system, map upon map upon map, and the act still gets done — because all those maps are appearances, and the actual production is one entry on one list. Inefficiency is map-proliferation, and map-proliferation is largely unbounded. It can grow for years, because none of it touches the irreducible act. A system can be grotesquely inefficient for decades and still produce, because somewhere someone still writes the name on the list.

[definition] But “still produces” and “still alive” are two different things, and the distinction matters. Separate operational feasibility — whether the irreducible act can still be performed at all — from organizational viability — whether the surrounding organization can still finance and sustain the act. Inefficiency erodes the second long before it touches the first. So a process may remain operationally feasible while becoming economically or institutionally non-viable: the act could still be done, but liquidity is gone, customers have left, suppliers have withdrawn, a regulator has intervened, or competition has made the whole arrangement unsupportable.

[tendency — falsifiable] So the precise claim is narrower than “a system never dies of inefficiency.” It is this: operational inefficiency does not, by itself, make the irreducible act impossible. Operational death — the act becoming explicitly impossible (the insolvency threshold of §2, where no consistent route can perform it), or that impossibility forced on purpose (§8) — is one ending. Viability death — the organization losing the means to keep financing a still-feasible act — is another, and the more common one. The theory’s distinctive claim concerns the first: that capability is lost by indeterminacy and the decision it forces, not by the irreducible act becoming undoable. Falsified if the irreducible acts of failed systems are typically found to have become genuinely impossible through inefficiency alone, rather than abandoned while still performable.

7. The closed-system destiny

Call a system closed when no structural awareness is injected from outside its local equilibrium — the only intelligence acting on the map lives inside the same constraints that produced the gap. The three series’ findings then point the same way: the price of clarity rises with maturity (information theory); the capacity to pay it falls as the coherence bottleneck tightens (organizational theory); and the gap is actively widened by rational action and the sorting of contributors to the shadow (economics).

[tendency] Inside a closed system these do not balance — they compound in one direction, so the gap tends to widen by the system’s own ordinary operation, and cannot be closed from within the local attribution equilibrium, because the agents who would reallocate are themselves the widening force. This is the gradual, defensible claim: a lawful drift toward inefficiency, not a prophecy of ruin. [conjecture] In the rare limit, absent any corrective injection, the drift can end in collapse — the map can no longer govern the flow, the failure crosses from the representation into the flow, and the flow gives way, governed past usefulness by a representation that came loose from it. The collapse is the terminal edge of the drift, not its usual state, and is fenced as conjecture. The exit requires a corrective function sufficiently insulated from the local attribution equilibrium — whether institutionally exogenous or internally constituted (an enterprise-architecture function, a protected transformation office, a special inquiry, a ring-fenced team). What matters is the insulation and the mode of action, not the location: this is the same “external by mode, not by position” point Paper 2 makes. That corrective function must revalue before clarifying, because clarity that resolves ownership without first recovering value cuts the unowned, and the unowned is the load-bearing thing. Upstream restructuring and downstream deviance are the same intelligence in two modes: the downstream patch defers the drift but never accumulates (it dies with the carrier); the upstream restructure is the only thing that closes the gap, but it costs precisely the scarce coherence-capacity that is missing.

8. The death is a decision: forced clarity, the parachute, the two stages

Because the irreducible act survives unbounded inefficiency, real death almost never comes from decay alone. It comes from a decision. There is a recurrent move where, precisely because no one knows how the system actually works, a governance body decides to force the collapse — to pierce the map, inject brutal clarity, and let the process “burst by itself” so it can be re-architected from the rubble. The blunt practitioner name for it, recorded here as the witnessed phrase it is, is the decision to “make the rot float.” It is a structural decision taken in ignorance: I do not know how this works, but it cannot go on, so bring it down.

[mechanism] Two structural pressures shape that decision. First, the clock: a newly arrived executive is held to account, within the first year, for not having acted — so there is direct pressure to take the rationalization decision fast, before the structural awareness needed to take it well could exist. The incentive is to decide blind. Second, the parachute: the incoming agent often arrives with one, signalling an explicit mission of change — a regime change, in the expected direction. The parachute is a crude form of the insulation Paper 1’s injector needs (it absolves the blame-sink problem), but it buys insulation, not coherence. This is the dangerous case: external by position but not by mode — an insulated maximizer under time pressure, which fires the destiny early rather than averting it. Only the coherent injector (Paper 2) averts it.

[tendency — falsifiable] Hence the two-stage death. First a slow decay — internal, non-competitive, the gap widening, competitiveness eroding; then a sudden death — a decision, taken under the stress the decay produced, that kills the still-live capabilities that would have let the system continue. Market stress is real and is the coupling between the inner and outer frames, but the death comes from the decision, not the stress: the stress is what makes the decision bad, by forcing it fast and blind. Falsified if, in system deaths, the killed capabilities were already non-productive at the moment of the cut (ordinary selection) rather than still productive (this mechanism).

[mechanism] Why the decision is taken blind: capabilities do not communicate themselves. The decider sees only the map, of necessity, because the off-map structure that keeps the system solvent does not travel upward. A capability cannot announce itself: to declare “this works only because I do what the map forbids” is to collapse the very openness the capability lives in (Paper 2, §9) and to name oneself the rule-breaker — the positive deviant’s head-down posture is not modesty, it is self-preservation. Three structural reasons compound, none of them incompetence: authority and information are different scarcities, rarely co-located, so the seat that decides is often the least connected to the floor (in large estates the most senior information officer is frequently the person who knows the fewest people by name, and this worsens with time in the seat); the channel is vertical, so speaking past one’s level reads as betrayal and the truth is gated at every rung; and maturity turns the real account into “gossip,” since the load-bearing facts take the form of turf and “I do this because no one else will,” which the flooded upper level filters out as noise. So clarity reaches the decider already stripped of exactly the structure it would need — not by oversight, but by the same forces that opened the gap. The decision is blind because capabilities are structurally silent.

9. The rival that would topple the theory: selection, nested not refuted

Efficiency selection is real, and the theory does not deny it. But the right distinction is not “selection operates on a map” tout court — that would be false for natural selection, which needs no explicitly declared fitness function, and false for markets, which can select on realized outcomes even when the internal causal structure is unknown. The defensible distinction is between two kinds of selection. External selection — the market, the environment — registers realized fitness at the system boundary: it can record that an organization is failing without identifying which internal capability was lost or why. Internal managerial selection — the rationalization decisions inside the organization — operates instead on declared metrics and proxies (the rational mode of Paper 2), and can therefore select against capabilities that remain valuable in the real flow.

Where neither kind of selection reaches is the divergence between map and flow itself. External selection sees the boundary result but not the internal map–flow gap that produced it; internal selection sees only the declared proxies, so the gap is invisible to it by construction. An agent or unit “fit” on the internal map can be value-destroying in the flow, and the internal process selects for it; meanwhile the external process registers only the eventual aggregate decline, too late and too coarse to name the cause. So the space this theory occupies — the internal divergence between representation and operation — is exactly the space external selection cannot resolve and internal selection actively mismeasures.

[mechanism] The measurement wall. The deep reason selection cannot supply the mechanism is the same wall as everywhere in the program: contribution is not identifiable from the local metrics and attribution structures that currently govern the process — estimating it requires additional evidence and a broader frame. Suppose a month is left with no metric to hit. Some agents spend it on genuine contribution — a teacher improving preparation and delivery — but that contribution is invisible to the local instruments; and what the local view cannot register it cannot locate, and what it cannot locate it cannot act upon. The system cannot find where its inefficiency is from inside its own metrics, only intuit it. So a maximizer — firm or market — acts only on what it can measure: the market eventually registers the outcome (the system competes worse) but never the cause (which capability was hollowed), because the cause left no trace in the governing metrics. Selection is blind to its own input.

[mechanism] Enriches, not contradicts. The market predicts that an inefficient system eventually stops competing; it does not predict when, or which, or why this one, because inefficiency is unbounded and survivable (§6). The two accounts are complementary, not rival: selection measures fitness at the boundary; this theory measures the correspondence between the boundary-map and the internal-map. Neither predicts the other’s quantity. This theory supplies exactly what selection cannot — the internal account of when and why a system stops being able to compete. (A clean fence: this model is deliberately closed, market pressure left out by construction. In the open model, the “make the rot float” decision of §8 is a market event — the forced-clarity regime change is triggered by the interaction of the internal system with the external one, both maximizing.)

[mechanism] Systems can turn against themselves — which is nesting, not a counterexample. One further point belongs here, because it is often raised as if it refuted the nesting: many systems, under selection, develop sub-parts that turn against the whole — a unit that maximizes a local signal while consuming the substrate that sustains it. This is the plague, or automatizer, mode: a part running the maximizing-on-the-map logic at one frame while degrading the system at the frame above it — consuming its own resources against its own intelligence and its own efficiency. It is not a counterexample to nesting; it is nesting — selection acting at one level while a maximizer consumes resources at another. (A vivid biological instance of this same pattern is offered, as an optional analogy that is load-bearing for nothing in the argument, in the Appendix.)

10. Nokia: the case that motivates the complementarity claim

Positionality. What follows is a first-person field observation, not documentary history. The author worked inside Nokia in this period (a Barcelona development centre, 2000–2004) and writes from direct presence in both the internal map and the work it governed. It is offered as witnessed practitioner evidence, with other causal accounts of the company’s later decline explicitly left open; it motivates the hypothesis below, and does not by itself establish it.

[Illustration — witnessed; motivating, not a causal account of the decline] In 2000–2004 the market and competitive-dynamics predictions available at the time did not point to Nokia’s fall — if anything, the more map-information one held, the more confidently one would have predicted the opposite. In one development centre sat three capabilities that turned out to matter for the future — social messaging, instant messaging, and location. Anyone reading the external map would conclude: three promising capabilities, co-located, a strong position. There was no external signal saying “weak, uncompetitive, should exit.” (The animal image is the intuition: you see a lion in its prime — full strength, full will, every capability — and a lesser one beside it, and you are asked which wins. From outside you cannot be sure; the thing that decides is not fully visible at the frame where you are scoring strength.) This is a single witnessed case, not a controlled demonstration, and other causal accounts of the decline cannot be excluded; its role here is to motivate, not to prove.

[mechanism] The inversion. The same three capabilities the external frame read as the decisive advantage were, in the internal frame, exactly the candidate to be cut — because that centre was the one furthest off the internal map. Its strength in the flow was its invisibility in the representation. The capability most likely to win in the territory was the capability most likely to be killed by the map. External frame: their greatest asset. Internal frame: the unowned, off-map node to be rationalized away. Same object, opposite readings — and only the internal map–mismatch view sees the kill coming.

[mechanism] Method, intervention, and falsification are one procedure. This is why the case is more than a story. The intervention is: inject an external view (the capabilities that actually compete, the territory) against the internal information (the map the agents maximize on), and measure the mismatch between the two maps. That mismatch — the divergence between the market-frame where the system competes and the internal-frame where decisions are taken — is the inefficiency, operationalized. So the study, the intervention, and the falsification are the same act: measure map against map. This is what motivates the non-redundancy claim: where the mismatch measure and the competitive reading disagree, at most one can be right about the cause — they are complementary instruments aimed at different quantities (boundary outcome vs internal divergence), so one is not derivable from the other. Nokia is the motivating illustration of that non-redundancy — a witnessed case in which the two readings pointed in different directions — not a proof of it; the demonstration that the instruments are genuinely non-redundant has to come from the systematic tests, not from a single case.

11. The forward implication: inefficiency before unemployment

[tendency — falsifiable] The loudest current forecast is that automation eliminates jobs. This theory predicts close to the opposite, at least first: inefficiency rises before anything dies, and rising inefficiency demands more work, not less. A maximizer (Paper 2) acts only on the declared map and is blind to the off-map, so inserting one does not clean the process; it deletes the integrating layer — the coherent work reconciling what the map leaves disconnected — and leaves a thin conformant output that produces little. A process that produces little requires another process to complement it, which requires another to complement that. Automation, applied to a divergent estate, does not collapse the headcount; it multiplies the complementary processes needed to compensate for what the maximizer can no longer integrate. The map fills with more boxes, not fewer — friction made visible as labour.

The non-fatalist turn: all those agents who appear on the map but produce little are handed an opportunity to become contributors — positive deviations (§1) that absorb the inefficiency shocks, using the automated tool as an auxiliary instrument for contribution rather than as a replacement for it. The consensus is wrong because it confuses a role’s map appearance with its contribution: automation removes the substitutable appearance-work and intensifies demand for the unsubstitutable contribution-work — which, the systems we build being maximizers, stays human (the coherence-capacity debt of Paper 2).

Two fences. The boundary: on a genuinely healthy, low-gap process, automation really does substitute cleanly and really does cut jobs; the inefficiency-inflation claim is about divergent, high-deviance-dependence estates — so the precise claim is that the effect rises with deviance-dependence. Test: track total process work (headcount plus complementary processes plus integration effort) after a maximizer is inserted, across estates of differing deviance-dependence; the theory predicts total work rises with deviance-dependence, not falls. The branch: the opening for contribution is an opportunity, not a guarantee — the coherence radar (Paper 2) predicts the system will suppress the very shift it now needs. So the outcome forks: either a healthy migration from appearance to contribution, or a swelling deviance layer that merely defers the destiny — and which branch is taken depends on whether the system tolerates the coherence it now depends on more heavily.

(The consolidated forecasts and the full battery of falsification tests — for all four series — are gathered in the dedicated companion papers, so that an examiner can test the whole program in one place.)

Appendix — a biological analogy (optional, load-bearing for nothing)

[analogy — not an identity, not a physical or biological claim] This analogy is offered only because readers often find it vivid; the argument does not rest on it, and a reader who distrusts biological analogies can discard this appendix without losing anything. The plague/automatizer pattern of §9 — a part that maximizes a local signal while consuming the substrate that sustains it — has a recognizable instance in cancer: a line of cells that maximizes its own replication and minimizes correspondence with the body it depends on, proliferating within the closed frame of the organism and, in the limit, killing the host and itself rather than ceasing to maximize. Read structurally, this is the same shape as the mechanical maximizer of Paper 2 — blind past its own local objective, it consumes the system above it — and it illustrates that selection and this theory operate at different frames: selection acts where organisms compete, while the maximizing-on-the-map logic runs at the frame of the cells within. The point is structural recurrence only; no claim is made about oncology, and the program’s use of “entropy” remains non-thermodynamic throughout.

— Iván Abril Palma