The Attribution Gap and Capability Loss

An architecture theory of why enterprise rationalization destroys the capabilities a firm most depends on — with a fenced conjecture toward the wider economy.

Abstract

Healthy, growing firms repeatedly destroy capabilities they still depend on — not the weakest applications in a portfolio, but, disproportionately, some of the strongest. This paper explains that paradox as an architecture problem, not a capability problem. In an enterprise, the real information flow (applications transforming information that contributes to what the enterprise sells, delivers, protects, or is required to maintain) and the architecture of that flow (the firm’s model of who contributes what, and who depends on whom) are two distinct levels. The capabilities live on the first level; rationalization decisions are made on the second. The central claim is that the architecture decays relative to reality — and that it decays for economic reasons, not technical ones. The wedge between the two is the attribution gap: the difference between an application’s contribution (the value it really adds, traceable through information lineage) and its attribution (the credit it receives, which is what moves budget). Because making a dependency explicit assigns the property right and transfers the rent to the source, every actor has a standing incentive to keep dependencies implicit while securing the inputs they rely on. Under that incentive the gap’s expected direction is widening across decision cycles — through a control-grab in bad times and a contest of heroes in good times — until the firm can no longer reliably see, on its own map, which capabilities are load-bearing. At the moment of the cut, even with information on the table, the decision turns on will: a governance body asks who will own and be accountable for the critical capability, and the answer is silence. The unowned high-contributor is cut, in the disguise of “retiring legacy” or “not deploying the new,” with its value still in it.

The paper proposes the firm-level theory as practice-grounded and put forward for testing — the phenomenon directly witnessed from both sides of these decisions across roughly twenty-five years of rationalization practice, the mechanism offered for refutation rather than claimed as proven — and states falsifiable predictions on records firms already hold: a discovery gap whose shadow set is enriched for high-contribution applications; the silence at the ownership question; a wedge between acquisition price and subsequent internal attribution; a build-versus-buy inversion traceable as rejected-build-then-external-buy of the same capability; and “rationalization scars,” critical capabilities fragmented into undocumented legacy remnants. A controlled comparison of two governance treatments — clarity-only versus revaluation-then-clarity — is the decisive test. A generalization to the wider economy is offered separately and fenced as conjecture, with its strongest rival (trade and comparative advantage) named.

The theory in one page

The causal chain. Incomplete visibility of real contribution → attribution distortion (credit and budget move away from the source) → resource and ownership withdrawal (the source is under-funded and made uninvestable) → accountability entropy (no one can or will own it; the candidate-owner set is indeterminate) → ownership silence at the decision → shadowing, non-investment, or destruction of the capability.

Two of these are distinct constructs, not one (sharpened in §6 and §9): the attribution distortion is a wedge between two maps — what an application contributes versus what it is credited; the accountability entropy is an indeterminacy over who will own the result. The first is the economic driver; the second is its downstream consequence; ownership silence is the observable; destruction is the outcome.

Four terms hold the argument.

• Reality — the real information flow: applications transforming information that contributes to what the enterprise sells, delivers, protects, or is required to maintain.
• Architecture — the firm’s model of that flow: its representation of contribution and dependency. Decisions are made here.
• Contribution — the value an application actually adds (estimated through lineage and validated by operational and economic consequence — and built only from signals independent of attribution; §6).
• Attribution — the credit it receives, which is what moves budget, ownership, and accountability (multidimensional: budget, ownership, recognition).

The attribution gap is the divergence between contribution and attribution. It opens because making a dependency explicit creates a legitimate claim for recognition, funding, and authority at the source — threatening the dependent actor’s captured rent — so everyone is incentivized to keep dependencies implicit. Decisions run on the architecture (attribution), not on reality (contribution); so the high-contribution, under-attributed capability becomes unowned, shadowed, and — with its value still in it — eventually cut. This is decay by incentive, not decay by neglect, and it makes the central claim an architecture problem, not a capability problem.

The predictions, in brief (the firm-level claims that make the theory refutable; full statements in Part VI):

Prediction	Observable indicator	Expected direction	Main confounders to control
Discovery gap / shadow enrichment	inventory estimate vs. discovered count; contribution of shadow apps (lineage + criticality)	discovered ≫ estimated; shadow set over-represents high-contribution sources	application age, scope, benign undocumentation
Ownership-silence	in decision records: importance affirmed vs. named accountable owner accepted	unanimous importance + no owner → raised probability of {non-investment, shadowing, elimination}	genuine obsolescence, true redundancy
Internalization wedge	capability-attributable acquisition value vs. later internal budget/credit	acquisition value > subsequent internal attribution	goodwill, synergy, control premium, growth options
Build-vs-buy inversion	rejected internal build (fully-loaded), then external buy of a comparable capability	same capability bought after being rejected as a build	scope, maturity, risk, time, IP, support, integration comparability
Rationalization scar	loosely-coupled legacy fragments + unused replacements per capability	critical capabilities more fragmented; traceable to past cuts	organic complexity, M&A history
Loss is informational, not performance	losses by measured performance failure vs. ownership/value indeterminacy	a significant share driven by indeterminacy, on still-productive capabilities	mislabelled performance failures
Governance experiment (decisive)	clarity-only vs. revaluation-then-clarity on matched portfolios	clarity-only ↑ erroneous destruction; revaluation-then-clarity ↓	performance, cost, duplication, obsolescence, criticality, sponsorship, replacement

What is asserted, and what is not. The phenomenon — that healthy firms recurrently destroy capabilities they still depend on, in the pattern above — is directly witnessed (§0). The mechanism offered to explain it is proposed and practice-grounded, and put forward for testing, not claimed as proven. Its currently exposed flank is the independent measurement of contribution (§18). The wider-economy generalization is conjecture (Part V). Monetary and thermodynamic readings are optional (Appendix A).

0. Scope, discipline, and what is and is not claimed

This section states the fences before the argument, because the argument’s credibility depends on them.

What is measured. The contribution of information to value creation, in the information-economy setting, where that value flows through applications. The unit is the application or initiative — concrete, enumerable, and carrying recorded ownership, budget, dependencies, and backlog.

Application and capability are not the same unit. This paper’s claim is about capabilities, but the traceable technical unit is the application; the two are related, not identical. One application may serve several capabilities; one capability may span several applications; an application may be replaced while its capability survives; a capability may be lost while some application remains. Throughout, an application stands in for the capability whose information transformations it performs, and the mapping is application ↔ information transformation ↔ business capability. Where it matters empirically — in the companion measurement protocol — the operational unit is the application–capability pair, which keeps technical traceability without pretending the two coincide.

What is not claimed. Not all firm value is information; not all of a firm’s worth flows through applications. The paper measures the information slice — the slice that can be traced — and claims nothing about the rest.

Two levels, kept distinct. There is the real information flow (applications genuinely transforming information) and there is the architecture — the firm’s meta-model of that flow, its representation of contribution and dependency. The real flow can be entirely healthy while the architecture rots. Decisions are made on the architecture. When this paper says capability is destroyed, the destruction runs from a degraded map to an intact territory — which is exactly why it is invisible.

Proposed, not yet proven. The firm-level theory is proposed and practice-grounded, and put forward for empirical testing. What is directly witnessed is the phenomenon — the recurring decision pattern, observed by the author from both the central rooms where rationalization decisions are made and the periphery where they are received, repeatedly, over roughly twenty-five years of practice as a rationalization architect. The explanatory mechanism is offered for refutation, not asserted as established. The generalization to the wider economy is conjecture — a field note offered with its rivals named, not a result. The independence of the firm theory from any monetary or early-warning framework is deliberate and is a feature, not an omission (see §18).

Witness versus proof. The author’s position is motivation: a credible practitioner reports a recurring pattern. The proof is the set of measurable predictions in Part VI, testable on records firms already hold. The war-stories illustrate; they do not carry the weight. The two are kept apart throughout.

Claim-typing. Claims are tagged: [definition] (a stipulated meaning), [identity] (true by construction within the stated boundary), [mechanism] (a causal pathway asserted for the class examined), [tendency] (holds under stated conditions, with counter-cases admitted as tests), [conjecture] (offered for examination, not asserted).

On the word “entropy.” It is used non-thermodynamically throughout: it denotes the indeterminacy of the answerability ledger — the spread of possible true states of “who is accountable for what” — measured concretely (§9). Any thermodynamic correspondence is an optional lens, quarantined to Appendix A, and is a model, not an identity.

Part I — The question and the foundations

1. The explanandum

The question is not “do organizations sometimes make bad cuts.” It is sharper and more uncomfortable: why do healthy companies kill their own best capabilities?

The standard view of the well-run firm predicts the opposite. With aligned incentives and competent governance, rationalization should cut the worst applications — the redundant, the obsolete, the unused. The phenomenon to be explained is that, under identifiable conditions, the best — the most depended-upon, the most load-bearing — are disproportionately at risk.

This looks like a contradiction: a firm that was doing well, then destroyed what made it well. It is not a contradiction in the argument. The paradox belongs to reality, and the theory’s job is to dissolve it. Stating “good, then bad” is not the author contradicting himself; it is the world contradicting itself, and the rest of this paper explains how.

2. Two properties, not two kinds of thing

The usual move is to split the world into productive assets and unproductive ones, capabilities versus mere claims. That split is too clean and the boundary is contestable. The better foundation is a proportion.

[definition] Productive things may carry two coexisting properties, in varying proportions — and the firm-level theory does not require every one of them to carry both (some narrow capabilities, such as team-based tacit knowledge, are barely tradable at all):

• exchange value — it can be traded; it is a claim; and
• generative capacity — it can produce more than itself; it is a capability.

A stored good is mostly exchange value with a sliver of generative capacity; a skill is mostly generative; a system in use is both. There is no need to adjudicate “is a fish an asset or a capability”; it is mostly the first with a little of the second.

The load-bearing claim is a one-directional asymmetry:

[mechanism] Yield accrues only to the generative property; exchange value only redistributes. You can always convert generated output into stored exchange value, but stored value cannot, by itself, become generative capacity — it can only fund a capacity that must independently exist. The arrow runs one way and cannot reverse.

(The label is not “money.” A stored, exchangeable asset is near-money in the sense that liquidity is a spectrum; conceding the word costs nothing and the substance survives intact under “exchangeable stored value.” The monetary reading is quarantined to Appendix A.)

3. No perfect information in rationalization — a fact, not an assumption

The strongest objection to everything that follows is the efficient-firm assumption: this cannot happen, because a well-run private firm has the information and the incentives to have already cut the non-contributors.

Grant the idealized firm everything — aligned incentives, competent governance, deep resources — except one thing, and the asymmetry follows. The exception is this:

[mechanism / fact] Complete, current, and agreed contribution information is practically unattainable in enterprise rationalization. No organization holds a complete, current, agreed map of what each application contributes; observed portfolios are systematically incomplete. This is not an assumption a reader must be persuaded to grant — in the enterprise it is the ordinary condition, and the argument needs only that it holds, not that it holds perfectly.

This is the pivot of the whole paper. In a macroeconomic treatment, incomplete information is a premise one must beg. In the rationalization setting it is free — and that asymmetry between the two cases is precisely why the firm-level theory is the part that can be defended. Because information is imperfect and contributions are unequal, an asymmetry necessarily opens; the theory’s work is to predict which misallocation results, and to show it is the surprising one.

The non-circularity is secured by measuring not the static fact but its evolution: how the gap behaves over cycles, and the surprising prediction it yields (§§5–6, Part VI).

4. The internal market, and the unit of analysis

Inside a firm there is a real market, and it is not the annual budget. It runs every meeting and every coffee: a continuous market of reputation, influence, positioning, and backlog-as-evidence-of-demand. That market is the engine of what follows, and it should be described as such.

But the measured currency is kept hard. The reputation market generates the claims; the recorded ledger — budget, headcount, funding, the cut decision, named ownership — is the gauge. The soft market explains; the hard ledger measures. (The same discipline recurs throughout: where a mechanism is soft, the variable counted next to it is recorded.)

[definition] The unit is the application (or initiative). Throughout, two neutral labels are used: Application D, a high-contribution application, and Application Z, a low-contribution one. The evocative terms diamond and zombie are reserved for the interpretation in §10, once the mechanism stands on its own; loading them into the formal development would make the argument read as persuasion.

Finally, the good being traded inside the firm is information. Money is the budget; the good is information; applications produce and transform it. Application D is the high-fan-out source of truth — the master data, the client information everyone consumes. Dependency is therefore traceable through information lineage: follow what the customer pays for back up the value-added chain to its irreplaceable source.

Part II — The mechanism

5. The ratchet: accountability, not backlog

The pressure on Application D recurs every cycle (the process is iterative, not a sequence of stages). In bad times, every owner can credibly say “we missed because D did not give us the input” — and demonstrate it, because D is the only one who could have delivered. In good times, everyone routes their roadmap through D and the demand on it explodes. Either way, commitment piles on D.

The naive version of this is “D’s backlog explodes while others’ shrinks.” That is wrong, and the correction is important: backlog is uniform. Every application is underwater; everyone has unserved features. Backlog volume is uninformative and a critic dismisses it in one line.

The variable that actually carries the asymmetry is the charge on the backlog — the blame-exposure attached to not serving it.

[mechanism] Z can wave an enormous backlog and nothing happens if it goes unserved, because nothing depends on Z. D’s unserved backlog detonates, because everything does. The asymmetry is on accountability, not on backlog.

And the engine beneath it: blame is concrete and sticks to the source; credit is diffuse and is captured downstream. When the information is wrong, everyone can point at D — attributable, undeniable. When the information produces a good outcome, the credit lands on whatever interface “delivered” it — diffuse, capturable. So D accumulates the blame and loses the credit.

[definition] Define net accountability = blame-exposure − retained credit. It is maximal on D and roughly zero on Z. That gap, not the backlog, is the diamond beginning to form.

[tendency] The ratchet holds under continued reliance. The counter-case — an application whose claims genuinely shrink (de-prioritization) — is admitted and is itself a test of the conditions.

6. The attribution gap

This is the root. Two quantities, defined within a single decision boundary (a portfolio assessed together):

[definition]

• Contribution — the value an application actually adds to the information chain. It is not read off lineage alone: lineage establishes dependency and propagation, not economic value (a highly connected application may be indispensable, or merely locked-in, a bottleneck, or central only because the architecture is poor). Contribution is therefore estimated — anchored in information lineage and validated by operational and economic consequence: process criticality, unique transformations performed, substitution availability, replacement cost and time, revenue or service impact, regulatory and quality impact, failure consequences, and tacit-knowledge concentration. The contribution estimate must be built only from signals independent of attribution — lineage, replacement cost, failure consequence, substitution availability — and not from budget, headcount, or recognized credit. If contribution is estimated from anything that also drives attribution, part of the gap is built in by construction; avoiding that circularity is the central measurement problem of the theory (§18), and the companion paper carries it behind an inter-rater-reliability gate before any contribution number is used.
• Attribution — the credit an application receives, which is what moves budget, ownership, and resources. Attribution is multidimensional and its dimensions need not move together: budget B, named ownership and authority O, and reputational credit R. An application may hold budget but no real authority, or credit but little funding. It is safer to treat attribution as a vector A = (B, O, R) and let the data show which dimension best predicts the downstream outcome than to impose weights too early; where a scalar is needed, A = w_b·B + w_o·O + w_r·R with transparent weights.

To compare contribution and attribution at all, normalize each within the decision boundary: C_i = estimated contribution of i ÷ Σ_j estimated contribution of j, and an equivalent normalized attribution share. Only after this normalization, and only within the boundary, does the “credit pie sum to one” — in ordinary discourse organizational credit is not naturally fixed-sum (several groups can all claim a success at once); the fixed sum is an artifact of normalizing within a decision.

[definition] The divergence is then a redistribution, clearest stated as two non-negative directional quantities:

• capture surplus = A_i − C_i ≥ 0 for the over-credited (capturers, interfaces, Z); and
• contribution deficit = C_i − A_i ≥ 0 for the under-credited (the source).

Because the normalized shares each sum to one, every unit of capture surplus is matched by a unit of contribution deficit elsewhere. Application D is the extreme of the second kind: high contribution, low attribution — a large and widening contribution deficit. (When the text below says the gap “widens,” it means this deficit on the source grows — its under-attribution becomes more severe — not that a signed number drifts toward zero.)

This attribution distortion is the economic driver of the accountability entropy defined in §9 — but it is not the same variable, and the two should not be conflated. The contribution deficit is a wedge between two maps (what an application contributes versus what it is credited). The accountability entropy is an indeterminacy over who will own the result. The deficit comes first and produces the entropy: as recognition and funding withdraw from the source, it becomes under-invested and uninvestable, and that drives ownership willingness toward zero and the candidate-owner set toward indeterminacy (§9). The causal order is distortion → resource and ownership withdrawal → entropy → silence → loss.

The whole tragedy is one line: rationalization allocates and cuts on attribution; D carries the largest contribution deficit; so D is cut because the ledger that decides its fate is the one that most understates it.

Why the deficit widens — the deep mechanism: explicit dependency is monetizable; implicit dependency is not. The moment an owner makes the dependency explicit — “ninety-nine percent of what I deliver rests on Application D” — they create a legitimate claim for recognition, funding, and governance authority at the source, which threatens their own captured attribution and rent. (The transfer is not automatic — a dependency can be documented without budget moving — but the threat to the capture is enough to deter naming it.) The rent an actor collects — realized value without acknowledged responsibility — therefore exists only while the dependency stays implicit. Every actor has a permanent incentive to lock the input while keeping the dependency unspoken: secure what you depend on; never name how much. Naming the dependency is what threatens the rent.

This widens on both arms of the cycle:

Bad times — the control-grab. When the input fails, no one fights to give D more budget; they fight for control over D. “We did our job; D failed; therefore we need oversight of D, KPIs on D, early warning on D.” This is control without accountability — the inverse of what would help. And note: making the dependency explicit through blame does not close the deficit; it widens it. Blame-explicitation is a deeper capture, not a recognition. Only revaluation — clarity that credits the contribution — closes it.

Good times — the contest of heroes. Success demands an explanation, and being a hero is cheap: everyone has a heroic attribution theory. So in good times attribution is handed out by competition to claim, not by contribution, and the real contributor — one hero among many claimants — is systematically under-credited. The same dynamic has an internal-freeware form: D’s information is an internal free good, consumed with no internal price that captures its value, so intermediaries cheaply build paid “value claims” on a free input. (The hero market is soft — narratives cannot be counted — so the measured quantity stays hard: contribution via lineage-and-consequence against attribution via budget; the heroism is the engine, not the gauge.)

[tendency] Under continued capture incentives and absent independent revaluation, the expected direction of the contribution deficit is widening across decision cycles. This is not a law of arithmetic, and not every observation moves one way: the deficit can narrow under a leadership change, a serious incident, a regulatory audit, architecture renewal, new ownership, platform consolidation, internal pricing, or a powerful sponsor. Each such reversal is itself a test of the conditions — and the one structural force that closes the deficit is clarity from outside the local incentive system (§11). (Demonstrating a widening requires a contribution estimate at two points in time; this is the hardest line to instrument and the place a longitudinal skeptic will dig.)

Neighbour. This lives next door to the hold-up problem and property-rights theory (Hart & Moore; Williamson) — inverted. The classic hold-up is under-investment when a contract is incomplete. Here it is the dependent party keeping the dependency implicit to retain a rent, so the critical source is never recognized — leading not to under-investment but to invisibility, and eventual wrongful elimination.

7. Why the firm grows while the gap widens

An economist reads §6 as “inefficiency is growing,” and objects: if so, the firm should be competed away — yet firms grow. The resolution turns the objection into the theory’s own prediction.

[mechanism] The firm grows by recognizing external capability while capturing internal capability — on opposite sides of its boundary.

• External capability is priced → it must be paid for → its value is explicit and recognized. You cannot capture what you do not own, so you pay full value (the acquisition: “we’ll pay a lot for that little competitor’s application”).
• Internal capability is unpriced → no market price disciplines it → its value goes implicit and is captured; the attribution gap opens; it is under-credited and extracted.

Internalization is the flip. The moment a capability is acquired, it crosses from priced-and-recognized to unpriced-and-captured. The boundary of the firm is, precisely, a recognition boundary.

The growth is the mask. Top-line growth is acquisition — recognizing value outside — and it is exactly what hides the extraction inside. The fact cited as disproof is the camouflage. This is a sign-flip at the level of the whole firm: clarity about internal decay is suppressed because the firm is visibly growing.

Dark complement to Coase–Williamson. The transaction-cost theory explains internalization as saving transaction costs. The complement: internalizing also destroys the price signal that kept the capability’s value explicit. The same act that saves transaction costs opens the capture gap.

The treadmill — “dies of its own success.” Because every internalized capability begins to be captured the moment it is inside, the firm comes to depend on a continuous external supply of fresh capability to replace what it internally erodes. Growth can outrun internal decay for a long time — which is why the firm looks healthy while hollowing out. The collapse comes when the treadmill cannot keep up. Famous corporate deaths are recognizable here; this paper treats them as motivating illustration, not as causal claims about any named firm.

(Lead with the recognition flip, not “deep pockets.” That big firms can pay more is true and trivial; the mechanism is the flip at the boundary, and deep pockets are merely what lets the firm keep buying recognition while it spends down internal capability.)

[F-A1] The portion of an acquisition price attributable to the capability itself exceeds the internal attribution that same capability subsequently receives: firms pay full recognized value to acquire, then under-credit it once it is inside. Testable by tracking an acquired capability from the capability-attributable share of its acquisition price — net of goodwill, expected synergy, control premium, and growth options — through its post-internalization budget, ownership, and strategic visibility.

8. The cost side and the build-vs-buy inversion

The attribution gap has a second face. §6 was the value side — D under-credited. There is also a cost side: an internal capability must carry the program manager, the program manager’s manager, the innovation department that never innovates, the deep-technology group that does no deep technology — it must “produce value for all of them,” and every claimant loads cost onto it.

[mechanism] So the gap is two-sided — value captured away (D under-credited) and cost loaded on (D over-charged) — and together they produce the sentence that makes this stage lethal: an internal capability with strongly positive standalone ROI shows negative fully-loaded ROI, not because it is not worth building, but because the internal ledger understates its value and overstates its cost at once. The external application, priced clean by an outside market with no overhead to absorb, looks like the better investment. The firm buys instead of builds even when it has more than enough resources to build — an accounting distortion, not a judgment of merit.

[F-A2] The build-vs-buy inversion leaves a trace: the firm rejects an internal build of capability X on a fully-loaded business case, then acquires a comparable capability externally. The test requires demonstrated comparability of the rejected build and the purchased capability — scope, maturity, delivery risk, time to market, intellectual property, support model, integration cost, and expected outcomes. With those controls, the wedge — standalone ROI versus fully-loaded ROI, and rejected-build-then-external-buy of the comparable capability — sits in investment-committee records and is among the paper’s strongest tests.

Killed for not being created. The firm does not only cut existing diamonds; at this stage it aborts nascent ones — the good internal proof-of-concept that never scales, because scaling means dragging the whole Z-estate into the deployment (the integration tax is the overhead-loading applied to scaling), so the business case dies before the capability is born. [conjecture / mechanism] This is the most damaging and the least measurable form: there is no regret data for a capability that never existed. The counterfactual is not claimed to be measurable; the test is anchored only on the rejected-build-then-buy proxy of [F-A2].

This is also where the “growth is the mask” lives operationally: the firm grows by acquisition precisely while its internal engine stalls, so the growth hides the stall.

Part II-B — Worked example: a customer master-data capability

[Illustration — composite and anonymized] To see the mechanism end-to-end on one case, trace a single application: the customer master-data source — the system of record for customer information that the billing, CRM, analytics, and fulfilment applications all read from.

• Contribution. Every customer-facing outcome ultimately consumes its data; the lineage fans out across the portfolio; years of cleansing and matching logic make it costly to replace; a failure degrades billing and fulfilment at once. Its estimated contribution — anchored in lineage and validated by criticality and replacement cost — is high.
• Attribution. It was built years ago, runs on older technology, and has a small team. The downstream applications present the dashboards and the customer experiences, and the credit — and the budget — follow them. The master-data source is recorded as “data plumbing.” Its attribution is low; its contribution deficit is large.
• The deficit widens. A data-quality incident arrives; everyone can point at the source (“the input was wrong”), and a downstream group bids for oversight of it — KPIs, control, early warning — without accepting ownership (the control-grab). Later, a celebrated analytics launch is credited to the analytics team; the clean data that made it possible is one contributor among many claimants (the contest of heroes). The deficit grows in both seasons.
• The shadowing. Anyone proposing to invest in it hears: it is legacy, it must be refactored, that needs the central teams, it must enter the central roadmap — so the burden swamps the improvement and no one asks. Its real maintainer keeps quiet (raising a head invites the blame routed through the source); its dependents will not own it (owning it means owning everyone’s data-quality blame). It is triple-hidden. It recedes to present but uncounted — still load-bearing, no longer in the decisions that matter.
• The discovery. A portfolio review finds it absent from the official inventory, or listed as minor; its true fan-out surfaces only by tracing the dependencies of the visible applications back to it.
• The cut. “We have several systems touching customer data; consolidate onto the modern platform.” Everyone affirms the source matters. Who will own it and be accountable? Silence. It is retired — in the costume of modernization.
• The scar. The cut hits undocumented downstream dependencies; a middleware bridge is added “for now”; the capability ends as several loosely-coupled remnants plus an unused module on the new platform. Years later, a discovery exercise finds the scar and asks why there are five systems where there should be one.

Read against Part VI, the same case is measurable: the contribution deficit (lineage-and-consequence versus budget), the ownership-silence (in the meeting record), the scar (countable in the estate). And had it been revalued before the consolidation decision rather than merely clarified, the theory predicts it would have survived.

Part III — The death

9. The travel to the shadows

The naive expectation is that the diamond gets cut. It usually does not — at least not first. Killing what you depend on is a wild move no one takes; the real mechanism is blurring.

[mechanism] When D’s attribution gap leaves it carrying everything and credited with little, it is not executed; it is shadowed. The death is the shadowing. The irreversible loss accrues in the dark — the team erodes, the knowledge decays, the technical debt compounds — long before any formal decommission. If a cut ever comes, it lands on something already hollowed. The loss is not a moment one can point to; it is a fade, which is exactly why it is invisible and why it is never recovered.

The entropy, operationalized — and downstream of the distortion. The indeterminacy of §0 is now concrete, and it sits after the attribution distortion of §6 in the causal chain: as recognition and funding withdraw from the source, it becomes uninvestable and no one will own it. Accountability entropy is the indeterminacy of who is accountable for D — formally, over a distribution p_k across the candidate owners k, a quantity like H = −Σ_k p_k log p_k, high when many parties are apparent owners and none is the accepted one. The state space is the set of candidate owners; the entropy is the uncertainty over which actually owns it; and it is actively maintained — the real owner will not raise their head, because it could be cut, and the dependents hide them, because they rely on them. D is triple-hidden: by its owner (self-protection) and by everyone who needs it. It is measured as the spread of candidate owners and the gap between responsibility claimed and accountability accepted — entropy over the accountability assignment, not over the data, which keeps flowing fine. The attribution distortion is the driver (why recognition left); the accountability entropy is the result (why no one will own what is left).

Responsible versus accountable — the engine. Everyone will claim responsibility for D’s input, because that justifies their credit and budget for “producing” it; everyone will refuse accountability for D, because accountability means owning its insolvency and its blame. Many responsibility-claims, zero accountability-acceptance: that is the unfindable owner; that is the entropy. And it is why clarity is lose-lose for both sides — the Z owner who reveals “I depend ninety-nine percent on D” loses the capture; the D owner who reveals the real situation makes themselves the named target. Both prefer the dark.

Uninvestability and recession. D becomes uninvestable for the same reason a good internal capability is not built (§8): any investment triggers “it is outdated, it must be refactored, that needs the central teams, it must enter the central roadmap,” and the burden swamps the small improvement, so no one even asks. The status quo is the shadow. The terminal state is not non-existence but present but uncounted, still load-bearing: D is known to exist, still runs, still produces, but has exited the attribution game. Shadow IT, properly understood, is not “what central IT does not know”; it is a gradient — central IT, regional IT, business IT — ending in a remnant that everyone knows exists and that no longer participates in the contested ledger while still bearing the load.

[F-C1] Empirical anchor — the discovery gap. In application-portfolio discovery, the estimated count understates the real count severely (in the author’s engagements, an estimate of one-to-two thousand applications against a discovered six thousand-plus — a four-thousand gap). That gap is the architecture-reality divergence, quantified. The falsifiable claim is sharper than the gap itself: the shadow set is enriched for high-contribution applications — not merely low-value leftovers — and they are found by tracing the hidden dependencies of the visible applications. Standard expectation says the undocumented are junk; this predicts that critical, high-dependency sources are disproportionately in the shadows. Testable across portfolios, controlling for the benign reasons an application is undocumented (age, small scope), so the enrichment is shown to be contribution, measured by dependency. (The claim is not “shadow equals valuable” — most shadow applications may well be junk; the claim is the non-random presence of high-contribution applications, above what chance or the efficient view predicts.)

10. The cut

At the moment of rationalization, even with clarity on the table, the decision does not run on information. It runs on will — and will means that every manager must put accountability on the table. The rationalization moment is an accountability market.

The scene is recurrent. There are twenty-five applications nominally doing the same thing; only one or a few carry the real differential, but nominally they are interchangeable, and the dependency on the differential one is not on paper. (Fifteen years after a migration to a modern platform, a legacy layer is still doing things in the manufacturing plant that nobody understands and nobody can replicate — recorded, if at all, as “a small dependency, kept just in case.”)

The centerpiece is a single, recordable observable:

[F-C2] The ownership-silence test. Everyone in the room affirms the capability is important. The governance body asks who will own it and be accountable. The answer is silence. The prediction is counterintuitive and clean, but it is probabilistic, not deterministic: importance-affirmation combined with ownership silence materially raises the probability of a destructive outcome — non-investment, continued shadow operation, or outright elimination — even though the room is unanimous that the capability matters, because importance-affirmation is not ownership-acceptance, and the decision runs on the second. (This is consistent with §9: the most common outcome is not an immediate cut but the shadow, with the cut as its terminal form; forced ownership is the rarer fourth path.) Assert value → request a named owner → measure the silence → predict the elevated hazard of {non-investment, shadowing, cut}.

The diamond’s disguise. The cut does not feel like destruction. The diamond usually wears the costume of “legacy to be retired” — the old, cheap, working, undocumented system, while the Z’s are the new, expensive, well-licensed, understood-but-dependent ones — so cutting it feels like modernization. The same mechanism kills the new diamond from the other side: nobody will own the burden of scaling it, so it is not deployed. The death has two costumes — “retire the old” and “don’t risk the new” — and one cause beneath both: silence at the ownership question.

[F-C3] Empirical anchor — the rationalization scar. The deployment of a cut, fought in the shadows by those who depend on the capability, produces a worse entropy than the cut was meant to resolve. You try to kill one application, hit an undocumented dependency, bandage it with a middleware “for now,” need another application to bridge the legacy capability, and end with five loosely-coupled legacy fragments — undocumented, “temporary” made permanent. Years later, someone finds five systems doing what one did, plus a modern system that supposedly has the capability and sits unused. (The anonymized canonical case: an inventory capability in a manufacturing plant, split into five loosely-connected legacy systems alongside an unused execution system.) This is a countable fossil. The claim: critical capabilities show more fragmentation, more undocumentation, and more persistence than non-critical ones, and where logs survive, the fragmentation traces to a past failed cut.

The brave executioner. Eventually someone acts: “I don’t care, I’ll take the decision nobody would take for ten years — kill it, one or five, all of them.” And that person is blamed in the postmortem for the consequences that pile up. The deep point is not “blame falls on the brave.” It is that the blurrers leave the executioner a forced gamble under entropy: by the time anyone acts, the value information is gone, so the cut is a bet — sometimes a real zombie, and the cut is right; sometimes a diamond, and it is catastrophic — and they cannot tell which, because everyone else spent years ensuring they couldn’t. The moral is therefore neither “be brave and cut” (the brave cut completes the destruction) nor “never act” (paralysis is the shadow). It is revalue before you cut — and neither the blurrers nor the executioner do it. This is also why the clarity must come from outside the local incentive system: breaking the blur makes you the blame-sink, so only an agent insulated from that blame-economy will do it. The blame-on-the-actor is the force that keeps the blur intact.

The witness, and the absence of a villain. What makes this account more than grievance is that the author has been on both sides — the central rooms where the decision is made and the periphery where it is received — and they do not contradict. The centre’s decision was rational (everyone affirmed importance; no one would own it; the cut followed) and the periphery’s loss was real (it was genuinely delivering). Both true at once, no incompetence, no malice, only the structure. (The recurrence is asserted, not the instances; the field cases are anonymized in Appendix B; none is a claim about why any firm failed.)

11. The antidote

Everything above tends to widen under one condition: that the only intelligence acting on the system is intelligence inside the local incentive equilibrium, which profits from the blur. The single structural force that reverses the deficit is clarity from outside it.

[F-C4 / decisive] Systems that bring in clarifying intelligence external to the incentive equilibrium — or accept an outside authority’s re-architecture — and revalue before they cut retain value-positive capabilities at materially higher rates than systems that pierce the blur without revaluation. The crucial and counterintuitive corollary: clarity without revaluation accelerates, rather than prevents, the loss of diamonds — because clarity that resolves ownership without first recovering value simply cuts the unowned, and the unowned high-value cell is the diamond.

The decisive test is therefore a controlled comparison of two governance treatments on matched rationalization portfolios:

1. Clarity-only — force ownership and funding decisions.
2. Revaluation-then-clarity — first reconstruct capability value, dependencies, tacit knowledge, and replacement consequences; then assign ownership.

The prediction is that clarity-only increases erroneous capability destruction while revaluation-then-clarity reduces it, controlling for actual performance, cost, duplication, technical obsolescence, operational criticality, sponsorship, and replacement availability. (A companion measurement paper specifies the protocol: the asymmetry measure, the contribution–attribution operationalization via lineage, the outcome and regret measures, and the confound controls.)

Part IV — The synthesis

12. It is an architecture problem, not a capability problem

The organizing interpretation, stated last because it ties the measurable results together. An enterprise architect’s object is the architecture: the meta-model of the information flow — the map of contribution and dependency. The attribution distortion of this paper is the divergence between that map and the territory it claims to represent — the architecture decaying away from real contribution; the accountability entropy of §9 is then the downstream indeterminacy of ownership that the decayed map produces. (Two constructs, in that order — §6, then §9.) The whole theory has three layers, each more fenced than the last, with the most defensible — and the one most native to the author’s profession — at the base.

• Bedrock (independently testable; needs no economics): the architecture decays relative to reality, and that decay — not the loss of underlying capability — drives the destruction of capabilities and can contribute to terminal organizational decline. The map says one thing; the territory is another; decisions run on the map. The discovery gap of [F-C1] is the first direct measure of that map–reality divergence; whether the divergence is caused by attribution incentives, whether the shadow set is enriched for high-contribution sources, and whether the divergence causes capability destruction are the further predictions that decide whether the wider theory holds. (Firm death is a possible terminal outcome, not an established general one.)
• Explanation (firm-level, proposed for testing): the architecture decays for economic reasons, not technical ones. This is the contribution, and it must be stated against the thing it is mistaken for. Every enterprise architect complains the documentation is stale — that is the boring version, decay by neglect and technical debt. This is different: the architecture decays because agents are economically incentivized to keep it unclear — the attribution gap is a rent, and obscuring the map is what protects it. Decay-by-incentive, not decay-by-neglect.
• Generalization (conjecture): the same at the scale of an economy (Part V).

The headline, with one precision so it does not collide with §9. At the point of decision, the capabilities are intact; the firm still holds enormous undeployed informational value; only the map is broken. That is the tragedy — the destruction is unnecessary, because nothing real had failed. But the decay does not leave the capabilities intact forever; it causes their destruction. The dead map kills the live territory, and the capabilities are destroyed with value still in them — the remnant the firm liquidates before it has been given. That “value still in them” is why the loss is waste, not merely cost.

Two levels, kept clean, pre-empts the obvious objection. A critic will ask: “if the information still flows and the capabilities still work, where is the problem?” The problem was never on the level of the real flow; it is one level up, in the representation that decisions are made on. The territory is fine; the map rots; the firm acts on the map.

One consequence for the paper’s standing: the enterprise architect is the protagonist of the cure as well as the witness of the disease, because the architect is precisely the role whose job is to keep the map true against the incentive to blur it. The author is not commenting from outside the mechanism; he is the named function the mechanism defeats.

Part V — A fenced conjecture: the wider economy

[conjecture — this entire Part] The mechanism appears to recur at the scale of an economy. It is offered as a field note, not asserted, and its strongest rival is named below.

13. The three moves, at the scale of an economy

The firm stages map move-for-move:

1. Not investing in internal diamonds → the economy courts external capability while extracting internal: tax breaks and benefits for foreign investors and imported founders, full burden from day one on the domestic first business. The recognition asymmetry at scale — external priced and recognized, internal captured and extracted.
2. Shadowing of capability-creation → shadow capabilities: the real ingredients of a capability — the budget, the ways of doing, the unwritten channels one must actually pass through to produce — are not represented in the formal market, exactly as the diamond’s dependencies were not on the firm’s paper. (Neighbours: informal institutions — North; tacit knowledge — Polanyi; embeddedness — Granovetter. The point is unrepresented practices inside the formal economy, not the informal economy as such.) [conjecture] A maturing economy may become more shadowy while a scaling one becomes less — unofficial paths accreting over time.
3. Killing → capabilities lost because no one will carry their unaccounted burden.

The carrier thread, and its image. The single thread that carries the generalization is the recognition asymmetry, and its image is the imported repairman. In a town built for generations on a craft — shoemaking, ceramics, baking — the local capability is extracted while the same capability is imported and paid for: the economy buys from outside the very thing it let die inside.

Name the rival, then aim past it. The standard explanation for such a town’s decline is trade and comparative advantage, and it is sufficient for the tradable part: cheaper imported shoes kill local manufacture, no further mechanism required. So the generalization must aim at the residual the rival cannot own — the non-tradable capability. Shoe repair cannot be imported as a service; yet it dies too, and is filled by imported labour. And the anomaly sharpens it further: the last carriers of these crafts hold high status and the work is lucrative — a help-wanted sign goes unanswered for years even for a salaried, no-risk post — and people in the same town train for equally hard new crafts (automotive mechanics) while refusing the inherited one. That pattern defeats, at once, low wage, low status, entrepreneurial risk, human capital and labour mobility (which predict the reverse — people moving toward lucrative work and outsiders filling the cheaper niche), and “it is too difficult to learn.”

[conjecture] What is left is the firm’s mechanism, seen from the carrier’s side: the carrier flees not the work and not the wage but the unaccounted burden of being the carrier — and that burden grows with the capability’s reputation. To be the last ceramist of a world-famous ceramics town is to inherit the accumulated standard of everyone who remembers; you must be exceptional and will be blamed if you are not. The price covers the product; the exam covers the meaning. One pays a euro for the cookie, but brings a grandmother’s taste in mind and tests the baker against it; the carrier is paid for the cookie and held to the standard of a whole tradition for free. So the most-revered capability loses its carrier first — the same sign-flip as the firm (reputation, which ought to attract a carrier, repels them past the point where the unpriced expectation exceeds the priced reward), and the same shape as the ownership-silence: no one will own what carries an unaccounted, ever-rising burden. (The residual rivals — craft-aversion, demographic collapse — are beaten by the fingerprint: abandonment intensifies where reputation is highest, which mere effort or scale would not predict.)

The epistemic asymmetry, stated plainly. In the firm, the author knows, having been in the rooms and heard what was stated. In the town, he can only suppose, not being in those heads. The firm rests on what was witnessed; the economy on what can only be supposed. That asymmetry is the discipline of the whole paper in one move. The thing the locals cannot quite articulate — the informal hot-potato — is, on this reading, the residue of the same travel to the shadows the capability underwent before it was abandoned: the firm leaves a rationalization scar in the estate; the town leaves a tacit one in the trade.

Part VI — The falsifiable core

The empirical case for the firm-level theory does not depend on the architecture interpretation or the economy conjecture. It rests on predictions testable on records firms already hold.

14. Firm-level claims (proposed for testing)

[F-1] The attribution gap is real and measurable within a decision boundary: the contribution deficit C_i − A_i is positive on high-contribution sources and matched by capture surplus A_i − C_i on capturers and interfaces — where contribution is estimated only from signals independent of attribution (lineage, replacement cost, failure consequence, substitution availability), and attribution is read as a vector of budget, ownership, and credit. The measure stands or falls on estimating contribution without circularity (§18); [F-10] is decisive partly because it sidesteps the estimate.

[F-2] Under continued capture incentives and absent independent revaluation, the expected direction of the contribution deficit is widening over decision cycles; the deficit is closed structurally only by clarity external to the incentive equilibrium. (The counter-case — a deficit that narrows without external clarity — is itself a test of the conditions.)

[F-3] The claim-to-accountability asymmetry of a capability — active claimants per accountable owner — predicts its probability of being cut and the latency of the decision; high-asymmetry, still-productive capabilities are cut disproportionately. Countable from a short survey; outcome and latency are in the portfolio’s own history.

[F-4] The ownership-silence test. Where importance is affirmed unanimously and no named owner accepts accountability, the probability of a destructive outcome — non-investment, shadowing, or elimination — rises materially, despite agreement on the capability’s value. (See [F-C2].)

[F-5] The discovery gap. The shadow set is enriched for high-contribution applications, found by lineage-tracing, controlling for age, scope, and criticality. (See [F-C1].)

[F-6] The internalization wedge. The price paid to acquire an external capability exceeds the internal attribution it subsequently receives. (See [F-A1].)

[F-7] The build-vs-buy inversion. Standalone ROI positive and fully-loaded ROI negative drives buy-not-build; the trace is rejected-build-then-external-acquisition of the equivalent capability. (See [F-A2].)

[F-8] The rationalization scar. Critical capabilities show more fragmentation, undocumentation, and persistence than non-critical ones; where logs survive, the fragmentation traces to past failed cuts. (See [F-C3].)

[F-9] Capability loss is informational, not performance-driven, within the class examined. Viable capabilities survive the original real shock and are then lost where answerability entropy — not measured performance failure — drives the decision. Testable by separating, in a rationalization record, losses driven by measured performance failure from those driven by ownership/value indeterminacy.

15. The decisive experiment

[F-10 / decisive] A controlled comparison of two governance treatments — clarity-only versus revaluation-then-clarity — on matched portfolios. Prediction: clarity-only increases erroneous capability destruction; revaluation-then-clarity reduces it. Controls: performance, cost, duplication, obsolescence, criticality, sponsorship, replacement availability. (See §11; protocol in the companion paper.) If this result holds consistently across portfolios, organizations, and transformation types, the mechanism is difficult to dismiss as a re-description of adjacent theories.

16. Fenced macro claims (conjecture)

[conjecture] The following are offered for examination, not asserted, and are vulnerable to a strong rival (trade and comparative advantage) the firm claims do not face:

• An economy courts external capability while extracting internal (the FDI/domestic-founder asymmetry).
• A maturing economy accretes shadow capabilities (unrepresented practices); a scaling one sheds them.
• Non-tradable capabilities are lost for want of a willing carrier, the unaccounted burden rising with reputation; the loss is filled by imported labour. The distinguishing fingerprint is that abandonment intensifies where reputation is highest.

Part VII — Positioning and method

17. Adjacent literature (positioning, not premises)

These locate the contribution; none is a premise. What is new here is not the individual ingredients — incomplete contracts, strategic opacity, and capability theory are all established — but their specific causal composition (incomplete contribution-visibility → attribution distortion → ownership and funding misallocation → destruction or non-scaling of load-bearing capability) and the measurable predictions it yields, above all the clarity-only-versus-revaluation experiment ([F-10]).

• Incomplete contracts, hold-up, property rights (Grossman, Hart, Moore; Williamson) — adjacent to §6, inverted: the dependent party keeps the dependency implicit to retain a rent.
• Transaction-cost theory of the firm (Coase; Williamson) — §7 is its dark complement: internalizing destroys the price signal even as it saves transaction costs.
• Resource-based view, dynamic capabilities, and the architecture of the firm (Penrose; Nelson & Winter; Teece; Jacobides; Baldwin on modularity; Felin & Foss) — the value and non-substitutability of organizational capability, and the boundary and architecture of who owns what; adjacent to §§6, 7, 9.
• Strategic opacity and the value of information (Hirshleifer; Grossman–Stiglitz; Gorton and Holmström on secrecy and informationally-insensitive arrangements) — adjacent to §9; the novelty is the threshold at which the private value of clarity changes sign for both parties.
• Zombie lending (Caballero, Hoshi & Kashyap) — adjacent to §9; their zombies receive subsidized credit, which is why the bilateral-blur logic is framed as continuation-beats-recognition, not high yield.
• Informal institutions, tacit knowledge, embeddedness (North; Polanyi; Granovetter) — adjacent to the economy conjecture (§13).
• Comparative advantage and trade — the named rival the economy conjecture must beat, and largely does not, on the tradable margin.
• The realization tradition and the financial-fragility tradition (Marx; Kalecki; Minsky; Koo) — adjacent only to the quarantined monetary appendix.

18. Methodological notes

• The firewall. The firm-level theory is independent of any monetary theory and of any early-warning / regime-detection framework. That independence is deliberate: a result that stands without them is harder to dismiss, and a convergence between independently derived lines is evidence only if the lines were in fact independent. Any such convergence is reserved for a separate paper.
• Witness is motivation; measurement is proof. The author’s two-sided position warrants taking the pattern seriously; the predictions of Part VI are what survive a skeptic.
• The central measurement problem: estimate contribution independently of attribution, or the gap is built in by construction. This is the paper’s currently exposed flank. Contribution must be estimated only from signals that do not also drive attribution — lineage, replacement cost, failure consequence, substitution availability — never from budget, headcount, or recognized credit. The companion measurement paper establishes inter-rater reliability on the contribution estimate (a “Study-0” gate) before any contribution number is published, and [F-10] is the decisive test partly because it is an intervention comparison that sidesteps the contribution estimate. The widening claim [F-2] is the most exposed of all, since it requires the estimate at two points in time.
• Anonymization. Field cases are rendered without identifying companies, programmes, or persons — both for confidentiality and because a named war-story reads as grievance, which weakens scholarly credibility while the anonymized pattern keeps its force.
• Tendencies, not laws. The ratchet, the masking, the build-vs-buy inversion, and the blurred-not-killed path are tendencies under stated conditions; counter-cases are admitted and are themselves tests.
• “Entropy” is non-thermodynamic (§0). The thermodynamic correspondence, if pursued, is a model in Appendix A, not an identity.
• The counterfactual is not claimed. The most damaging loss — diamonds never built — cannot be measured directly; it is anchored only on the rejected-build-then-buy proxy.

Appendix A — The monetary and thermodynamic correspondences (quarantined)

Two optional lenses, held apart from the argument so that an objection to either cannot touch the firm-level theory.

A.1 The monetary reading. The two-properties asymmetry of §2 has a monetary statement: exchangeable stored value is an accepted claim; only generative capacity produces new value; financial claims cannot raise the production ceiling, only capability can; and in a consolidated boundary the net of financial claims is zero, so aggregate real losses fall, after claims reprice, on the unencumbered capability rather than on the net-zero claims. This reading is coherent but borrowed and partly vulnerable; it is not required for anything in Parts I–VI.

A.2 The thermodynamic reading. A functional architecture has an exergy-like quantity: a capacity to produce valid, sellable information as output. The answerability entropy then behaves like a disorder term that, under the capture incentive, rises over the system’s maturity, until the system liquidates value still latent in its initial architecture. This is a model, not an identity; the formal correspondence (including any explicit entropy functional over candidate-owner distributions) is deferred to the companion measurement paper.

Appendix B — Illustrative field notes (anonymized; tagged)

These illustrate the mechanism; they do not carry the proof, and each is [Illustration] unless marked [Conjecture].

• [Illustration] The undocumented legacy layer. Years after a migration to a modern platform, a legacy system still performs irreplaceable transformations in a manufacturing plant, recorded — if at all — as “a small dependency, kept just in case.” (§10)
• [Illustration] The split inventory capability. A critical inventory capability in a manufacturing plant, cut once, regrown as five loosely-coupled legacy fragments plus an unused modern execution system: a rationalization scar. (§10, [F-C3])
• [Illustration] Killed in absentia. A social-platform deployment cut in a meeting its owner could not attend — because he was, at that hour, delivering the very capability being cut; then served in the shadows against orders, and finally executed two years later. The mechanism in one scene; not a claim about why the firm declined. (§10)
• [Illustration] The call from headquarters. A regional public-private infrastructure programme, advanced and partly built, cancelled by a call from headquarters mid-meeting with its committed partners. The periphery receiving a central decision. (§10)
• [Conjecture] The imported repairman. A town built for generations on a craft, its non-tradable residual now filled by imported labour while the high-status, lucrative work goes unanswered locally: the carrier’s unaccounted burden, rising with reputation. Offered as a field note with its rival (trade) named. (§13)

End of working paper (v8). The title, the section titles, and the [F-] numbering are working choices. Open items: a companion measurement paper specifying the protocol for [F-10], the normalization of contribution against attribution within a decision boundary, and an inter-rater-reliability gate on the contribution estimate (the central measurement problem, §18); verification of any confidentiality constraints before the field notes are published in any form; a decision on whether Part V (the macro conjecture) is better lifted out as a separate field note; and a separate paper, if warranted, on the convergence of this independently derived account with adjacent lines of work.

— Iván Abril Palma