Coverage

Why three maps and not one

A chain is a property of an object class, and this theory specifies three different kinds of object. The three chains do not share a unit, and that is checkable rather than a matter of taste: one moves downward in abstraction inside a single firm, one moves forward in time across two firms, and one moves outward from an author. Forced onto a single ladder, papers with no position on each other's chains would land at adjacent stages and imply coverage of everything in between.

The separation also keeps one number honest. The largest specification-advantage effect sizes here live on map C, and their subjects are language-model agents, not firms. On a single map they would read as evidence that specified organizations perform better. They are not evidence of that, and keeping the maps apart is what makes the difference visible.

Two papers that sit off the maps

Both are covered work. Neither is a stage on any of the three chains, for the reason given beside it.

• 2026bo is a gate in front of all three maps, not a stage on one. It supplies a three-condition criterion for deciding where an organization ends. Every chain here already assumes that boundary is settled, so this question runs before all of them rather than alongside any one of them. One dual-coded acquisition serves as the existence proof.
• 2026bc is the join to the brand-side map. The producer-observer seam terminates this stack at the rendered instance and hands off to the observer filter. That is why these three maps and the brand chain meet at exactly one node — the rendered artifact — and nowhere else.

The reading key

Owned — a paper makes a defended claim about this stage. Touched — a claim exists but is partial, or is about a neighbouring object. Ceded — deliberately not ours, by a recorded decision. Empty — nothing. Designed and unrun — pre-registered but not executed, which is a different state from empty.

Evidence grades. A executed pre-registered study with public logs · B executed but single-case or interpretive · C calibrated simulation or Monte Carlo, labelled by its own paper as not a test · D formal result, no data · — nothing.

A paper counts as coverage; an article, a post or a seed does not.

Map A — Internal configuration

The object specified is one firm's own commitments. Read against: intent → strategy → operating model → process design → procedures → execution → outcome → feedback Movement along it is downward in abstraction, inside a single entity.

#	Stage	Verdict	Papers	What is actually claimed	Evidence
1	Intent	touched	2026ag2026m	Owner intent is named as the top tier, with a distinct governor and transfer mode, and as the origin of the constraint hierarchy. There is no theory of how intent forms, or of what makes one intent better than another. It is an ontological slot, occupied and not developed.	D
2	Strategy	touched	2026m2026aj	Business model is a tier, and cross-tier allocation has a closed-form rule under Jorgensonian user costs. Neither says anything about the content of a strategy.	D
3	Operating model	owned	2026i2026h2026m2026af	The home ground on this map: a six-level specification cascade with acceptance contracts, a geometric bound proving exhaustive specification impossible, a cascade-compression result, and an observer-relative equivalence condition on configurations. Demonstration is one constructed domain.	D + B
4	Process design	owned	2026i2026m	Process contracts as unit tests, executor-invariance, and rank-deficient projection between tiers.	D
5	Procedures	owned	2026i2026ar	Where the cascade bottoms out, plus a six-level diagnostic that examines each level for what a healthy specification looks like. Worked on the same constructed case.	D + B
6	Execution	ceded	—	Ceded by decision, and the reasoning travels with it, because a settled position read as a hole is a different thing from a gap. Execution is not a tier; it is the rendering of the bottom tier by the tier below, an axis orthogonal to the stack. A running instance has no stable identity and no transfer mode, so it fails the tier test.	—
7	Outcome	designed and unrun	2026an2026am	Designed and unrun — which is a different state from empty, and the difference is the credit. An information-theoretic friction-tax model and a full archival identification template exist (continuous within-firm treatment, staggered difference-in-differences, regulatory instruments, event studies, Oster bounds). The estimates have never been run. What is reported is a Monte Carlo mechanism test and a powered placebo that bounds the measure: the index does not separate 100 zero-activity from 100 matched operating filers, d = .166, p = .243.	C
8	Feedback	owned	2026ae2026bn2026ar2026bd	The strong end. Verification as a spectral projection whose rank bounds what an organization can detect about itself; conventional audit as a degenerate rank-1 projection; an effective-sample-size treatment showing randomly oriented evaluators saturate, so diversity must be engineered rather than sampled; a six-level audit protocol; and an optimization-depth instrument executed on a pinned 40-organization filing panel. Bounded by its own pre-registered floors: 4 of 35 organizations resolve, so the decoupling hypotheses demote to descriptive reporting by rule. It measures the public record, never internal ground truth.	D + A

Arrow	Status
intent → strategy	Unclaimed.
strategy → operating model	Owned as a formal operator, with rank deficiency and a bounded information loss. No data.
operating model → process design	The same: a cascade junction, formal, no data.
process design → procedures	The same, plus the constructed demonstration.
procedures → execution	Ceded — see the execution stage.
execution → outcome	Empty. This is the arrow every practitioner reading silently assumes.
outcome → feedback	Measurement of the described organization, not of outcome. The arrow as drawn is not tested.

Of the seven arrows on this map, none carries field evidence. Four are owned as formal results, which is a materially different posture from the brand-side chain map — where the corpus owns zero of its seven arrows even as theory. The difference should not be flattened in either direction: this side owns cascade arrows as mathematics and none as measurement.

Map B — Cross-firm transfer

The object specified is what crosses an ownership boundary. Read against: screen → diligence → value → structure → close → integrate → realise Movement along it is forward in time, across two entities.

#	Stage	Verdict	Papers	What is actually claimed	Evidence
1	Screen	owned	2026bi	A pre-close negative screen on one failure mode, tested on 350 completed transactions. Conditional on no detected gap, the cascade-type failure rate is .073, with an exact upper 95% bound of .112 — low, but not zero.	A
2	Diligence	owned	2026ag2026ah2026bm2026ai	Theory. A six-tier ontology with per-tier governors and transferability modes, a separability diagnostic, a persistence column separating sale from succession, and portfolio-capacity and recovery-salvage instruments.	D
3	Value	touched	2026ai2026aj	A convex-kinked value geometry with a slope discontinuity at a separability threshold, and a cross-tier allocation rule. Both formal; neither estimated.	D
4	Structure	owned	2026bj	Theory, and it is strong. An acquisition reframed as a typed bundle of fork operations over the six tiers, recovering all fourteen named deal types as signature instances of a generative rule rather than as an inductive list.	D
5	Close	owned	2026bi2026bj	The cascade gap is defined at closing, and it is measured on real deals.	A
6	Integrate	touched	2026ag2026bi	Failure-propagation propositions along the service hierarchy, as theory; one of them tested, at bounded strength.	D + A
7	Realise	empty	—	No post-deal performance result exists anywhere in the corpus. The one field study’s outcome is the occurrence of a single failure type, not value realised.	—

Arrow	Status
screen → diligence	Unclaimed.
diligence → value	Empty. Value is modelled from tier structure, but nothing links a diligence finding to a price.
value → structure	Empty. The structure work is structure-side only; it does not price a bundle.
structure → close	Formal: which bundles are admissible, and what a transaction leaves behind.
close → integrate	The one arrow in the whole of OST with field evidence — and it is a necessary-condition claim at half strength. Necessity consistency is .517, so about half of the failures carried no gap. A bounded safe harbour and a negative screen, never a predictor of deal success.
integrate → realise	Unclaimed.

Of the six arrows on this map, exactly one carries field evidence, and it is a necessary-condition claim at bounded strength.

Map C — Knowledge artifacts

The object specified is a document, a claim, a research result. Read against: author → specify → render → transmit → verify → reuse Movement along it is outward from an author.

#	Stage	Verdict	Papers	What is actually claimed	Evidence
1	Author	touched	2026ao2026bg	Cost asymmetry across an artifact’s layers, and invention as a typed structural operation.	D
2	Specify	owned	2026t2026ao	A machine-readable standard for scientific claims.	D + B
3	Render	owned	2026ap	Executed: spine preservation and rendering equivalence, demonstrated on independently authored renderings.	A
4	Transmit	owned	2026u2026w	Evaluation infrastructure, and canon as a repository.	D
5	Verify	owned	2026bh2026bl2026bk2026at	Executed: blinded multi-model coding with inter-coder reliability, a pre-registered decomposition of extraction disagreement, a deterministic six-class compatibility check for federated vocabularies, and an internalization study whose validation failed at its own gate — which is recorded rather than quietly dropped.	A
6	Reuse	touched	2026aq	Executed, on a non-firm subject. Specification-first prompting beats interpersonal style once value headroom exists: d = .314, p = .009 at mid capability, d = .569, p < .001 at the frontier. Ablation shows the advantage comes from teaching logrolling, it survives paraphrase, and it replicates across model families. The subject is a dyad of language-model agents.	A

This is the only map where executed studies outnumber formal results. It is also the only one whose subjects are not organizations — they are documents, claims and language-model agents. Nothing here is evidence about how a firm performs.

The arrow question

Asked plainly: is there any field evidence that a specified organization performs better?

Three things come near it, and each falls short in a different way. The differences are the useful part, and none of the three may stand in for the claim.

1. The design exists; the estimate does not. The specification-readiness work builds the measure and the full identification template, and reports a Monte Carlo and a powered placebo. Its own text flags the archival execution as the next deliverable.
2. Real field data, different arrow, half strength. The one large-N field study tests whether a closing-time structural gap is necessary for one integration-failure mode, and finds a bounded safe harbour with necessity consistency .517. That is failure-avoidance on one mode, not performance — and the paper says so in its own voice.
3. A clean causal specification advantage, on the wrong subject. The most quotable "specification pays" result in the corpus has a dyad of language-model agents as its subject, not a firm. It belongs to map C.

The warning that matters most

The count

• Map A — of 8 stages, 4 owned (operating model, process design, procedures, feedback), 2 touched, 1 ceded by recorded decision, 1 designed and unrun. Of 7 arrows, 0 carry field evidence; four are owned as formal results.
• Map B — of 7 stages, 4 owned (screen, diligence, structure, close), 2 touched, 1 empty (realise). Of 6 arrows, exactly 1 carries field evidence, at bounded strength.
• Map C — of 6 stages, 4 owned and 2 touched, with four of the six resting on executed studies.

Across all three maps there is one arrow with field data behind it, and it is a necessary-condition claim about avoiding one failure mode.