---
title: "Privacy is Value (V6): the seventh capital is soil, and four model evolutions from the Programme runtime"
kind: research-note
series: pvm-v6
date: 2026-07-15
provenance: Programme runtime (V6 Rehydration Pipeline), the WP-14 seventh-capital arc and its lit-review runtime
evidence: pipeline/reviews/critiques_ledger.md L130, L132, L134, L137, L138, L139; pipeline/litreview/ (runs wf_ba400906-67b, wf_aea45393-332); pipeline/rehydrations/academic/weis_seventh_capital.md (draft-v3)
status: >
  The organizing reclassification (the seventh capital is land/soil-based rentier
  capital) is a First-Person ruling of 2026-07-15 and is recorded here as adopted.
  The four amendments it organises are candidate canon evolutions; their propagation
  into the spec and the register is the register process, not this note.
supersedes_nothing: true
---

# The seventh capital is soil

This note records what the Programme runtime surfaced about the Privacy is Value
model itself, as distinct from the WEIS paper (WP-14) that occasioned it. It was
generated the way the Programme was meant to generate it: by running the research
harness on real numbers and against the actual literature, so that the process of
building the paper produced the evidence the model needed. The findings are traced
to the critiques ledger throughout; nothing here is asserted that the runtime did
not establish.

The organizing move is a reclassification, ruled by the First Person on
2026-07-15: **the seventh capital is land-based, rentier capital, and the right
economic analogy is data as soil.** Everything below is easier to state, and
truer, once the model stops treating behavioural data as a compounding, aggregating
asset and starts treating it as soil: a positional, rent-yielding stock with a
fertility that can be cultivated and an exposure that erodes over time.

Soil is the better analogy because it carries all four of the evolutions at once,
and because each thing the runtime found maps onto a property soil already has.

- Soil yields **rent by position**, not value by aggregation. You cannot make more
  land by piling up more dirt, and the marginal product of one more grain is
  essentially nil. That is the returns-to-scale finding exactly.
- Soil's worth is **locational**: it is set by where the plot sits relative to the
  market it can reach, not by a headline multiple. That is the appropriation-share
  finding.
- Soil **erodes**. A plot loses protective cover as the weather that works against
  it grows stronger, on a calendar clock, even when nothing is added to it. That is
  the moving-ceiling R(t), read correctly.
- You can **lease the harvest without conveying the land**. A tenant may take a
  season's crop; he cannot carry off the soil. That is structural inalienability by
  non-reconstruction.
- And the one thing that genuinely accumulates is **fertility**, built by
  cultivation. That is the reputational relationship stock, and it is the only place
  the word "compounding" survives.

The rest of this note develops each of the four in turn, with its evidence, what
it retires, and what in canon it touches.

## Evolution 1: rentier, not compounding (the soil is worked, not multiplied)

**What the model held.** Behavioural data compounds as capital. The aggregate is
super-additive: the same records are worth little atomised and much in combination,
and it is that latent compounding value the subject is denied. The prior essay
lineage carried this as an accessible-volume gap of order tens of thousands to one.

**What the runtime found (ledger L134, on the L130/L132 substrate).** The
returns-to-scale evidence refutes the super-additivity. Forecast error falls only at
the diminishing root-N rate; the data-variety dimension is robustly flat to negative
(Bajari, Chernozhukov, Hortacsu and Suzuki 2019); a small subsample suffices for a
given estimation task, so the marginal analytic value of additional volume at scale
is near zero (Varian 2014); combined-attribute broker profiling is barely above
chance (Neumann, Tucker and Whitfield 2019). The observer-to-subject value gap does
not decompose into compounding aggregation. It decomposes into an atomisation
discount, market-position rent, and lifetime accumulation. The single record's
modelling premium saturates at order one; the large gap is rent, not aggregation
value.

**What the model becomes.** Behavioural data is rentier capital, and soil is its
figure. The value the subject is denied is not a latent compounding she cannot reach
without a platform's scale; it is a positional rent she is denied because she is not
admitted to the venue where the soil's produce clears. This is not a demotion. Land
is the canonical capital asset, and four of the five capital predicates (returns,
investability, opportunity access, cross-context transfer) hold for a rent-yielding
positional stock without any appeal to compounding. Compounding survives in exactly
one place, and honestly labelled: the cultivated fertility of the relationship, the
reputational stock, which is a conjecture and not a measured aggregation of records.

**What it retires, and the honest cost.** It retires "data compounds as capital" as
a general claim and, with it, the accessible-volume gap as a valuation figure. This
is the single largest change, because "data is the new oil" and its compounding
intuition is the premise most of the field runs on, and the model now stands against
it with evidence rather than beside it with a slogan.

## Evolution 2: value is the appropriation share the interface sets, not a figure

**What the model held.** A value function with headline multiples: a present-day
per-person data-value gap of order several hundred to one, and the accessible-volume
gap above. These lived in the canonical-figures material and the essay lineage.

**What the runtime found (ledger L130, L132).** Neither multiple survives as an
asserted fact. The ratio of potential to captured value is ill-posed as the
subject's captured value tends to zero: it is a division artefact, not a
measurement, and because value is aggregation-dependent any single ratio silently
fixes a dataset composition, a model class, and a counterfactual. The
accessible-volume figure had already been reframed inside the canon from an
arithmetic ratio to a non-arithmetic difference in accessible volume on the same
manifold, and a non-arithmetic quantity is not a valuation figure at all. A later
attempt to rescue a figure by convergence of the observer's revenue flow and the
subject's reservation value was refuted within its own run as a flow-versus-stock
annualisation artefact (L132).

**What the model becomes.** Value is a bounded surplus-appropriation share, and its
position is the subgame-perfect equilibrium of the bargaining game whose rules are
the consent interface. The share is near zero under notice-and-consent read as an
ultimatum extensive form, because the subject is not admitted to the clearing venue;
it moves into an interior range under a propose-and-respond protocol (IEEE 7012) that
inverts the proposer role. The appropriation gap is then the movement of one
dimensionless share between two regimes, a difference of stated intervals, never a
ratio to a near-zero base. In the soil figure: the plot's worth is the rent its
position commands, and the interface is the road that does or does not connect it to
the market. Privacy is value now means the share and the mechanism that moves it, not
a number.

**What it retires.** The headline multiples as asserted facts. They remain lineage
and essay material behind the existing figure fence; they are not reconstructed as
measured values.

## Evolution 3: the moving ceiling is an erosion clock, and it runs on the adversary's information, not his compute

**What the model held.** The reconstruction residual R(t) is a moving ceiling: it
drifts upward over time as adversary capability grows against a fixed archive, so the
protection has a finite, adversary-relative horizon rather than a static one.
"Adversary capability" was left unspecified between compute and information.

**What the runtime found (ledger L139, the DP-refinements stress test).** The
ambiguity was hiding an over-claim, and resolving it sharpened the model. None of
eleven differential-privacy refinements runs R(t)'s clock. The differential-privacy
family already saturates compute: its guarantees hold against an adversary of
unbounded computation, so "the adversary gets more compute" does not erode an
information-theoretic guarantee. What actually drifts is the adversary's
informational capability, his accumulating linkage corpus and side priors, against a
release that is never re-queried. This is a third clock, distinct from the
differential-privacy composition clock (the budget depletes as more queries are
answered, the adversary held fixed) and from the cryptographic shelf-life clock
(Mosca's break-time, where the cipher weakens). Where differential privacy does run
a wall clock, in age-dependent and temporally-discounted formulations, its polarity
is the opposite: the guarantee strengthens as data ages through relevance decay.

**What the model becomes.** The moving ceiling is precisely typed: R(t) is a
calendar-time, adversary-informational-capability erosion clock. Soil loses its cover
to the weather, on a calendar, while nothing is added to the plot. This imposes a
requirement the model did not previously state: the guarantee the durability leg
imports (WP-07) must bound reconstruction under growing linkage and side information,
not under growing compute, or the security half and the economic half of the model
describe different objects. The distinction is stated as current-formalisation-
contingent, not a categorical barrier, because adversary-aware differential privacy
already carries adversary capability as a static parameter and computational
differential privacy already gestures informally at the same erosion.

## Evolution 4: structural inalienability is where the security guarantee becomes the ownership

**What the model held, implicitly.** The separation architecture and the
reconstruction bound were the security substrate; the seventh-capital thesis was the
economics; they were parallel tracks.

**What the runtime made explicit (WP-14 draft-v3, Section 2; the propertisation
answer).** The two tracks fuse, and the fusion is the model's answer to the oldest
objection against owning personal data. Propertisation failed for decades on
alienability: a right the subject can sell is a right she will be pressured to sell,
and once sold the protection is gone. The model's answer is not legal (a statute
making the right inalienable) but structural. What the subject discloses is a scoped,
single-purpose disclosure, not a handover of the stock, because the observing side
cannot reconstruct the underlying record from it. The stock cannot be fully alienated
by disclosure for the term R(t) permits, because disclosure does not transfer what it
cannot reconstruct. In the soil figure: a tenant leases the harvest, a scoped and
time-boxed right to a season's produce, and cannot carry off the soil itself. The
security guarantee is therefore not adjacent to the economics; it is the mechanism
that manufactures the ownability, and it hands the asset the same finite horizon it
has: the ownership is durable for a term, on the erosion clock, and not forever.

**What the model gains.** A named bridge between the moving-ceiling security material
and the seventh-capital economics: structural inalienability by non-reconstruction.
This is a candidate conjecture, and its assignment is the First Person's.

## The epistemic evolution: the model earns belief by narrowing, not by claiming

The runtime also changed how the model should carry itself. Its lit-review pass put
five broad novelty claims to an adversarial prior-art search and every one came back
a mirage: the field already owns the broad objects. It prices privacy system
properties as commodities (the differential-privacy-markets strand, Ghosh and Roth
2011 and successors); it has propertised personal data and criticised it (Laudon,
Samuelson, Schwartz, Purtova, Prins); it has proposed data-as-labour and its unions
(Posner and Weyl, Arrieta-Ibarra); it has the adversary-relative shrinking horizon
(Mosca). What survives is narrow: a specific conjunction none of them occupies. The
lesson for the model is that its credibility now comes from what it refuses to claim
and from knowing precisely which neighbour owns each broad move. The model sits
beside these literatures, not above them, and it is stronger for saying so
(ledger L137, L138).

## Canon propagation, if adopted

The reclassification is a First-Person ruling; the amendments it organises touch the
following canon surfaces, and moving them is the register process, not this note.

- **The seventh-capital thesis (register conjecture C55).** Its character changes
  from compounding-aggregation capital to land-based, rentier, soil-like capital.
  The five predicates survive; the compounding predicate is scoped to reputational
  fertility as a conjecture. The seventh-capital narrative material (the tide-orbit
  poem, the relevant Tome) carries the soil figure.
- **The canonical-figures material (spec Section 30) and the essay lineage.** The
  per-person and accessible-volume multiples retire as asserted facts, behind the
  existing figure fence; the value function is restated as an appropriation share set
  by the bargaining game.
- **The moving-ceiling material (extraction E2; the spec's reconstruction sections;
  the moving-ceiling SoK and essay).** R(t) is typed as an informational-capability
  erosion clock, with the metric requirement it imposes on the theory companion.
- **The durability and separation material (WP-07 and its extraction).** The
  guarantee is required to be an informational-capability reconstruction bound, and
  structural inalienability by non-reconstruction is named as the bridge to the
  economics.

## Honest limits and what remains the register's

- The reclassification to soil is adopted; the propagation of the four amendments
  into the spec and the register is not done here and is not assumed.
- "Compounding" is retired as a general claim but retained as a reputational-fertility
  conjecture; the model must not let that conjecture quietly re-import the aggregation
  intuition it just refuted.
- The R(t) informational-capability typing imposes a real requirement on WP-07 that
  the theory may or may not currently meet; if the proven bound is a compute bound,
  the halves diverge and the durability leg reverts to an open problem.
- The candidate bridge conjecture (structural inalienability by non-reconstruction)
  and the lit-review's CTR-series residues are candidates; conjecture assignment and
  the canon dispositions are the First Person's (GR-10). Nothing here marks a
  completion gate.

The seventh capital is soil. It yields rent by position, not value by aggregation;
its worth is the share the interface lets the subject keep; its exposure erodes on
the adversary's informational clock; and it can be leased by the season without ever
being carried away. The runtime did not invent any of this. It ran the process, and
the process returned the analogy the economics was always asking for.