The ground migration, executed — v5.13.0 proposed

README.md

@@ -238,7 +238,7 @@ See [`core/06-composition.md`](core/06-composition.md).
 
 For the region in which the acceptance predicate closes, the informal conservation principle admits an exact information-theoretic construction.
 
-Let `V` be the verdict induced by the task over the ground distribution.
+Let `V` be the verdict induced by the task over the deployment distribution.
 
 Then determination demand is identified as:
 

core/00-primitives.md

@@ -95,7 +95,8 @@ The apparatus holds against this without strain:
 <!-- ddd:embed id=term:poisoned-ground -->
 > **Poisoned ground** — ground that is present but false: the substrate a determination reads has
 > been corrupted, so a correct determiner resolves wrongly with full authority. The logic is
-> sound; the ground is the attack surface.
+> sound; the ground is the attack surface. The state it names: **ground that reached the act
+> false** — what arrived is wrong, not missing.
 <!-- /ddd:embed -->
 
 <!-- ddd:embed id=term:encode-verify-split -->

core/06-composition.md

@@ -172,7 +172,7 @@ judgment into encoding, and an **inheritance path** from encoding to the next ru
 A composed arrangement executing one act is the two primitives twice over. From outside the
 declared boundary, the composite is **one actor**: one capability envelope (`04`), one verdict
 owed. Inside the boundary, the wiring — who reads what, which member resolves which slice, what
-sits at the seam — is declared structure, exactly as the decomposition and the ground
+sits at the seam — is declared structure, exactly as the decomposition and the deployment
 distribution are. Nothing about being composed requires anything beyond the act.
 
 Synchronic does not mean timeless. Sequence exists within the episode — members fire in order,

core/09-the-measure.md

@@ -28,7 +28,7 @@ exposition (canon authority lives in the claim files; see `core/claims/README.md
 | §7 | The construction is available to an arrangement exactly where its predicate closes; existence and estimability are separate conditions | `DDD-measure-16` |
 | §7 | The construction's domain and the non-zero-floor region coincide — principled, not evidential | `DDD-measure-17` |
 | §7, §9 | The measure prices the verdict, not the search | `DDD-measure-11` |
-| §9 | Demand is relative to the ground distribution | `DDD-measure-12` |
+| §9 | Demand is relative to the deployment distribution | `DDD-measure-12` |
 | §2.1 | Admissibility: the engineering reading holds for admissible `X` only | `DDD-measure-15` |
 | §9 | The chain rule iterates; the conditional term is an internal seam | `DDD-measure-14` |
 | §7a | Outcome and verdict: the determinate's two registers | `DDD-frame-14` |
@@ -63,7 +63,7 @@ So demand is not a count. It is a **measure**. And the measure is **Shannon info
 > arrangement (`term:closure`; *decidable* is the formal special case, not the requirement),
 > the predicate evaluates outcomes, and the **task class** supplies one correct output per
 > input point. The **verdict** is that induced assignment — the correct output over each point
-> of the input space. Let `P` be the distribution over inputs (the *ground distribution*). The
+> of the input space. Let `P` be the distribution over inputs (the *deployment distribution*). The
 > **determination demand** of the task is the Shannon entropy of the verdict:
 >
 > **D = H(verdict)**, measured in **bits**.
@@ -266,7 +266,7 @@ form of the unification of conservation with the actor model — and it corrects
 > mere relabelled difficulty). Demand is "constant *across* actors, *allocated by* actor."** The same
 > `H(verdict)` faces every actor; the actor sets only how it splits between encoded and judgment.
 
-`H(verdict)` never mentions the actor. It is a property of the verdict function and the ground
+`H(verdict)` never mentions the actor. It is a property of the verdict function and the deployment
 distribution — the task. That is exactly why it is "fixed by the task, never by the system."
 
 ### 6.3 X = what is supplied before the act → the encode/verify split
@@ -459,7 +459,7 @@ concession. A distribution-free quantity cannot express it.
 Second, the objects differ. MDL is computable — that is its purpose — but it prices the description of
 a hypothesis against data. It is model-selection machinery. Verdict entropy prices the distinctions a
 task requires of whatever resolves it. The two reconcile rather than compete: by the source-coding
-theorem, `H(V)` is the expected length of an optimal description of the verdict under the ground
+theorem, `H(V)` is the expected length of an optimal description of the verdict under the deployment
 distribution. The measure proposed here is therefore itself a description-length quantity — the
 distribution-relative expectation, which is the form a deployed system faces.
 
@@ -498,7 +498,7 @@ from it (`core/11` §4), and that bound supplies the escape/judgment split that
 supply is partial: the mechanism is sufficient for escape and not necessary for it (`core/11` §7).
 
 **Brooks (1987).** "No Silver Bullet" claims an essential complexity fixed by the problem. `H(V)` gives
-that claim an exact form and a boundary: fixed by the task, the tolerance, and the ground distribution
+that claim an exact form and a boundary: fixed by the task, the tolerance, and the deployment distribution
 — and never by the actor — where the predicate closes, and only there.
 
 ### References
@@ -533,21 +533,21 @@ Three, none fatal, all required in any write-up:
    decomposition with conditioning. That identification is falsifiable and was vindicated on the
    example — but it must be claimed as a modelling result, never as a mathematical discovery.
 
-2. **Demand is relative to the ground distribution.** `H(verdict)` depends on `P(input)`; the
+2. **Demand is relative to the deployment distribution.** `H(verdict)` depends on `P(input)`; the
    example used uniform. So *"fixed by the task"* must be stated as *"fixed by the task, the
-   tolerance, and the ground distribution."* This is arguably more correct — the same validator
+   tolerance, and the deployment distribution."* This is arguably more correct — the same validator
    faces different demand in different deployment environments — but it is an added parameter, not a
    free lunch.
 
 3. **Five instances is credibility, not certification.** The identity is general (it is the chain
    rule), and it has now been exercised on five conditioning variables — decomposition, actor
    encoding, what is supplied before the act (empirically, with distractors), a decomposition
-   applied *twice*, and a decomposition under varied ground — all on closing-predicate tasks. That
+   applied *twice*, and a decomposition under varied deployment distributions — all on closing-predicate tasks. That
    is real triangulation, not a single toy. Two of the three cases this caveat previously named as
    owed are now worked: **chained seams** (`core/assets/measure-chained-seams.py`, claim
    `DDD-measure-14` — both chain orders re-split the seam and leave the parts residual invariant)
-   and **non-uniform ground** (`core/assets/measure-nonuniform-ground.py`, claim `DDD-measure-12` —
-   the identity holds exactly under three ground distributions while the demand itself moves by a
+   and **non-uniform deployment distribution** (`core/assets/measure-nonuniform-ground.py`, claim `DDD-measure-12` —
+   the identity holds exactly under three deployment distributions while the demand itself moves by a
    factor of six). **Multi-actor compositions remain owed**, and the debt is not reduced by the
    chained-seam instance: there the conditioning variables are sub-decompositions of one task, and
    the composition case conditions on *actor encodings*. Same arithmetic, distinct instance,

core/11-the-floor-mechanism.md

@@ -50,12 +50,12 @@ Effective capacity is `min(C_hold, C_resolve)`, but the two produce different es
 <!-- /ddd:embed -->
 
 <!-- ddd:embed id=term:overflow -->
-> **Hold-overflow** — the decision's governing ground does not fit. **Resolve-overflow** —
+> **Hold-overflow** — the decision's governing ground exceeds hold capacity. **Resolve-overflow** —
 > the ground fits and is held, but the bits that must be *jointly resolved* exceed resolve
 > capacity.
 <!-- /ddd:embed -->
 
-- **Hold-overflow** — the decision's governing ground does not fit; the actor decides against a
+- **Hold-overflow** — the decision's governing ground exceeds hold capacity; the actor decides against a
   *partial view*. This is **encodable away**: pre-resolve some ground into the constraint (raise
   `I(verdict;X)`) and the residual that must be held shrinks until it fits. RAG is this move. Not
   floor.

core/13-delivery.md

@@ -90,7 +90,7 @@ timeout has a *when-to-stop* alternative, predetermined by the deadline, and a *
 alternative, drawn from whatever partial state obtained.
 
 **Exhaustiveness here is provable, not enumerated.** Given the standing configuration and the
-ground, the resolution is determined or it is not; if it is not, what determines it is inside the
+ground at the act, the resolution is determined or it is not; if it is not, what determines it is inside the
 arrangement's control or outside it. Two dichotomies, three values, no remainder — and a
 dichotomy, unlike a list, cannot be counterexampled. The claim's content sits in the
 answerability of that first test, which is where its falsifier now points.
@@ -138,7 +138,7 @@ supply is produced per act; there is no act-free discharge. Governance never cho
 demand is supplied — only *by what*, chosen in advance or defaulted at the act.
 
 **Distribution-weighting — an exposition note, deliberately not a claim.** Discharge is
-distribution-weighted: demand comes due where acts concentrate, at the rate the ground
+distribution-weighted: demand comes due where acts concentrate, at the rate the deployment
 distribution `P` supplies them. This is the measure's own `P` (`core/09`) read back, a
 projection of the measure rather than a further claim, and it files as this paragraph — the
 flag is the finding.

core/assets/measure-aggregate-discharge.py

@@ -21,7 +21,7 @@
 The i.i.d. control redraws the month at every act. It must return exactly
 N*H(V) — the equality limb of the inequality, demonstrated rather than stated.
 
-Task, denomination (bits x n) and ground distribution as in measure-toy.py;
+Task, denomination (bits x n) and deployment distribution as in measure-toy.py;
 entropies are exact (exhaustive, log-space), nothing is sampled.
 Exercises DDD-frame-16 (discharge is act-indexed) as worked arithmetic.
 
@@ -31,7 +31,7 @@
 
 LOG2E = 1.0 / log(2.0)
 
-# Task: validate (M, D). VALID iff D <= days_in_month(M). Uniform ground.
+# Task: validate (M, D). VALID iff D <= days_in_month(M). Uniform deployment distribution.
 days = {1: 31, 2: 28, 3: 31, 4: 30}
 points = [(m, d) for m in days for d in range(1, 32)]
 verdict = {(m, d): (d <= days[m]) for (m, d) in points}

core/assets/measure-mdl-demo.py

@@ -25,7 +25,7 @@
 projected rate-split (DDD-cost-03). Coefficients stipulated, not measured.
 
 Task (identical to measure-toy.py figures): (M, D), M in 1..4, D in 1..31,
-uniform ground, verdict VALID iff D <= days(M), days = {31, 28, 31, 30}.
+uniform deployment distribution, verdict VALID iff D <= days(M), days = {31, 28, 31, 30}.
 """
 
 import itertools

core/assets/measure-nonuniform-ground.py

@@ -1,12 +1,12 @@
 """
-Non-uniform ground for "Specification Demand Is Verdict Entropy."
+Non-uniform deployment distribution for "Specification Demand Is Verdict Entropy."
 Re-runs the date task under skewed input distributions and verifies the identity
 
     H(V) = I(V;S) + H(V|S)
 
 per deployment: same task, same code, same decompositions — demand moved by the
-ground distribution alone. Exercises DDD-measure-12 (demand is relative to the
-ground distribution) as a worked instance rather than a stated caveat.
+deployment distribution alone. Exercises DDD-measure-12 (demand is relative to the
+deployment distribution) as a worked instance rather than a stated caveat.
 Task and denomination (bits x n) as in measure-toy.py; entropies are exact
 (weighted, exhaustive), nothing is sampled.
 
@@ -29,7 +29,7 @@ def H(pairs):
     return -(p * log2(p) + (1 - p) * log2(1 - p))
 
 def analyze(split, weight):
-    """(parts, seam, whole) in bits x n under the weighted ground distribution."""
+    """(parts, seam, whole) in bits x n under the weighted deployment distribution."""
     all_pairs = [(weight(p), verdict[p]) for p in points]
     tot = sum(w for w, _ in all_pairs)
     Hv = H(all_pairs)
@@ -42,7 +42,7 @@ def analyze(split, weight):
         Hc += (sum(w for w, _ in pairs) / tot) * H(pairs)
     return n * Hc, n * (Hv - Hc), n * Hv
 
-# Deployments: the ground distribution the task faces, nothing else varied.
+# Deployments: the deployment distribution the task faces, nothing else varied.
 deployments = [
     ("benign      (valid 9x invalid)", lambda p: 9.0 if verdict[p] else 1.0),
     ("uniform     (the worked example)", lambda p: 1.0),

core/claims/DDD-frame-17.yaml

@@ -57,7 +57,7 @@ evidence:
     ref: core/13-delivery.md §4
     note: derivation by dichotomy; premises by ID and steps in notes
 falsifier: |
-  An outcome-relevant alternative at a completed act for which there is no fact of the matter whether the standing configuration and the ground together determine the resolution; or a completed act with an outcome-relevant alternative resolved by nothing at all.
+  An outcome-relevant alternative at a completed act for which there is no fact of the matter whether the standing configuration and the ground at the act together determine the resolution; or a completed act with an outcome-relevant alternative resolved by nothing at all.
 test: |
   Coding reliability over recorded acts: independent coders assigning every completed act's outcome-relevant alternatives to predetermined, exercised, or drawn without remainder, without collapsing the discharge coding into the store coding, and agreeing on where the fixing test is answerable.
 owner: none

core/claims/DDD-measure-01.yaml

@@ -2,7 +2,7 @@ format: 1
 id: DDD-measure-01
 kind: empirical
 statement: |
-  Specification demand is verdict entropy: for a task whose acceptance predicate closes, the demand engineers experience as specification burden is H(V) over the ground distribution.
+  Specification demand is verdict entropy: for a task whose acceptance predicate closes, the demand engineers experience as specification burden is H(V) over the deployment distribution.
 status: projected
 region: closing predicates only
 evidence:

core/claims/DDD-measure-11.yaml

@@ -3,7 +3,7 @@ id: DDD-measure-11
 kind: conceptual
 statement: >
   The measure prices the verdict, not the search: H(verdict) is a property of the verdict
-  function and the ground distribution and says nothing about the cost of computing a
+  function and the deployment distribution and says nothing about the cost of computing a
   correct answer. Two tasks with identical verdict entropy can differ unboundedly in
   generation cost, so the measure must not be read as pricing generation.
 status: reported

core/claims/DDD-measure-12.yaml

@@ -2,8 +2,8 @@ format: 1
 id: DDD-measure-12
 kind: conceptual
 statement: >
-  Demand is relative to the ground distribution: H(verdict) depends on P(input), so "fixed
-  by the task" must be read as "fixed by the task, the tolerance, and the ground
+  Demand is relative to the deployment distribution: H(verdict) depends on P(input), so "fixed
+  by the task" must be read as "fixed by the task, the tolerance, and the deployment
   distribution" — the same validator faces different demand in different deployment
   environments.
 status: reported
@@ -15,7 +15,7 @@ evidence:
 - kind: asset
   ref: core/assets/measure-nonuniform-ground.py
   note: >-
-    the date task under three ground distributions — uniform, adversarial (invalid 9x valid), and
+    the date task under three deployment distributions — uniform, adversarial (invalid 9x valid), and
     benign. Whole demand moves from 25.493 to 96.639 to a sixth of uniform while the identity
     holds exactly in every deployment, and a fixed decomposition's seam share moves with the
     distribution (41% benign, 57% uniform, 75% adversarial). Re-run 2026-08-17 after promotion

core/claims/DDD-measure-16.yaml

@@ -2,7 +2,7 @@ format: 1
 id: DDD-measure-16
 kind: formal
 statement: |
-  The verdict-entropy construction is available to an arrangement exactly where the task's acceptance predicate closes for that arrangement; availability is a property of the arrangement, not of the task, and it is neither the existence of the verdict function nor the estimability of the ground distribution.
+  The verdict-entropy construction is available to an arrangement exactly where the task's acceptance predicate closes for that arrangement; availability is a property of the arrangement, not of the task, and it is neither the existence of the verdict function nor the estimability of the deployment distribution.
 status: established
 region: |
   Arrangements, not tasks: closure is closure for an arrangement (term:closure), so availability
@@ -21,7 +21,7 @@ region: |
   exists until the task supplies a tie-breaker, a canonical form, or a declared selection
   distribution.
 
-  ESTIMABILITY. H(V) is taken with respect to P, the ground distribution (core/09 §2), so P must
+  ESTIMABILITY. H(V) is taken with respect to P, the deployment distribution (core/09 §2), so P must
   be known well enough to estimate. Where the deployment distribution is unknown, unstable,
   non-stationary, or only partly observable, the demand is well defined and cannot be computed
   (DDD-measure-12). A verdict function alone does not deliver a number.

core/graph/terms.yaml

@@ -86,7 +86,8 @@ terms:
     canonical_md: |
       > **Poisoned ground** — ground that is present but false: the substrate a determination reads has
       > been corrupted, so a correct determiner resolves wrongly with full authority. The logic is
-      > sound; the ground is the attack surface.
+      > sound; the ground is the attack surface. The state it names: **ground that reached the act
+      > false** — what arrived is wrong, not missing.
   - id: term:granularity-bound
     term: granularity-bound
     aliases: [granularity bound]
@@ -384,7 +385,7 @@ terms:
       > arrangement (`term:closure`; *decidable* is the formal special case, not the requirement),
       > the predicate evaluates outcomes, and the **task class** supplies one correct output per
       > input point. The **verdict** is that induced assignment — the correct output over each point
-      > of the input space. Let `P` be the distribution over inputs (the *ground distribution*). The
+      > of the input space. Let `P` be the distribution over inputs (the *deployment distribution*). The
       > **determination demand** of the task is the Shannon entropy of the verdict:
       >
       > **D = H(verdict)**, measured in **bits**.
@@ -441,7 +442,7 @@ terms:
     established_by: 11-the-floor-mechanism.md
     status: settled
     canonical_md: |
-      > **Hold-overflow** — the decision's governing ground does not fit. **Resolve-overflow** —
+      > **Hold-overflow** — the decision's governing ground exceeds hold capacity. **Resolve-overflow** —
       > the ground fits and is held, but the bits that must be *jointly resolved* exceed resolve
       > capacity.
   - id: term:escape-mechanism