<!--
Created 2026-10-12 under ASSEMBLY-WP-0004-T01 (causal-loop diagram) and
ASSEMBLY-WP-0004-T02 (link signs and dependencies). Base: GADM-0.1
(GenerativeAssemblyDynamicsModel.md). Links marked "0.2" are the information-
organism additions adopted in GADM-0.2.md.
-->

# GADM Causal Structure — Causal-Loop Diagram and Link Table

**Status:** v0.1 for GADM-0.1 plus the GADM-0.2 additions, exploratory.

## 1. Causal-loop diagram

Solid arrows are GADM-0.1. Dotted arrows are the GADM-0.2 organism
additions. Edge labels give the polarity: `+` means same direction, `−` means
opposite direction.

```mermaid
flowchart LR
  H[H Human capability] -->|+| V
  X[X Assembly commons] -->|+| V
  F[F Risk capacity] -->|+| V
  G[G Effective agency] -->|+| V
  N[N Diversity & composability] -->|+| V
  Q[Q Epistemic capacity] -->|+| V
  P[P Domination] -->|−| V
  V((V Experimentation)) -->|+| VS((V_sel Selected))
  SIG((σ Selection quality)) -->|+| VS
  VS -->|+ via ρ| X
  VS -->|+| K[K Productive capacity]
  K -->|+| Y((Y* Output))
  E[E Ecological headroom] -->|+| Y
  Y -->|+| I((I Claims))
  I -->|+| D((D Demand))
  S[S Social cohesion] -->|+| D
  D -->|+| SIG
  I -->|+ existence security| G
  F -->|+ restartability| G
  P -->|−| G
  VS -->|+ private return| P
  P -->|+ network effects| P
  Q -->|− checks| P
  X -->|− interoperability| P
  K -->|− throughput| E
  G -->|+| S
  P -->|−| S
  X -->|+| H
  S -->|+| H
  X -->|+| Q
  P -->|−| Q
  V -->|− if V > Q| Q
  V -->|+ entry| N
  P -->|−| N
  X -->|+ standards| N

  Ko[K_Ω Organism capability] -.->|+| V
  Ko -.->|+ learning with use| Ko
  VS -.->|+| Ko
  Ko -.->|+ membrane & lineage control| P
  Ko -.->|− deskilling| H
  Ko -.->|− lock-in| G
  Ko -.->|− monoculture| N
  Ko -.->|− labor share| I
  Ko -.->|− compute draw| E
  Ko -.->|− extraction| A[A Attention headroom]
  A -.->|+ adoption| SIG
  A -.->|+| H
  A -.->|+| S
  I -.->|+ recovery| A
```

## 2. Loops

| ID | Type | Path | Reading |
|---|---|---|---|
| R1 | reinforcing | X → V → V_sel → X | **Generative reuse.** Commons enable experiments, and selected experiments become commons. Strength depends on ρ. |
| R2 | reinforcing | V_sel → K → Y* → I → D → σ → V_sel | **Demand-backed growth.** Works only while claims stay broad. |
| R3 | reinforcing | V_sel → P (private return) → P (network effects) | **Success becomes control** (GADM-0.1 § 13). Path closure when unchecked. |
| R4 | reinforcing | Y* → I → G → V → V_sel → K → Y* | **Agency-backed growth.** Existence security widens who can attempt. |
| R5 | reinforcing | G → S → D → σ → V_sel → K → Y* → I → G | **Cohesion.** Legitimate participation sustains demand and selection. |
| R6 | reinforcing | X → Q → V → V_sel → X | **Epistemic commons.** Open evidence improves the experiments that feed the commons. |
| R7 (0.2) | reinforcing | K_Ω → K_Ω (learning with use) | **Organism learning.** Organisms improve fastest where they are used most. |
| R8 (0.2) | reinforcing | K_Ω → P → G, S → D → V_sel → K_Ω | **Lineage concentration.** Combined with R7, this is the natural-monopoly loop of the mapping note's open question 1. |
| B1 | balancing | P → V → V_sel → P (private return) | **Domination throttles its own source.** High P shrinks experimentation, which slows the success that feeds P. Weak relative to R3. |
| B2 | balancing | K → E → Y* → I → D → V_sel → K | **Ecological limit.** Throughput erodes headroom and caps output. |
| B3 | balancing | V → Q → V (when V > Q) | **Epistemic brake.** Experimentation that outruns model correction degrades Q, which slows V (GADM-0.1 § 15). |
| B4 | balancing | Q → P → Q (checks) | **Accountability.** Epistemic capacity checks domination, and domination captures information. In the model, a reinforcing pair in disguise: whichever wins compounds. |
| B5 (0.2) | balancing | K_Ω → A → σ → V_sel → K_Ω | **Attention limit.** Extraction depletes the attention that adoption and selection need. |
| B6 (0.2) | balancing | K_Ω → I (labor share) → D → V_sel → K_Ω | **Demand erosion.** Automation without broad claims undermines its own demand. |

## 3. Link table

`Conf.` is confidence in the sign: **H** (stated in GADM-0.1 or definitional),
**M** (plausible, inferred from GADM text), **L** (contested). `Src.` is the
source section in GADM-0.1, or the mapping note (`MN`).

| # | Source | Target | Sign | Mechanism | Conf. | Src. |
|---|---|---|---|---|---|---|
| 1 | H | V | + | capability to attempt | H | § 5 |
| 2 | X | V | + | reusable components | H | § 5 |
| 3 | F | V | + | risk-bearing capacity | H | § 5 |
| 4 | G | V | + | effective freedom to attempt | H | § 5 |
| 5 | N | V | + | recombination potential | H | § 5 |
| 6 | Q | V | + | model quality | H | § 5 |
| 7 | P | V | − | gatekeeping | H | § 5 |
| 8 | V | V_sel | + | definitional | H | § 6 |
| 9 | σ | V_sel | + | definitional | H | § 6 |
| 10 | V_sel | X | + | preservation via ρ | H | § 7 |
| 11 | P | X | − | enclosure L_X | M | § 7 |
| 12 | V_sel | K | + | commercial conversion | H | § 8 |
| 13 | K | Y* | + | definitional | H | § 8 |
| 14 | E | Y* | + | ecological constraint f(E) | H | § 8 |
| 15 | Y* | I | + | claims on output | H | § 9 |
| 16 | I | D | + | purchasing power | H | § 10 |
| 17 | S | D | + | social confidence | H | § 10 |
| 18 | D | σ (market) | + | adoption | M | § 19 |
| 19 | I | G | + | existence security | H | § 12 |
| 20 | F | G | + | restartability | M | § 12 |
| 21 | P | G | − | domination term | H | § 12 |
| 22 | V_sel | P | + | private return → control | H | § 13 |
| 23 | P | P | + | network effects, capture | H | § 13 |
| 24 | Q | P | − | checks (audit, evidence) | M | § 13 |
| 25 | X | P | − | interoperability | M | § 13 |
| 26 | K | E | − | physical throughput | H | § 14 |
| 27 | G | S | + | agency, future opportunity | H | § 17 |
| 28 | P | S | − | domination, lack of voice | H | § 17 |
| 29 | X | H | + | education, transfer T_g | M | §§ 11, 18 |
| 30 | S | H | + | care, participation | M | § 11 |
| 31 | X | Q | + | open evidence, research infrastructure | M | § 15 |
| 32 | P | Q | − | information capture | H | § 15 |
| 33 | V | Q | − (if V > Q) | experimentation outruns correction | L | § 15 |
| 34 | V | N | + | entry | M | § 16 |
| 35 | P | N | − | monoculture, lock-in | H | § 16 |
| 36 | X | N | + | open standards | M | § 16 |
| 37 | K_Ω | V | + | cheap experiments, gated by Q | M | MN § 3.1 |
| 38 | K_Ω | K_Ω | + | learning with use | M | MN § 1 |
| 39 | K_Ω | P | + | membrane control, lineage concentration (moderated by m, L) | M | MN § 3.5 |
| 40 | K_Ω | H | − | deskilling (moderated by m) | L | MN § 2 |
| 41 | K_Ω | G | ± | access (+, with m) vs. lock-in (−, without m) | L | MN § 2 |
| 42 | K_Ω | N | − | monoculture (moderated by L) | M | MN § 2 |
| 43 | K_Ω | I (labor share) | − | automation | H | MN § 3.4 |
| 44 | K_Ω | E | − | compute draw | M | MN § 4 |
| 45 | K_Ω | A | − | extraction (moderated by x_att) | M | MN § 4 |
| 46 | K_Ω | Q | ± | audit/research (+, with expl) vs. opacity (−) | L | MN § 2 |
| 47 | A | σ (market) | + | adoption needs attention | M | MN § 4 |
| 48 | A | H, S | + | self-directed development, participation | M | MN § 4 |
| 49 | I | A | + | economic security frees attention | L | new |

### Contested links (Conf. L)

- **#33.** GADM-0.1 states the damage hypothesis qualitatively. GADM-0.2
  implements it as a degradation of Q proportional to `max(0, V − Q)`. The
  threshold form is a modeling choice.
- **#40, #41, #46.** The sign depends on institutional levers. GADM-0.2 makes
  this explicit: membrane control `m` and explainability `expl` switch the
  dominant sign.
- **#49.** Not in either source. Introduced so that precarity reduces
  attention headroom. Needs empirical grounding (WS-14).