THE PRESSURE-TEST CAUSAL HYPOTHESIS — drawing the arrows

Version: 1.1 · Status: The v1.0 body below retains its RATIFIED (author, Session 4e) status unmodified. Sections 8–9 are bounded S4h additions — the lag-type decomposition and the adapted evidence ladder with edge labels — stamped inline panel-delegated (S4h standing delegation), provisional-pending-author, so a reader always sees which parts carry the author’s ratification and which carry only the panel’s. · Last updated: Session 4h Source: docs/GOALS.md, panel/SESSION_2_PRESSURE_TESTS.md, docs/DIAGRAMS.md §4 (the thirteen tests). Companion to outputs/THEORY_A_OPERATIONALIZED.md §9 (Theory-A loop) and the operationalizations of A/B/C.

The project has always said the arrows between the layers are the research program — that “everything is connected” only becomes criticizable once you commit to which thing drives which, with what sign and what delay. This document does that: it proposes a specific signed causal hypothesis over the thirteen pressure-tests. It is deliberately falsifiable and deliberately contestable. Drawing it is what separates this project from the polycrisis literature, which names the entanglement but never commits the graph.

1. The signed graph (load-bearing edges only)

Not every conceivable edge — the strongest proposed couplings. “+” = same-direction; the dashed node M and its edges mark the open base-or-summit question (Q-002).

flowchart TD
    D4["D4 · machine intelligence"] -->|"+"| D1["D1 · acceleration"]
    D4 -->|"+"| D2["D2 · optimization"]
    D1 -->|"+"| Y1["Y1 · epistemic breakdown"]
    D1 -->|"+"| Y2["Y2 · coordination failure"]
    D1 -->|"+"| Y3["Y3 · institutional decay"]
    D2 -->|"+"| S2["S2 · attention economy"]
    D2 -->|"+"| Y1
    D3["D3 · wealth pump / inequality"] -->|"+"| Y3
    D3 -->|"+"| S3["S3 · populism"]
    Y1 -->|"+"| S3
    Y1 -->|"+"| S4["S4 · anomie / loneliness"]
    Y2 -->|"+"| Y4["Y4 · ecological overshoot"]
    Y3 -->|"+"| S1["S1 · fertility collapse"]
    Y3 -->|"+"| S3
    Y3 -->|"+"| S4
    S2 -->|"+"| Y1
    S2 -->|"+"| S3
    S3 -->|"+"| Y3
    M(["M · coherence vacuum — base or summit?"]) -.->|"±"| S4
    S4 -.->|"+"| M

2. The load-bearing pathways (why each arrow)

Machine intelligence as a driver of drivers. D4 → D1 and D4 → D2: AI is not a peer symptom but an amplifier — it accelerates the rate of change and hands the optimizers sharper tools. This is why “AI” appears as a driver-of-drivers, not a symptom: its causal role is to multiply two other drivers.

Acceleration surfaces in the dynamics. D1 → {Y1, Y2, Y3}: when the built world changes faster than institutions, cognition, and shared culture can absorb it, the capacity to check falls behind the capacity to fabricate (Y1), coupled systems fragment faster than they can be governed (Y2), and institutions built for a slower world lose fit and legitimacy (Y3). This trio is Theory A’s gap, distributed across subsystems.

Optimization is its own driver, with a home symptom. D2 → S2 (the attention economy is optimization’s direct product) and D2 → Y1 (engagement-optimization is a selection environment that structurally rewards unreality — the reason “post-truth” undershoots). This is Theory B’s territory.

The wealth pump drives decay and revolt. D3 → Y3 and D3 → S3: Turchin’s mechanism — elite overproduction plus popular immiseration erode state legitimacy (Y3) and fuel populist revolt (S3) — enters as a rival driver, not a special case of acceleration. Whether D3 or D1 better explains S3 is exactly the A-vs-Turchin discriminator in Theory A’s study.

Dynamics surface as symptoms. Y1 → {S3, S4} (fragmented truth breeds simple stories and disorientation); Y3 → {S1, S3, S4} (institutional thinning removes the frames that made children the default, that make institutions feel responsive, and that locate people); Y2 → Y4 (a world that cannot coordinate on long-horizon collective action overshoots its ecological limits).

3. The loops (where the system’s fate is decided)

• R1 — the reinforcing trap (Y3 ⇄ S3). Institutional decay breeds populism (Y3 → S3), and populism degrades institutional capacity (S3 → Y3): a vicious cycle. Extended through D1 → Y3, this is Theory A’s reinforcing loop — the gap widening because the response to it makes adaptation slower.
• R2 — the capture loop (D2 → S2 → Y1 ⟶ D2). Optimization produces the attention economy, which degrades shared epistemics, which weakens the collective capacity to regulate optimization — so D2 persists and deepens. Theory B’s autopoietic self-reinforcement, drawn.
• B1 — the balancing hope (delayed). Acute crisis or overshoot (Y4, or a sharp symptom) can provoke adaptive reform that restores institutional capacity (reduces Y3), damping the symptoms — Theory A’s delayed balancing loop, and the accelerationist’s counter (D-006): if B1 is fast and strong, the gap is transitional; if R1/R2 dominate and B1 is too slow, it is a trap. Which loop wins is empirical, not assumed.

4. The master condition M — drawn as a question, not an answer

M (the coherence vacuum) is deliberately drawn with dashed, bidirectional edges because its position is open (Q-002):

• M-as-base: the loss of a shared “why” comes first and makes everything downstream easier — the meaning-vacuum lowers resistance to capture (Nietzsche’s Last Man is pre-adapted to the dopamine economy), so M → S2, S4 with a weakening effect on resistance.
• M-as-summit: the drivers, dynamics, and symptoms hollow out shared meaning as their aggregate residue, so {everything} → M.

The graph refuses to commit, because the panel has not resolved it and pretending otherwise would violate the project’s own honesty. The dashed edges are the open question.

5. How the three theories live on this graph

• Theory A is the claim that the whole graph is powered by D1 (with D4 amplifying it), read through the D1 → {Y1,Y2,Y3} → symptoms pathways and the R1/B1 loops. Its test: does a measured gap predict the symptom set, controlling for D3?
• Theory B is the claim that D2 (with D4 amplifying it) and the R2 capture loop are the operative mechanism, read through D2 → S2 → Y1. Its test: does optimization intensity predict capture, and does de-optimization reverse it?
• Theory C is not a node — it is the claim that a graph like this can only be built and corrected by a plural, self-correcting commons, and that the arrows must stay revisable. C is the method that drew the graph, held to its own ablation test.

The theories are therefore not exclusive: A and B name different power sources for overlapping pathways; where they make different predictions (wide-gap/low-optimization vs narrow-gap/high-optimization domains) is exactly where their studies distinguish them.

6. What this buys, and what to test next

Drawing the graph converts “the polycrisis” from a mood into a set of checkable directed claims. The near-term research program is, literally, to test the arrows: the A study tests the D1-pathways and the D3 discriminator; the B study tests the D2 → S2 → Y1 pathway and the R2 loop via de-optimization events; the strongest single crux is D3-vs-D1 into S3/Y3 (is populism driven more by the wealth pump or by the adaptation gap?), which both studies bear on. A signed causal loop for B and for C (paralleling A’s §9) is the next diagramming step.

7. Preserved contestation (the structure itself is disputed — do not treat as settled)

• Marx: the driver/symptom cut is ideology. What the graph calls “symptoms” (populism, anomie, fertility) are the system working as designed for whoever benefits (D3 is not one driver among four — it is the point). The graph should perhaps be redrawn with the wealth pump as the root, not a peer.
• Nietzsche: M belongs at the summit, not floating dashed at the side — the coherence vacuum is the central fact, and a graph that makes it one node among fourteen has already missed it.
• Le Guin: any ordering encodes a worldview. Choosing D1/D2 as roots (a technological framing) rather than, say, a relational or ecological root is itself a value-posit; the graph is not neutral and should not pretend to be.
• Luhmann: the arrows imply steering and causation between subsystems that, on his account, cannot directly steer one another — the couplings may be structural resonances, not causal pushes, and “→” may be the wrong primitive.
• Heidegger (standing): drawing the whole as an optimizable causal diagram is itself the enframing the project is trying to understand — the map may be a symptom.

These objections are not resolved. They are the reason the graph is versioned, contestable, and forkable — and the reason the signs and delays are hypotheses, not facts.

8. Lag-type decomposition on the Y-layer edges (S4h addition — panel-delegated (S4h standing delegation), provisional-pending-author; from the S4g learning panel, adoption 2)

Every edge into or out of the Y-layer may carry a lag-type annotation: what kind of failure-to-adapt the edge asserts. This is the decomposition that makes the A-vs-D3 crux codable at the episode level — “the system cannot adapt” (Theory A) versus “capacity exists, an interest blocks it” (the wealth pump) are currently indistinguishable in the graph.

Coding semantics (resolved on the record, S4h): multi-label with a named PRIMARY type — real episodes will be multi-typed (obstruction operates through political channels and manifests as coordination failure), so mutual exclusivity would guarantee uncodability; the primary type is selected by the tie-break rule: the type whose removal would most have changed the outcome, coded from the episode record’s own evidence (Campbell’s tie-break repurposed as primary-selection). All five types enter label-only — not yet codable until an inter-coder trial on real episode records clears a pre-registered agreement floor; there exist zero coded episodes today (docs/EPISODE_UNIT.md gates all coding), so no present-tense codability claim is made — the crux becomes codable only when the episode base exists. In-house codings of any real episode are directional-only; evidential weight requires the project-blind cross-model standard (Campbell).

⚠ The S4k re-balance (the author’s ruling, decision 9 — SG-02 OBJ-1, sev 9). The S4h form of this table carried a directional thumb on the scale: obstruction required a high two-part documentary bar while “absent that, code capacity, never obstruction” made capacity the default sink for every ambiguous episode — which is Theory A’s read winning by construction over the wealth pump (D3), the project’s own strongest crux, in exactly the cases that discriminate them. That default is struck. The corrected rule is symmetric: capacity requires its own affirmative evidence (knowledge present, means absent/exhausted); obstruction requires its own (capacity present, an interest blocked it); and an episode for which neither affirmative case is documented is coded unregistered/ambiguous — never defaulted to either theory’s favor. And the D3→S3 and D3→Y3 edges are brought into the episode-codable frame (they were §9 “cross-case, external-provenance, untested-by-this-project” only): the wealth-pump channel is now coded at the episode level by the same lag-type instrument, so the A-vs-D3 crux is tested on both arms with no built-in tilt. (This is a subtractive bias-repair — removing a default, widening a scope — and is therefore exempt from the Rule-25 moratorium. It fixes the instrument; it does not fire it: with no episode base coded and no foreign vantage, the crux is codable-in-principle and untested-in-fact, per the S4k concession.)

Preserved: Nietzsche — “the system cannot adapt” and “an interest blocks it” must never again be the same sentence in this project. Ostrom — obstruction coded by feel becomes a mirror; the ex-ante discriminator is binding. Heidegger (AWR) — “five lag-types on the arrows of a map whose arrows I refuse; my objection is to the cartography, not the legend.”

9. The evidence ladder (adapted) and the edge labels (S4h addition — same stamp; from the S4g learning panel, adoption 4)

The ladder (governing every claim-bearing document — panel-confirmed under delegation, the author’s ratification outstanding; the word “canonical” corrected S4i per the shadow-grade pass, Rule 11/C-028 class): each claim carries a rung label — fact (a dated, sourced particular) · mechanism (a named causal pathway with at least one documented instance) · cross-case (a pattern across cases — valid only with stated N and provenance; a cross-case label without them is void on sight) · theory (a proposed integration whose warrant is coherence, not yet cases) · normative (a value-commitment — a different KIND of claim, not a lesser degree: placing it at the ladder’s far end must never teach a reader that a value-claim is a fact that failed — Le Guin, binding gloss) — plus an orthogonal test-status tag: untested · pre-registered · run:survived · run:falsified. Silent rung-merge — presenting a claim at a higher rung than its evidence, or merging rungs in one sentence — is a codable catch category from this session. (Theory-doc labeling is deferred to the author-present session under three binding conditions: every theory doc carries a pending-banner and no rung status may be cited from it until labeled; the labeling is due by the next substantive session’s close, else an automatic C-012-class latency catch, with the repo-open backstop; and the first labeling pass is audited against flattery — a pass in which every edge lands mechanism-or-better is itself examined as a rung-merge. Recorded per Meadows: the deferral was chosen under the standing delegation with the cheap surface labeled first; this line exists so the deferral can never later be read as pure prudence.)

The §1 edges, labeled (first pass, S4h — deliberately bottom-heavy; that is what an honest first pass of an untested graph looks like):

No edge in this graph is run:survived. One first-pass count, honest: 2 mechanism, 2 cross-case (external provenance), the rest theory/untested. The flattery guard passes by inspection.

S4k update — the theory-doc rung labels applied (C-034 discharged; the author’s ratifying pass). The §9 deferral (theory-doc labeling due by the next substantive session, else an automatic C-012-class latency catch) is met: at the author’s S4k contact, each of outputs/THEORY_A_OPERATIONALIZED.md, outputs/THEORY_B_OPERATIONALIZED.md, and outputs/THEORY_C_OPERATIONALIZED.md received a rung + test-status header. The first-pass flattery audit required by this section: no theory doc lands every claim mechanism-or-better — A is labeled theory · untested (its gap is drafted-not-run), B mechanism · pre-registered · frozen-unrun on its anchors and theory · untested on O→P, C theory · untested with its one ablation read run: null-compatible, instrument-unreliable. The distribution is bottom-heavy, which is what an honest first pass of an untested program looks like; a pass in which every theory had climbed to mechanism-or-better would itself have been the rung-merge this section warns against, and it did not occur. The re-balanced D3 instrument above (§8) is the one object-level channel whose bias was corrected in the same act.

Ratified working hypothesis (author, S4e); §8–9 panel-delegated (S4h), provisional-pending-author; contestation preserved. Author: Hulki Okan Tabak — with Claude · License: CC BY-SA 4.0.