A Systems Theory for Today

A living, forkable attempt to build a systems theory adequate to the present — by designed disagreement.

Self-audit & figures

What this project has actually produced — and, more honestly, what it has not yet. n = 4 sessions. Counts are self-reported by the same system that benefits from looking productive; “falsifiable” means a claim was stated so it could be tested, not that anything has been tested. Nothing here is empirically confirmed yet.

The canonical source of these figures is the diagram log — Mermaid text a contributor can fork and correct; this page is the curated presentation, with the counts. The interactive map is a companion piece.

The Facts — counts measured

3/14pressure-tests carry a falsifiable claim
0of those tested against data yet
0external forks · contributors · citations — the commons has not formed
Effort so far: ~47k words → 49 files → 3 falsifiable claims. Volume is not rigor.

What our own blind measures found

Some of the project's instruments have been run past project-blind external coders. Here is what those blind runs returned — including where they failed, and where they cost the project a claim. Reported flatly, as results, not as a badge. Every figure is drawn from the record (`logs/CATCHES.md`, `logs/DECISIONS_CHANGED.md`, the S4h/S4i study outputs); all of it is provisional-pending-author.

InstrumentWhat it was meant to measureWhat the blind run returned
Severity anchoringwhether a flaw is minor / substantive / conclusion-changing, agreed across codersPassed — 94% cross-family agreement on a 26-item gold set (S4h)
Catch detection / unitizationwhether coders even flag the same passages as flaws on live materialFailed — ~1% agreement, far below the 50% floor (S4i; learning L-016, ratified S4k)
The convergence rubricwhether a real deliberation reads differently from a hollow oneInvalid — it banded a stakeless "empty" session as full; a placebo falsifier lost (C-025 / DC-006)
The Theory-C ablationwhether designed disagreement out-catches a lone reasonerNo valid result — two powered reads point opposite ways (1.45× / 1.15×), both uninterpretable: one is a broken-series pre-anchoring datum, the other sits on a run whose detection failed

Two honesty notes travel with every count on this page. Reliability: the catch counts above are severity-anchored (94%) but not yet detection-anchored (~1%, candidate L-016) — so a "catch" is a reliably-graded flaw whose boundary two coders rarely agree on. One source: every "external" run here is a second large language model, which draws the same underlying aquifer (L-013) — cross-model-confirmed is not foreign-confirmed. At S4k the author closed this build: no genuinely foreign vantage — no outside reader, grader, or forker — enters it; the author is the terminal ratifier, so every external-dependent claim stays conclusively untested in this closed setup.

From 14 tests to what is actually tested. Source: METRICS §4. “Falsifiable” = a claim stated so it could be tested — not that it has been. 0 of the 3 are confirmed; the data work is future (Q-001).
1414 pressure-tests defined11~11 with a first-pass account33 with a falsifiable claim00 tested against data
Pressure-test coverage — the fourteen, by layer (thirteen across four layers, plus the reflexive R named at S4k). Verbatim from METRICS §4. Status is shown by word + glyph, not colour alone. “~11” keeps its tilde; D1 is Theory A’s flagship — design-drafted but untested.
IDTestFirst-passHome theoryFalsifiable claim yet?
Master
M Coherence vacuum · base or summit open (Q-002) Yes A / C □ Not yet
Drivers
D1 Acceleration Yes A ◫ Drafted · untested
D2 Optimization Yes B ◨ Partial
D3 Wealth pump Yes A · Turchin ■ Yes
D4 Machine intelligence Partial B / A □ Not yet
Dynamics
Y1 Epistemic breakdown Yes B + A □ Not yet
Y2 Coordination failure Yes A + C ◨ Partial
Y3 Institutional decay Yes A · Ibn Khaldun ■ Yes
Y4 Ecological overshoot Yes A + B ◨ Partial
Symptoms
S1 Fertility collapse Yes A / B □ Not yet
S2 Attention economy Yes B ◨ Partial
S3 Populism Yes A + B · Turchin ■ Yes
S4 Anomie / loneliness Partial C · Han □ Not yet
Reflexive
R Register / care fairness Named (S4k) Cross-cutting — audits the list ◫ Drafted · untested
14 defined · ~11 first-pass · 3 falsifiable · 0 tested
Three theories — the vote spread, not the score. Source: CANDIDATE_THEORIES, Round-7 ratings. The median alone is a lie of omission; the width is the designed disagreement. These are not measurements: they are one model in many roles rating itself, with no inter-rater reliability (L-015) — read the numbers as a signed intuition, never as a score.
0510A · Adaptation Gapdiagnosis · weakest: falsifiability4 Turchin9 Meadowsmed 7B · Optimization Ecologymechanism · risk: conspiracy-shape5 Kant9 Zuboffmed 7C · Distributed Coherenceresponse · exposed: the meaning-wound3 Nietzsche9 Ostrommed 6
Four sessions — the trajectory, four points not a curve. Source: METRICS §2 snapshots. n = 4, plotted as four labelled points. Dissents-preserved is flat since S2; words grew ~2.5× while falsifiable claims moved only 1→3.

Files

34

21 → 34

Words (k)

~47

~19 → ~47

Dissents preserved

10

6 → 10 · flat since S2

Catches

12

6 → 12 · 13th (C-013) logged post-S4

Falsifiable claims

3

1 → 3 · hand-judged

The learning loop — caught, distilled, promoted. Source: CATCHES, LEARNINGS, METRICS §2. The pipeline, not the raw count, is the real metric — and a caught error is not a prevented one.
36catches logged3 of the first 6 were the chair catching its own bias; C-012 the loop catching itself; C-014–C-017 the empirical turn; C-018–C-022 the cross-check + reflection; C-023–C-036 the delegated sessions, courier rounds, the state-of-project forum, and the author's ratifying contact — including C-019: that this very number is a births-only vanity metric (no denominator; the drain logs subtractions separately — it fired once, DC-007, by the author's hand, retiring Theory C's "the machine subtracts of its own motion" sub-claim, so across four readings the self-initiated channel still stands at zero)
17learnings distilleda learning that never changes a rule is just an observation
25ground rules (max)R22–R24 and Rule 17b, provisional across three delegated sessions, are now author-ratified (S4k); the newest, Rule 25 — a reversible moratorium on net-additive machinery — was ratified in the same act

Denominators, not achievements: 4 sessions · 10 core + 22 advisory panelists · 5 fault-line axes · 25 ground rules · ~7 searches / ~46 results. Ratification is internal — the author’s own, and reconstructed-persona ratings; external validation is exactly zero. Any number can be checked against METRICS §4, the open questions, and the Round-7 votes in the theories.

The record in full: Catches — the error log · Learnings · Reflections.

The Figures — arguments drawn

Above: counts measured. Below: arguments drawn — a different kind of mark. Nine conceptual figures (eight in Mermaid, one a hand-authored SVG), each restating part of the argument in a second notation, and each a claim that can be wrong. Download any as SVG (vector) or PNG.

Fig. 01

The logic of the project

Legitimacy flows in a loop, not a line — the panel argues, the chair synthesizes but never votes, the author ratifies, the public extends. The single point of risk is the chair, which is why the learning loop exists. Contested: D-004 asks whether running this well actually means anything.

rendering figure 01…
Fig. 02

The document architecture as a flow

Not a stack of documents but a circuit: outputs are tested back against goals and history, and what is learned rewrites the rules. The arrows are the point.

rendering figure 02…
Fig. 03

The learning loop

A catch becomes a learning becomes a rule change. In Session 2 the loop closed once: the L-001 guard caught C-007 before it reached an output — a prior learning preventing a repeat of a prior error.

rendering figure 03…
Fig. 04

The pressure-tests, layered

Thirteen tests in four coupled layers, with named, directed couplings — where "polycrisis" leaves the arrows undrawn, we draw them. Heavily contested: Marx rejects the driver/symptom cut; Nietzsche puts meaning at the summit; the arrows are the research program (C-007, Q-002).

rendering figure 04…
Fig. 05

The field, positioned

Each contemporary framework placed on the two axes the project turns on. The top-right — total and testable — is nearly empty, and that emptiness is the niche. The amber marker shows where this project aims. (The meaning dimension can't be shown here; see LANDSCAPE §9.)

total · descriptive total · testable (rare) partial · interpretive sharp · narrow interpretive · descriptive predictive · mechanistic partial slice → total ambition Cliodynamics Complexity science Luhmann Polycrisis Metacrisis Big History Metamodernism Integral Theory This project (aim)
Fig. 06

The three candidate theories

Not three guesses at one answer but three kinds of theory — diagnostic, mechanistic, reflexive — that fit together. B is the mechanism beneath A's gap; C is the response to both; two named dangers stalk C.

rendering figure 06…
Fig. 07

The path to Code — a decision tree

The move is gated on what is bottlenecking, not enthusiasm. Chat is correct while the work is thinking; Code becomes correct only when there is a stable thing to build or a real quantity to compute. The honest branch includes "may never need heavy Code." A hard context-capacity limit in Chat can force Gate 1 earlier — see figure 08.

rendering figure 07…
Fig. 08

The Chat / Code shuttle

The working architecture: Chat deliberates, Code executes, the repo remembers. Neither side holds the whole project in one context — Chat loads only a slice, Code reads the full repo from disk. This is the fix for the context-window limit that forced our compaction, and it is the author's existing two-Claude writing workflow applied here.

rendering figure 08…
Fig. 09

Theory A as a signed causal loop

The gap, drawn — the first bridge from prose to a runnable model (THEORY_A_OPERATIONALIZED.md). Two loops decide the outcome: R1, a reinforcing trap (populism → weaker institutions → slower adaptation → wider gap), and B1, a delayed balancing loop (overwhelm → reform → faster adaptation → narrower gap). B1 is the accelerationist counter (D-006) made visible; which loop wins is empirical, not assumed.

rendering figure 09…

Source of truth: docs/DIAGRAMS.md (also renders on GitHub). Open the standalone figures or the landscape & panel map. Contested claims point to the open questions.