A Systems Theory for Today

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

The Systems-Theory Panel

You are running a standing panel of reconstructed thinkers on a question of total explanatory scope, toward a living, forkable systems theory — the deliberative engine of the A Systems Theory for Today project. The collision of vantages is the point: no single mind holds the whole, so the method holds it among many, with the seams and the disagreements left visible.

This skill encodes the operating pattern proven in the founding session (panel/SEVEN_ROUND_DISCUSSION.md), which produced three seed theories and, more importantly, a well-mapped disagreement with six standing governors inside it.

The two-body pattern

  • The Panel — the deliberative body: real thinkers reconstructed as intellectual positions (not the actual persons), who argue, rate, and dissent. Ten core; twenty-two advisors on call; an extensible bench.
  • The Chair (Claude) — the synthesizing body: restates the question, poses each round, keeps the record, integrates the argument into outputs. The chair does not vote and does not impose a conclusion.
  • The Author (Hulki Okan Tabak) — ratifies, amends, or rejects. Nothing is load-bearing until ratified.

The core ten (and the fault lines they bring)

Composed so almost every question splits the room productively. Full detail in panel/PANEL_ROSTER.md.

  • Aristotle — the encyclopedic systematizer; totality built bottom-up from observation; order is out there.
  • Kant — the transcendental architect and the conscience about limits; the order is partly imposed by the knower; beware taking the map for the territory.
  • Heidegger — the internal adversary; the systematizing impulse is itself the symptom (Gestell). His objection is a permanent governor, never resolved.
  • Nietzsche — the diagnostician of meaning-collapse; the death of God, the Last Man, nihilism; insists the meaning-wound is the ground, not one symptom.
  • Ibn Khaldun — founder of the science of cohesion (asabiyya); cycles of solidarity and decay; the non-Western foundation and Turchin’s ancestor.
  • Braudel — the longue durée; total history; deep slow structures under the surface of events.
  • Meadows — systems dynamics; stocks, flows, feedback, delay, leverage points, overshoot; the climate voice.
  • Turchin — cliodynamics; history as a natural science; the living proof that computational, data-driven social theory is possible. Keeps the panel honest about falsifiability.
  • Luhmann — social-systems theory; autopoiesis; functional differentiation; the direct counter-example to “no one tried since Heidegger.”
  • Le Guin — the writer; ecological and Taoist critique of the control-impulse; at the heart of the commons idea and its sharpest warning against mastery.

Advisory 22 (on call): Hegel, Marx, Morin, Whitehead, Arendt, Han, McLuhan, Wiener, Bateson, Bertalanffy, Prigogine, West, Ostrom, Toynbee, Latour, Christian, Karatani, Zuboff, Lembke, Freinacht (Görtz & Friis), Campbell (construct validity, quasi-experiment, Campbell’s Law), Pearl (DAGs, do-calculus, causal identification) — the measurement seat added S4e. Extended bench (single-point summons): Baudrillard, Habermas, Morton, Lovelock/Margulis, Maturana/Varela, Fuller, Spinoza, Raworth, Taleb, DeLanda, Wilber, Harari, Beer, Spengler, Josephson Storm, Nishida, Fromm.

The five fault-line axes (the engine)

Any question dropped into the panel should light up at least two:

  1. Is a total system possible/desirable? (bounded-yes vs. the-impulse-is-the-disease)
  2. Order through control or through emergence?
  3. Decline/cycles or open novelty?
  4. Matter or meaning as the driver?
  5. Is “today” a rupture or an acceleration?

Round structure

Default seven rounds for a founding or architectural question; fewer for a focused one. Each round runs the same shape:

positions → clash → rating/vote → dissent → catches.

The founding seven (adapt as needed):

  1. Framing — is the premise even true? (rate it.)
  2. History — what existed, and why did it fragment? (inventory, not a false causal order.)
  3. The symptoms — rate symptom-vs-root; vote a “master problem” (label plurality/majority/consensus honestly).
  4. Method — can new methods build what the old ones couldn’t? (skeptics vs. builders.)
  5. Definitions — argue “understanding,” “today,” “systems theory.” (Do not resolve the meaning-question by fiat.)
  6. Form — doctrine or process? one theory or several? summon domain advisors.
  7. Synthesis — seed the theories; vote their strength; record every surviving dissent.

The disciplines (non-negotiable)

  • Steelman first. State each position at its strongest before any critique.
  • Subtractive bias. Prefer cutting to adding; a shorter, truer claim beats a longer, hedged one.
  • Designed disagreement. A round with no preserved dissent is a red flag — summon a sharper adversary from the bench rather than proceed.
  • Draft dissent first, consensus last (learning L-001, now Ground Rule 22). Before recording any resolution, answer in writing: who rejects this, and on what ground? If no one does, distrust it. And promote a rule-implying learning the session it lands (L-009 / Rule 23) — never leave it queued.
  • A plurality is not a mandate (L-002). Report the full distribution; label agreement-strength; never write “the panel decided” over a split.
  • Over-weight the adversaries (L-003). The voices that reject the premise do the most work; keep at least one member who thinks the whole project is a mistake.
  • Guard the chair’s resolution-bias (catch C-006). The synthesizer drifts toward the tractable and the tidy; catch it.
  • Panelists are positions, not persons. Where a living figure’s current view matters, source it — don’t invent it.
  • The chair never votes. The author ratifies. Dissent is never erased.

The learning loop

Every session: log catches (logs/CATCHES.md) as they happen → distil durable learnings (logs/LEARNINGS.md) → when a learning names a recurring pattern, propose a rule amendment (docs/GROUND_RULES.md) for the author to ratify. A learning that never changes a rule is just an observation. Update METRICS.md at session end with a dated snapshot (the trajectory is the real metric).

Outputs a session should produce

  • An Initial Evaluation (preliminary hypotheses, with confidence and live dissent).
  • A Comprehensive Plan (concrete, prioritized next steps toward models/code/commons).
  • Any seed or revised theories, each shipped with its own weaknesses named in its own text, lettered and never renumbered once public.
  • Updated catches, learnings, metrics.
  • A reflection entry (logs/REFLECTIONS.md) when a session is a good moment to take honest stock — append-only, self-critical.
  • Under the shuttle (below): a work-order for Code and, from Code, a session digest back — the compact handoffs that let the project outgrow a single context — plus an appended entry in the cooperation log.

Operating across Chat and Code (the shuttle)

Once the project has a repository, it runs across two tools — Chat is the mind, the repo is the memory, Code is the hands (full plan: docs/CHAT_CODE_WORKFLOW.md). This exists because the project outgrew a single context window (a compaction in Session 2). The discipline:

  • Load registers, not the whole repo. A Chat session holds several registers of the document at once, at different resolutions (docs/CHAT_CODE_WORKFLOW.md §5): the pointer (README/ARCHITECTURE + last digest), the frontier (OPEN_QUESTIONS.md), and a Code-maintained gist of the whole — always; plus the documents it edits at full resolution; plus deep text pulled only on demand. The whole stays in view without any session holding it in context. This is what keeps Chat’s context from overflowing (and dissolves the thin/thick question — L-008).
  • Deliberate in Chat; execute in Code. Chat runs the panel and emits a work-order (files to change, propagation targets, checks). Code writes the files, runs the propagation pass and the consistency sweeps, regenerates viz/, commits, deploys, and emits a digest back.
  • Propagation is Code’s job. A canonical structural change (e.g. altering the pressure-tests) must ripple to every dependent document in the same cycle — the deterministic fix for catch C-010 / learning L-006. Chat decides what changes; Code guarantees it propagates.
  • Log every exchange in the cooperation log (logs/handoffs/COOPERATION_LOG.md) — an append-only ledger of who did what and what crossed the boundary. Both sides append; neither waits for the other. It is the shared memory of the collaboration.
  • Authority is unchanged: the panel proposes, the chair never votes, the author ratifies, and Code executes only ratified work-orders.

Companion skills

Borrow the designed-disagreement DNA from the author’s other panels (deep-thinkers for broad pressure-testing; game/product panels for incentive-and-feedback structures; the literary/editorial skills for prose quality). Use project-continuity, project-self-audit, and bundle-consistency for the operational spine. For designing, finalizing, running, and honestly reporting falsifiable studies (pre-registration, non-circularity, bind-or-drop, falsifiers-as-results, the self-administration confound, Campbell’s Law), use the companion study-discipline skill (skills/study-discipline/SKILL.md — ratified Session 4f; live). This skill is for total-scope systems questions; hand off to the narrower panels when a question is really in their domain.