COMPREHENSIVE PLAN

Version: 0.6 · Status: Draft (author to ratify) · Last updated: Session 4j (the S4j forum’s leverage-ordered roadmap appended as the operative plan — see the final section; forum-delegated (S4j), provisional-pending-author)

The second deliverable from the founding session. Where INITIAL_EVALUATION.md says what the panel currently thinks, this says what we build next and in what order — the path from three prose theories to explicit models, runnable simulations, a public forkable commons, and a website. Edited and sequenced by the chair; the author ratifies scope and priority. Phases mirror the path-to-code in ARCHITECTURE.md.

Guiding principles for the plan

1. Falsifiability is the through-line. The project’s biggest current weakness is that most of the thirteen pressure-tests carry no falsifiable claim (roughly three do: D3, Y3, S3), and the strongest theory (A) has the weakest testability. Every phase is judged first by whether it moves claims from evocative to checkable.
2. Prose → model → simulation → commons. Each theory descends a ladder: readable argument, then an explicit variables-and-couplings map, then a runnable model, then a public artifact others can fork. We do not skip rungs; a simulation built on an un-mapped theory is false precision.
3. The method is a deliverable too. Because Theory C claims the form is part of the answer, building and testing the commons-governance is not overhead — it is content, and it gets planned like content.
4. Preserve the seams. Integration never means erasing dissent. The six standing governors travel with every phase.
5. Small, honest, shippable increments. Better a modest thing that works and is public than a grand thing perpetually almost-ready.

The phase map

Phase 0 — Foundation (Session 1 — COMPLETE)

Goal: a coherent, self-governing repository that a stranger could read, understand, fork, and contribute to. Delivered: README, ARCHITECTURE, the full reference set (Goals, Ground Rules, Method, Metrics, Skills, Visualization, Glossary, History), the panel roster, the seven-round deliberation, the two output documents, three seed theories, the catches/learnings loop, the skill file, contribution/licence terms, and a first interactive visualization. Definition of done: ✅ all of the above exist, cross-reference cleanly, and carry version/status headers.

Phase 1 — From prose to explicit models (next 1–3 sessions)

Goal: turn the three theories from arguments into explicit, inspectable structures — and raise falsifiability. Deliverables:

• The Seven-Problems Concept Map — drafted in Session 2 (DIAGRAMS.md §4) as a thirteen-node, four-layer directed graph. A directed, signed graph: nodes = the key variables across the layered tests (acceleration, optimization-intensity, wealth-pump/inequality, machine-intelligence, institutional-adaptation-rate, trust, fertility, overshoot, polarization, anomie, …); edges = couplings with a sign (+/−) and, where known, a delay. It makes the “everything is connected” intuition explicit and criticizable — the thing the polycrisis literature never did. Next refinement: add the signs and delays, and split into per-theory causal-loop diagrams.
• Theory A, operationalized. ✅ Drafted (S3): outputs/THEORY_A_OPERATIONALIZED.md. Candidate proxies for the two master rates: rate-of-technological-change (adoption curves, capability benchmarks, product-cycle times) and rate-of-institutional-adaptation (time-to-regulate, legislative cycle time, curriculum-update lag), a normalization discipline, the divergence prediction (cross-sectional + longitudinal), crisp falsification conditions, the discriminator against Turchin’s wealth pump, and a pre-registered first study. This was the top intellectual priority of the phase; the next step is to pre-register and run the study (the concrete heavy-Code trigger).
• Theory B, bounded against conspiracy. ✅ Drafted (S4): outputs/THEORY_B_OPERATIONALIZED.md. Optimization intensity (O) as the measurable driver of capture/pathology (P); the explicit test that distinguishes selected-to-capture from designed-to-capture (B’s integrity condition); the natural experiment (engagement-optimized vs. deliberately de-optimized systems); and the B-vs-A decomposition. The cheapest of the three to run — the recommended first empirical study.
• Theory C, instrumented. Define the reflexive metric: the rate at which the panel/commons process catches errors and retains live objections, versus a single-author baseline. Begin logging it (Session 1’s six catches are the first entry).
• A falsifiability pass across all thirteen pressure-tests: for each, either state a checkable claim or explicitly mark it “conceptual — not currently falsifiable” and say why. (Extends the Session-2 coverage tracker in METRICS.md §4.) Definition of done: the concept map exists and is criticizable; each theory has at least one precisely-stated, potentially-falsifiable claim; pressure-test falsifiable-claim coverage rises from ~3/13 toward the majority — beginning with D1 (the gap, Q-001).

Phase 2 — Public living commons (parallel with Phase 1, then ongoing)

Goal: put the repository where others can actually fork and contribute, and stand up the governance that keeps a commons coherent. Deliverables:

• Public GitHub repository with the contribution model live (CONTRIBUTING.md), issue/PR templates for the three contribution types (strengthen / break / propose-rival), and the “recruit for the fault line” admission criterion for new panelists.
• Commons governance v1, Ostrom-informed: clear boundaries (what’s in scope), graduated contribution, a transparent process for how rival theories are admitted and how dissent is preserved, and a rule that no contribution is merged by silently deleting a live objection.
• A first companion website (static, honest, fast) presenting the question, the three theories, the concept map, and the open disagreements — with the visualizations from viz/ embedded. (Frontend/design skills and the design tokens apply here.)
• The bilingual step: an English-primary / Turkish-sibling presentation, treated as equal-weight editions.
• Operating model: the Chat/Code shuttle (docs/CHAT_CODE_WORKFLOW.md). Phase 2 is where Light Code begins — the repo, the build, the deploy — and where deliberation and execution split across tools: Chat deliberates; Code executes, propagates, sweeps, and deploys; the repo is the shared memory. The context-capacity limit hit in Session 2 makes standing this up the immediate next step, not a distant one. Definition of done: a stranger can read the site, open the repo, file a rival theory or a red-team objection through a defined process, and see their dissent preserved rather than dissolved.

Phase 3 — From models to simulation (later; gated on Phase 1)

Goal: make selected couplings runnable, so the theories can be pushed to see where they break — with loud humility about the limits of prediction. Deliverables:

• A minimal system-dynamics model of one well-scoped loop (candidate: the attention-capture / polarization / trust loop, or a Meadows-style overshoot loop for climate), as stocks/flows/feedback that can be simulated.
• Where apt, an agent-based or network model for a symptom that is fundamentally about many interacting units (fertility norms diffusing; contagion of unreality; Turchin-style elite dynamics).
• A hard-coded discipline of epistemic humility: every simulation ships with its assumptions exposed, its computational-irreducibility caveats stated, and Kant’s warning attached — this is a system of appearances structured by our own categories, not the territory. No model is presented as prediction; models are intuition pumps and consistency checks, explicitly. Definition of done: at least one loop runs, its assumptions are fully exposed, and it has already been used to catch a flaw or tension in the prose theory it came from (that catch is logged).

Phase 4 — Integration and iteration (continuous)

Goal: keep the whole alive — re-run the panel on new questions, integrate external contributions, refresh the empirical grounding, and let the theories evolve, split, merge, or die. Deliverables (recurring):

• Periodic panel sessions on new questions and on external rival theories, each producing catches → learnings → rule amendments.
• A standing refresh cadence for time-sensitive claims (the contemporary-landscape scan, and any live data series) — because “what’s out there now” goes stale fast.
• Regular self-audits (are learnings reaching the rules? has a new skill emerged? is the plurality real or cosmetic?) and metrics snapshots that show the trajectory, not just the latest state. Definition of done: never — this phase is the living state. Its health is measured by the trend lines in METRICS.md, not by completion.

Immediate next-session task list (prioritized, as of Session 4)

1. Ratifications — done. A’s operationalization and the living-document concept ratified (S4); Theories B and C operationalized to the same falsifiable standard and ratified (S4). The layering stays a hypothesis; the coherence vacuum M stays open.
2. Pivot to Code (the active step). The first work-order is written (logs/handoffs/WORK_ORDER_S4.md): initialize the private repo, stand up the public reading site, adopt the shuttle templates, make the propagation/sweep scripts permanent, and build the R3 gist. This is Code’s job now — Chat has taken it as far as a single context safely can.
3. All three theories operationalized (✅ S3–S4). A — THEORY_A_OPERATIONALIZED.md (the gap; the flagship). B — THEORY_B_OPERATIONALIZED.md (optimization intensity; the cheapest to test → the recommended first empirical study). C — THEORY_C_OPERATIONALIZED.md (the reflexive ablation; runnable now, no data pipeline, bounded by the baseline problem Q-012). Next: pre-register and run — starting with B — which crosses into heavy Code (Gate 2) and needs a fresh ratified work-order.
4. Add signs and delays to the concept map (DIAGRAMS.md §4); draw signed causal loops for B and C (A’s is done, §9).
5. Promote L-001 into the Ground Rules (draft-dissent-first; label agreement-strength) — evidenced by C-007.
6. Guard the standing risks: the chair’s resolution-bias (C-006, now measurable via C’s catch-provenance metric); the propagation discipline (C-010 / L-006), which the shuttle makes Code’s deterministic job.

Risk register (carried into every phase)

What “success” would look like (from GOALS.md, made concrete)

• S1 A stranger reads the repo and understands the question, the landscape, and the three theories. (Phase 0–2.)
• S2 Falsifiable-claim coverage rises from ~3/13 toward the majority. (Phase 1.)
• S3 At least one coupling runs as a model and has caught a flaw in its own prose theory. (Phase 3.)
• S4 Real external rivals and red-team objections arrive and are integrated with dissent preserved. (Phase 2–4.)
• S5 The commons demonstrably out-catches a lone author — the reflexive proof of Theory C. (Measured continuously from Session 1.)

The plan is deliberately more modest than the ambition. The ambition is a systems theory for today; the plan is a sequence of small, honest, checkable steps toward it, each of which is itself worth having even if the whole is never finished — which, if Theory C is right, it never will be, because “finished” is the wrong word for a commons.

Updated Plan — the next phase (Session 4f, from a 5-round planning panel)

Synthesized by the chair (not a vote) from a 5-round designed-disagreement planning panel (Turchin, Ostrom, Meadows, Nietzsche, Heidegger, Le Guin, Campbell), informed by the S4f reflection (logs/REFLECTIONS.md). Code-synthesized and PENDING author ratification — especially the eight forks below. It supersedes nothing above; it sequences the next phase.

The phase in one sentence: spend this phase proving the machine can drain and can be emptied — deliberately choosing rigor over reach, paying in another parochial, private cycle with the meaning-question (D-003) left open. “An empty result is the first real one this project has produced.”

The spine (to ratify)

costless subtractions now → lock the CONSTRUCT → run the PLACEBO → install the DRAIN → seat the foreign GRADER → widen Study-B → Q-001 / open the repo — with Heidegger & Nietzsche’s inversion preserved (open the repo FIRST — the only foreignness you cannot seat; holding is the ossification the reflection flagged).

Prioritized moves (each carries its gate)

1. Costless subtractions now — un-count the “4/4” to its true n=1 (keep only the Facebook/YouTube split); strike the ledger-sum and “32 voices” from any claim of health or independence; downgrade “plurality” to one standpoint until a foreign grader earns the word back. Gate: author ratifies the strike list. (Partly done in S4f: the cross-check headline is re-framed to the split and a METRICS honesty-note is in place; the governor-strike + Theory-C retirement below are not done.)
2. Lock the held-out CONSTRUCT for “threatens the metric” (with a numeric kill-threshold) before any new number or Study-B widening — a classifier reads the outcome and names it the cause, so the measure must be derived from a held-out construct. Ship it bundled with a quiet/non-Western strain case + the opposite-pointing near-control. Gate: author settles construct-vs-classifier (Q-014).
3. Run the PLACEBO — a substantively empty session through identical machinery, blind-scored against a real session (one that risked a value) on catch-count + convergence. The cheapest falsifier, and the only test that can end the project. Gate: author cuts Fork 1 + names the control session (Q-016-adjacent).
4. Install the SUBTRACTION operator (the drain) — a decisions-changed log + a pre-written single result that retires Theory C; strike one never-bit governor as its first bite (which must shield the machine, never prune a dissenter or D-003). Gate: author ratifies the trigger (Fork 2) + names the first governor struck (Q-017).
5. Seat the foreign GRADER — able to cost the author a claim, or it is merely a fourth LLM. Gate: author ratifies its nature + placement (Fork 3).
6. Widen Study-B — the opposite-pointing near-control first (selection vs revenue diverge), then WeChat + a quiet care-system case (Q-015). Gate: construct lock ratified; author sets the widen/grade order (Fork 4).
7. Q-001 + the REPO — held on a clock, as a bounded nested folder now, opening by default if the window is missed. Gate: all upstream + Fork 5 (open-first vs held).

What to STOP / RETIRE (the metabolism’s first outflow — the reflection’s zero-retraction fix, C-022)

The “4/4” as evidence (→ n=1; keep the split) · the ledger-as-scoreboard (keep the log as memory; strike the sum from any health claim; track decisions-changed instead) · “plurality / 32 voices” as an independence claim (→ n=1 until a foreign grader) · one never-bit governor, struck on the record (first strike shields the machine) · Theory C — candidate for retirement (gated; Nietzsche: the only subtraction with a value inside it; Ostrom refuses before the grader tests it) · and pre-empt the retraction-count becoming the next vanity number.

The hard tradeoffs (where it lands, and the cost)

Rigor vs reach/meaning → pay it; name the wound (one author/one model, construct validity ~0), don’t design around it. Placebo blinding vs foreignness → blind it, but the control must have risked a value. Adding machine vs Heidegger → the drain must be a decided graduated sanction, not a self-triggering counter; his refusal recorded unanswered (he is D-003). Opening the repo vs stability → held on the clock as a bounded nested fork; “open first” stands as a live inversion. Relieving vs wounding cuts → only retiring Theory C costs a value; pre-write its death-condition now.

Decisions for the author (the eight forks — the plan does NOT resolve these)

1. Pre-registration: held-out CONSTRUCT (majority) vs the CLASSIFIER (Turchin). 2. Fork 1 — drain placement: placebo gates the drain (Campbell) vs drain-first as the placebo’s reading instrument (Meadows). 3. Fork 2 — drain trigger: graduated ladder (Ostrom) / strike-bound-to-a-null (Turchin) / decided-by-hand (Heidegger, Le Guin, Nietzsche, Meadows) — near-consensus against a bare idle-clock. 4. First governor struck: any never-bit one, but it must shield the machine, not prune D-003 / a dissenter. 5. Fork 3 — grader: foreign-veto-bearing vs held-out-internal; upstream (appropriator-first, Ostrom) vs downstream. 6. Fork 4 — Theory C + order: retire C now / on-a-null / after-grader; widen-then-grade (Le Guin) vs grade-then-widen. 7. Fork 5 — Q-001 + repo: now vs later; open-first (Heidegger/Nietzsche) vs held-on-clock. 8. Placebo control: any real session vs only one that risked a value (Nietzsche).

Success vs falsification

Success: the phase produces ≥3 discriminating numbers (not confirmations) — the placebo reads a real empty-vs-live gap; the near-control separates selection from revenue; the foreign grader names ≥1 outcome the plurality changed — AND the synthesis visibly moves when a truly other standpoint sits down. A clean null (placebo passes) is also success in the weak sense — the first real result — and ends the project honestly. Falsified (any one of five, none yet run): the placebo passes → the ledger is theatre; a lone adversary catches everything the panel caught; the foreign grader names nothing the plurality changed; the synthesis fails to move for a genuinely foreign standpoint (C-006 confirmed); or the coherence-vacuum reduces without remainder to Turchin’s wealth pump (D-004 answered against us). Register nothing you cannot later fail.

Preserved dissent (the planning panel, unresolved)

Heidegger (construct-validity is the disease in a lab coat; every monitor / decided-drain is enframing; he is D-003). Nietzsche (learn the construct by wounding; the placebo’s control must risk a value; retire C by hand now; open the repo first). Ostrom (seat a human/institutional appropriator first; don’t retire C before the grader tests it). Meadows (install the drain first, as the placebo’s instrument). Campbell (protect the split; construct not classifier; gate the drain behind the placebo; name the wound). Turchin (placebo in parallel + blinded; strike bound to a null not a clock; near-control first; keep Q-001). Le Guin (widen before grading; a grader over four loud American cases isn’t foreign). Standing governors D-001 / D-004 / D-005 / D-006 remain open; C-006 applies to this synthesis itself — the chair did not vote and did not collapse the five forks.

Ratified & reordered (Session 4g — author)

The eight forks above are now ratified. They are kept in place (nothing erased); their resolutions are recorded here. The author’s Fork-2 choice reorders the spine: the drain is built first, as the placebo’s reading instrument.

Fork resolutions. 1 (pre-registration) → held-out CONSTRUCT + validity first (Heidegger’s refusal preserved as dissent). 2 (drain placement) → drain FIRST (Meadows). 3 (drain trigger) → graduated sanction ladder (Ostrom); not a bare idle-clock. 4 (first strike) → a never-bit over-claim guardrail (must shield the machine; never a dissenter or D-003). 5 (grader) → held-out internal, no veto; placement upstream / appropriator-first (Ostrom) — a no-veto internal grader does not meet the “genuinely foreign” bar, so Move 1’s “plurality” downgrade stands until an opened repo / a true outside vantage. 6 (Theory C + order) → retire C on a null (Turchin — death-condition pre-written now, bound to a pre-registered null), and widen before grading (Le Guin). 7 (Q-001 + repo) → held on a clock (repo private; opens by default if the window is missed). 8 (placebo control) → any real substantive session.

The ratified sequence (reordered by Fork 2): Move 1 costless subtractions ✓ → lock the held-out CONSTRUCT → install the DRAIN (as the placebo's instrument) → run the PLACEBO → widen Study-B → seat the held-out internal GRADER → Q-001 / open the repo (held on a clock)

Still preserved and unresolved: Heidegger’s dissent (the apparatus is itself enframing; D-003); the standing governors D-001 / D-004 / D-005 / D-006; and C-006 over this record. Ratifying how to run the phase did not resolve whether the fragility is methodological or ontological — that split stays open, by design.

Executed & recalibrated (Session 4g autonomous loops — appended, nothing above erased)

Execution stamps on the ratified sequence. Move 1 ✓ (DC-001…003) · Move 2 ✓ (TTM construct locked, studies/study-B-optimization/CONSTRUCT_THREAT_TO_METRIC.md) · Move 3 ✓ installed-not-proven (the drain, logs/DECISIONS_CHANGED.md; one live entry DC-004; first strike HELD bound to a null on a clock; no value-costing subtraction yet) · Move 4 ✓ ARMED (pre-reg v0.5; arms generated blind-clean; kits + runbook in the author’s Downloads; hashes studies/FROZEN_HASHES_S4G.md; nothing scored in-house) · Move 4½ — seal the scoring (inserted by the S4g panel): deterministic scorer scripts written, fixture-tested on both decisive branches (DISCRIMINATES/PASSES; powered/under-powered), sealed before any verdict exists (studies/placebo-control/src/, studies/study-C-ablation/src/) · Move 5 in progress: near-control pool now holds documented divergences in both directions — YT2012 (selection beat revenue) + YT2017 Adpocalypse (revenue beat selection — the falsification-capable case) as frozen dossier pairs, the Shorts gap as an evidence map, the CAC-2022 provisions as the first fetched non-Western primary, the care-systems debt logged (.../near-control/); nothing TTM-coded (widen-before-grade) · Move 6 redefined = design, then seat: the held-out internal grader is drafted (panel/GRADER_DESIGN_DRAFT.md — upstream/appropriator-first, no veto; cannot lift the plurality downgrade) · Move 7 weight raised: with the non-LLM appropriator deferred to a future forker, the opened repo is the only remaining road to a genuinely-foreign vantage — checklist + a named default-open date (2026-10-01) drafted (docs/REPO_OPEN_CHECKLIST.md), opening itself author-only, with the hard sequencing rule: verdicts collected (or seals escrowed) before the door opens.

The firing-vantage recalibration (author, supersedes the Fork-5/6 gate wording above): cross-model (project-blind other LLMs, author-mediated) fires the retire/end levers; the non-LLM genuine appropriator is dropped as a current gate and transferred to the future-forker slot. Every firing logs cross-model-confirmed, not genuinely-foreign, n=1 of a kind (L-013), revisitable by a future fork (Ostrom/Campbell/Turchin dissent travels).

The verdict gate is a plan element: verdicts return only by the author’s hand (the author-mediated paste is part of the design); Code’s whole role at scoring is running the sealed deterministic scorer.

Provenance (Q1 guard): after all seals, the author recorded an outcome-preference (“let’s have the placebo discriminate”) — ruled a wish-to-record, not an instruction (studies/placebo-control/PRE_REGISTRATION.md §8.6); the pre-registered conditional stands: any in-house “discriminates = vindicated” text takes a Rung-0 flag.

Success-criteria note: the phase’s “≥3 discriminating numbers” now map to the three pending instruments (placebo · N=20 ablation · near-control TTM) — with the caveat that all three return through the same external-model set, so the evidence base shares one aquifer batch (L-013 rides on all three).

S4h — the standing delegation spent (appended; nothing above erased)

The author’s standing delegation (“fully delegated until you finalize all that has to be done”) was booked as C-027 and spent as the C-023 correction demands — on ratifying the queue and restoring external checks. Full record: panel/SESSION_S4H_RATIFICATION.md; every stamp panel-delegated (S4h standing delegation), provisional-pending-author.

Executed: the whole ratification queue (results readings; the Q-001 admission as a split; all six learning-panel adoptions enacted-as-amended — episode unit docs/EPISODE_UNIT.md (criteria DRAFT, freeze = the author’s act), lag-types + adapted evidence ladder (docs/PRESSURE_TESTS_CAUSAL_HYPOTHESIS.md v1.1), D-007 installed (first bite C-028), Rule 17b, episode-base candidates named-not-coded; rule wordings in force with the author’s ratification recorded as OUTSTANDING; the drain’s rulings — the first strike still held on a second empty audit, the §5 clock’s pre-registered NULL posted (logs/DECISIONS_CHANGED.md §5.1, three unnetted entries), margins confirmed with the anchoring gate) · the two instrument repairs designed under sealed requirements (content-sensitive convergence v2, two-axis; anchored severity + gold calibration set) with adversarial break-and-revise, RECODE validation pre-registered (instrument-acceptance only; non-firing; dead-man clock 2026-10-01) · the widen-grade TTM kit built (five ex-ante dossiers incl. the anchors and the first non-Western case, Douyin-2021/CAC — the pool’s register skew booked as dated debt) · the grader v0.2 seat-ready (seating gated on widen verdicts; disposition-log reconciliation; absent-standpoint field). The author’s courier round is the next external act: run the validation kit and the TTM kit through project-blind external models per the S4h runbook; then confirm/amend/reverse the S4h stamps, cut the §3 fork in docs/EPISODE_UNIT.md, and rule on the repo clock’s advance-or-hold (queued, not moved).

S4i — the second standing delegation spent (appended; nothing above erased)

Booked as C-032. The two S4h courier rounds returned (severity anchoring PASSED 94%; the convergence instrument caught the placebo’s own empty arm as non-empty → DC-006; TTM codes blind 30/30 but cannot separate selection from revenue; the blue ablation returned FINAL Indeterminate — the detection layer failed at 1%, a NEW unreliability, candidate L-016 — and the one powered ratio landed null-side at 1.15×). An 8-lens + 4-adversary panel processed the delegable docket (three adversary flips); the grader ran in SHADOW (85 objections, all chair-answered, seat held); the detection repair was designed and its blue-v2 + criteria-red-team kits sealed to Downloads; the author’s five-user site playtest ran to completion and STOPPED (36 content findings queued). The zero-strike clock posted its second null. Everything panel-delegated (S4i), provisional-pending-author. Full records: panel/SESSION_S4I_RATIFICATION.md, panel/GRADER_SHADOW_LOG_S4I.md, playtests/.

S4j — the state-of-project forum’s roadmap = the operative plan from here (appended; nothing above erased)

Booked as C-035 (the third standing delegation; the SG-10 guard’s first trigger, fired and overridden on the record). An exhaustive forum (8 lens briefs → 3 drafting tables → adversary+revise) produced three documents — a systems-theory report, a wrongs analysis, and a leverage-ordered improvement roadmap (six PDFs in Downloads; record panel/SESSION_S4J_FORUM.md). This section installs the roadmap’s spine as the operative plan. Authored by the improvements table (Meadows · Ostrom · Le Guin · a delivery advisor) and a wrongs table (Campbell · the sentinel · a hostile skeptic); one model in many roles (L-015); forum-delegated (S4j), provisional-pending-author. No summary may cite this as remediation performed — a provisional critique of the tower is a fourth storey of the tower until the author confirms (the forum’s D-2).

The keystone diagnosis (why the spine changed). The project built machinery across the whole Meadows leverage 6–12 band (parameters, the drain at 8, the instruments/grader/site at 6) and moved zero object-level rungs, because nearly every structure governs the apparatus rather than the world — “the machinery deliberating about its right to deliberate.” The plan’s new organizing discipline is therefore leverage-ordered, weak→strong, and its first instruction is to stop building the machine and use it against the world.

The operative spine (the leverage-sorted short list; the author’s calls marked ⟐):

1. ⟐ Name the stage honestly — “a research program with strong hygiene, three untested seed hypotheses, one borrowed falsifiable chain, and no data yet.” The one leverage-3 act that is cheap-now; recalibrates every reader and header. (Code surfaced the honest-stage facts on the site’s self-audit + /standing/ pages S4j; the canonical header rename from “systems theory” is the author’s.)
2. ⟐ Declare THE MORATORIUM — freeze net-additive meta machinery (new rules, seats, instruments, graders, programs, standing delegations that GROW the apparatus) until an object-level rung moves. Subtractive/bounding repairs (re-arm SG-10; an empirical-drought downgrade clock; the 7th repo-open gate) and object-level work (a 14th pressure-test; an actual empirical run) are EXEMPT and land in the same ratifying act. This is the plan’s new top priority and the thing every other item depends on.
3. ⟐ Fire the drain once of the project’s own motion — or make a channel fireable (re-seal C-031 in an equivalence form + a bite-event definition), or concede on the record it cannot. The act that distinguishes an apparatus from a compulsion; the drain’s own §9 reserves it to the author.
4. Surface the tested-and-FAILED instruments on the self-audit page (severity PASSED 94%, detection FAILED ~1%, convergence INVALID, ablation NO VALID RESULT) — the strongest asset, out of the basement, stated precisely and flatly. (Code did this S4j, provisionally; the deeper authored entries are the author’s.)
5. ⟐ Rebuild the front door + split the reader-site from the maintainer repo; ship the Minimum Lovable Version — a welcome (not a changelog), the question in plain words, the three theories in two paragraphs each, the failed instruments above the fold, a glossary, a code decoder (done S4j), one person in the register of a life, and OPEN_QUESTIONS offered as the contributor front door. Requires the author to lift the self-imposed “home is canon README” rule.
6. ⟐ Let the 2026-10-01 clock fire — onto named real forkers and one human grader, after one genuinely external human reads the rendered site. The only move that touches the boundary and escapes the one-model paradigm; necessary-but-not-sufficient (the commons has not formed).

The deep author’s-calls (live fault lines, not deferrals): the person-vs-coldness fork (write a warm house voice, Le Guin — vs make the coldness the Luhmannian thesis); the chair as coupling-observer vs synthesizer (Luhmann’s structural challenge to the method’s core role); the two senses of “total” and whether differentiation theory competes as a rival rather than serving as Theory B’s backbone; Heidegger’s firing condition (what would confirm the ontological objection — the state of the world in which the honest move is to stop building); the fourteenth pressure-test (register-fairness / care promoted from dissent to a falsifiable test, or ruled out in the house voice); and the artifact question (repository, site, or book — the ambiguity is itself a delivery failure). Full detail: the improvement-roadmap PDF (Downloads) and panel/SESSION_S4J_FORUM.md §4.

The wrongs the plan now answers (ranked, root-first): the reflexivity trap (Tier 0 — one aquifer; only a foreign vantage lifts it → moves 5, 6); the empirical void + the honesty paradox (Tier 1 — nothing tested and by C-017 nothing testable in-setup; the confession may be persuasion not audit → moves 1, 2, 6); the unfireable falsifiers, the accretion, and register blindness (Tier 2 → moves 2, 3, and the fault lines); the delegation spiral and the front-door failure (Tier 3 → the moratorium + move 5). Full analysis: the wrongs PDF (Downloads) and panel/SESSION_S4J_FORUM.md §3.

Preserved dissent (the forum, unresolved): Heidegger (the forum deepens what it diagnoses — the critique became fuel); Le Guin (the apparatus manufactures obligations for one tired author; a living metabolism needs a rest cadence); Luhmann (a synthesizing chair presupposes the meta-position differentiation abolishes); Campbell (the honesty itself is now the gamed indicator). C-006 applies to this synthesis; the chair did not vote.