THE REPAIR-EPISODE UNIT — the case-coding unit (DRAFT criteria; nothing coded; the freeze is the author’s)

Version: 0.1 · Status: DRAFT — panel-delegated (S4h standing delegation), provisional-pending-author. The boundary criteria below are DRAFT until the author confirms them on return: the freeze event is the author’s confirmation, never a panel act under the standing delegation, and no “pre-registered / frozen” claim is made before that act (panel/SESSION_S4H_RATIFICATION.md B1). This document carries ZERO coded episodes and ZERO empirical claims — definitions, criteria, and an uncoded candidates list only; any sentence in it asserting a historical claim is out of scope and takes a Rung-0 flag (the S4h zero-claims clause). · Last updated: Session 4h (Code)

Governs under: docs/PRESSURE_TESTS_CAUSAL_HYPOTHESIS.md (the thirteen-node map; §8 lag types) · From: panel/LEARNING_FROM_PARALLEL_S4g.md adoptions 1 + 6 · Discipline: skills/study-discipline/SKILL.md.

Why this unit exists. The thirteen-node causal hypothesis has arrows but no sampling unit — nothing tells a coder what one observation is. The repair episode is that unit: a bounded historical or contemporary passage in which a system is destabilized and a repair is attempted. It unblocks Q-001 with codable clocks (model plus coded episodes is exactly the cliodynamics pairing — Turchin), and it is where the §8 lag-type decomposition gets its data. Turchin’s binding condition governs everything here: inclusion and boundary criteria precede any coding, or episode coding is anecdote collection with a schema.

1. The unit

A repair episode is a passage through six stages: destabilize → sense → legitimate → coordinate → act → recover — with the last stage explicitly allowed to be absent (§3, right-censoring). The unit is the episode, not the polity: one polity may hold many episodes; one episode may span institutions.

The four clocks (the initial instrument set — pinned; any additional clock or sub-clock requires its own pre-registration entry before first use, Meadows):

No clock-denominated claim is reported below a pre-registered inter-coder reliability floor on clock readings (Campbell); clock readings from unfrozen criteria are void (§4).

2. Boundary and inclusion criteria (DRAFT — the author’s confirmation is the freeze event)

1. Inclusion: a candidate episode enters the base only if (a) a dated destabilization event is documentable from admissible sources (§2.3); (b) at least two of the six stages are independently documentable; (c) the episode’s boundaries — onset and close/censor — can be stated so that a project-blind external coder could apply them from a dossier alone, without house context (Ostrom’s decidability bar).
2. Exclusion (stated with the same precision — Meadows): NOT an episode: gradual secular drift with no dateable destabilization; a pure policy proposal never legitimated or acted on; a crisis narrative whose only source is retrospective synthesis (no contemporaneous documentary layer); an episode whose boundaries can only be drawn by outcome-knowledge (boundary-by-hindsight is selection on the dependent variable).
3. The case-universe (declared sampling frame — Turchin): the population from which episodes are drawn is declared before coding: polities/systems with a surviving contemporaneous documentary record, drawn from the admissible source classes — published documentary editions, court/administrative records, dated official acts, contemporaneous correspondence/chronicles with established provenance, and (for contemporary episodes) the dated public record under the two-dossier discipline. Selection-on-the-dependent-variable is only checkable against a stated frame; episodes may not be nominated from outside it.
4. Right-censoring (Turchin, binding): an episode that never reaches recover is data, not a non-case — coded with time-to-repair censored at the record’s end. Failed repairs must be codable, or the base is survivorship-biased at birth.
5. Repair-for-whom (Nietzsche, binding): every episode coding carries a repair-for-whom field — whose order was “recovered,” and who paid for the recovery. “Repaired” is a valuation; the coder signs it.
6. Void-plus-catch (Turchin): any episode coded before the author-confirmed freeze of these criteria is void and logged as a catch.

3. What this unit cannot see (mandatory section — Le Guin, binding; enacted as a real design fork, not a caveat)

The destabilize-first grammar requires a destabilization event to open an episode — so it is structurally blind to maintenance-shaped repair: care systems, mutual aid, prevention that succeeds, where the repair consists of the crisis never arriving. The open design fork (for the author, with the criteria freeze):

• Option (i) — a maintenance-episode variant: a second opener — a documented near-miss or sustained-load evidence (the flood that did not breach because the levee was maintained; the epidemic curve that stayed flat) — with its own clock semantics defined before first use (time-to-sense from the near-miss’s documentation, not from a destabilization that never occurred). This is a real second unit-type, not a smaller episode.
• Option (ii) — the exclusion recorded as widening debt, dated, in this document, compounding beside the case pool’s existing debt (D-005).

Until the author cuts this fork, no maintenance-shaped candidate is coded and the blindness stands named. An episode grammar that opens on destabilization will code the world’s loud repairs and call the coverage complete; the care that prevents the episode from existing is the invisible load-bearing wall (Le Guin, as preserved in-session).

4. Sequencing gates (all must clear before any episode is coded)

1. The author confirms §2 (the freeze event — Rule 11; pre-registration under delegation fixes criteria against post-hoc fitting, it does not substitute for the one non-self-administered check on the criteria themselves, C-027).
2. The E-cluster validations return (the repaired severity/convergence instruments pass their cross-model acceptance — coding a new unit through an instrumentarium the project’s own external coders just broke would inherit C-025/C-026 by construction — Ostrom).
3. A project-blind cross-model red-team of the draft criteria (via the author-mediated kit route) is run before the freeze — the drafting hand demonstrably held the candidate list below, so the criteria themselves get the external read (S4h adversary, adopted).
4. The §3 fork is cut by the author.

Any coding before all four gates clear is void-plus-catch (§2.6). (Per item H of the S4h session: this document’s entry into canon opens no claim-surface — it is a sampling-frame definition, codable by no one until the gates clear.)

5. The historical episode base — candidates (names only; nothing coded; episode base first, canon expansion second)

The anchoring caveat (Turchin, as preserved in-session, binding): these candidates were named before the boundary criteria existed and were known to the criteria’s drafters; that anchoring is disclosed, not cured. (And per the S4i shadow-grade pass: the why/why-fail sentences below are the drafting hand’s unverified impressions, signed as such — not findings; nothing in this table has been checked against any source.) The no-priority rule: the coded episode base is drawn by applying the frozen criteria to the declared case-universe; the named candidates receive no sampling priority and confirm nothing by having been named first. Each entry is name + period + one why-a-candidate sentence + one why-it-might-FAIL-inclusion sentence (Ostrom’s adversarial line) — no episode narrative, no empirical claim.

The signed selection note (Nietzsche, binding): the first four were chosen for documentary density and familiarity to the drafting model — the archive’s favorites. Named omissions, as load-bearing as the list: no African episode, no pre-Columbian episode, no subaltern-led repair among the first four; the two quiet-register candidates were named by the same contaminated hand and extend, not cure, the anchoring. If no quiet candidate can be documented to §2’s criteria, that absence is recorded, dated, as widening debt — not silently dropped. (Campbell’s in-file freeze-order — criteria section hash-frozen before this section is appended, hashes recorded — is retained as non-curative process hygiene, subordinate to the disclosure above: a hash proves textual sequence, not epistemic independence.)

6. Preserved dissent (travels with the unit)

• Heidegger (rejects the unit entirely — the panel’s reconstruction of his position, preserved from the deliberation; not a published quotation, per Ground Rule 16): “To code repair is to have decided that healing is a process with a clock. What recovers without a legitimating act — quietly, slowly, outside the six stages — will be coded as nothing having happened.” And on the candidates: “Rome’s third century was a world, not a data point. To shelve it as a ‘candidate natural experiment’ is to have already lost what it could teach.”
• Turchin (his yes is conditional, standing): episode coding without pre-registered inclusion and boundary criteria, applied to a declared case-universe, is selection on the dependent variable with a schema attached; if coding ever begins before the freeze, every episode so coded is anecdote, and that will be said in every reading that cites them.
• Le Guin (standing): four empires reforming is not a sample of repair; it is a sample of what archives keep. The episode base is born owing the same debt as the case pool — dated above, so the interest is legible.
• Nietzsche (standing): “repair” unsigned is the victor’s word for what happened; the repair-for-whom field is what keeps the coder honest, and the omissions list is as load-bearing as the candidates list.
• The sentinel: these criteria were pre-registered under a standing delegation; the author’s confirmation is not a formality but the only non-self-administered check the criteria will ever get before they start selecting history.

A sampling unit, its criteria drafted-not-frozen, its candidates named-not-coded, its blindness named on its face — built the session after the project’s first externally-coded results, under a delegation booked as a cost. Panelist names in this document denote the project’s simulated panel personas (Ground Rule 16), not statements by the persons. Author: Hulki Okan Tabak — with Claude · License: CC BY-SA 4.0.