# SEDZ Scoring Codebook (audit remediation A3)
**Roman Economic & Production Centers project** · 2026-07-07

> ## SCOPE — this is the ANCIENT instrument (`ancient_v1`)
> Every rule below was written to discriminate among **ancient production régimes**: what counts as a
> perimeter of differential law, what counts as administrative allocation of a deposit, what counts as a
> fiscal apparatus. Its anchors are Roman *metalla*.
>
> **It was nevertheless applied, unchanged, to three modern zones** (Shenzhen SEZ, Mauritius EPZ, the
> free-port type), which then sat in the same 28-row table as Vipasca and scored on the same 0–2 scale.
> That is a category error the table could not express: Shenzhen's 1 and Vipasca's 1 are the same
> *magnitude* of categorically different things, and — on the fiscal axis — of **opposite sign**.
>
> The modern cases are now additionally scored by a second instrument, `modern_v1`
> ([SCORING-CODEBOOK-MODERN.md](SCORING-CODEBOOK-MODERN.md)), which issues a `structural_analogy`
> verdict in its own vocabulary. The `ancient_v1` genus verdicts on those three cases are **retained,
> not deleted**, and labelled as `ancient_v1` artifacts — see that codebook for why deletion would have
> been the worse dishonesty.
>
> **Do not read an ancient_v1 magnitude and a modern_v1 magnitude as commensurable.** They are
> comparable at the level of *type* (does the architecture recur?), never at the level of *degree*.

Anchored 0/1/2 decision rules for the `sedz_assessment` axes, so scores are reproducible and the fragile 0/1 boundaries are explicit. This addresses the audit's finding that magnitudes were single-rater soft-coding without a codebook. **A blind second-rater pass + inter-rater κ remains outstanding** (see "Reliability" below) — this codebook is the prerequisite for it.

## Standing rules
- Score **only** from evidence in the entry's `evidence[]`; if a feature is not attested, score by what IS attested, not by analogy to Vipasca.
- Every magnitude cites `evidence_ref` indices. "Inferred from a parallel" caps a score at 1.
- **Genus = all three convergence axes ≥ 1.** A score at the 0/1 boundary that flips genus membership must be flagged in the entry note as fragile.

## Convergence axes

### 1. bounded_jurisdiction — *a perimeter of differential law + administration displacing civic government*
- **2 (full):** attested *fines/territorium* with differential law at the edge **AND** a resident authority exercising jurisdiction where no civic *ordo* governs (*extra civitatem*). Test: is there a surviving boundary clause / penal expulsion *finibus* / a procurator with delegated *cognitio*? (Vipasca, Mons Claudianus, Simitthus, Mons Porphyrites, Dokimeion.)
- **1 (partial):** administrative distinctness (imperial estate/district, resident official) but **either** no attested differential-law perimeter **or** the boundary/jurisdiction is inferred/contested. Also: a special legal regime that does **not** displace civic government (a modern SEZ over an intact municipality; an estate within a *civitas*). (Karystos, Alburnus, Noricum [contested], Henchir Mettich, Shenzhen, EPZ.)
- **0 (absent):** ordinary civic/municipal territory, no differential law, no jurisdictional perimeter — **an accounts office (*tabularii*) or mere imperial ownership is NOT a jurisdiction.** (Luna [corrected], Proconnesus, La Graufesenque, garum cetariae, Baetica, Aphrodisias-civic, imperial figlinae on ordinary land.)
- ⚠ **0/1 boundary (the fragile line):** imperial *property/administration* without a legal perimeter = **0**, not 1. Luna was corrected 1→0 on this rule.

### 2. allocated_exploitation — *the administration allocates productive access* (mode-neutral)
- **2 (full):** the administration itself grants/assigns productive access — by concession/*occupatio*/*adsignatio* with revenue-sharing and deadlines, **OR** by direct fisc+army working. Test: does the state decide *who may extract/produce here*? (Vipasca, Mons Claudianus, Ptolemaic monopoly, Henchir Mettich.)
- **1 (partial):** access allocated but indirectly or weakly — the state farms a *vectigal* over self-occupying producers (the fisc does not itself grid/assign), or firm-licensing of private producers. (Noricum [corrected], Luna, Baetica, Shenzhen, EPZ.)
- **0 (absent):** private market entry; the administration allocates no production (it taxes transit — the *portorium*). (La Graufesenque, garum, portorium, Aphrodisias-civic.)

### 3. fiscal_monopoly_apparatus — *dedicated vectigalia + money-of-account + licensed service monopolies as one system*
- **2 (full):** a bounded revenue system fencing the site — farmed *vectigalia* AND licensed service monopolies AND/OR a distinct money-of-account or price/output controls. (Vipasca [service monopolies]; a state commodity monopoly with fixed price = 2, e.g. Ptolemaic, Sisapo-minium.)
- **1 (partial):** a real fiscal apparatus (procuratorial fisc, rents, consignment, portorium) but NOT the farmed local monopolies + money-of-account. (most metalla/quarries; Baetica annona; Shenzhen customs.)
- **0 (absent):** no dedicated fiscal apparatus fencing the site. (La Graufesenque, garum, Aphrodisias-civic.)

## Divergence axes (magnitude 0/1/2 + signed polarity)
- **regime_direction:** polarity `dirigiste` (the special regime is *heavier* than the surrounding norm — but note: heaviness via *farmed/auctioned* instruments is still delegated/market-mediated, not central planning) vs `liberalizing` (lighter — tax/customs/labor relaxation). Magnitude = strength of divergence from the provincial/national norm.
- **orientation_beneficiary:** `state_fiscal_revenue` (fisc/basilikon) vs `private_export_profit` (FDI/firm). **Mixed** where a profit-taking contractor (publican *societas*) captures surplus alongside the fisc (Sisapo).
- **labor_regime:** `convict_coerced_penal` (judicial *damnatio*) vs `incentivized_free_waged` vs `mixed`. **Do not code chattel slavery as penal** — chattel slavery is privately-owned market-supplied capital (Laurion, corrected penal→mixed), categorically distinct from a judicial sentence.
- **capital_attraction:** `participatory_liberalitas_grant` (sovereign grant/liberalitas) vs `tax_exemption_holiday` (SEZ instrument) vs `market_lease` (competitive auction, e.g. Laurion/Sisapo).

## bounded_jurisdiction tie-breaker (IRR 2026-07-07) — the fragile 0/1 line
The blind-rater pass found the sole genus disagreements at BJ for **state-regime-within-a-functioning-city** cases (Sisapo: state commodity monopoly + a live town; Laurion: mining-law leases in Attic civic territory). The "1 = special regime not displacing civic govt" clause and the "0 = ordinary civic territory" clause both apply. **Rule:** for such cases, record BOTH readings and treat genus membership as **conditional/fragile** (do not present it as settled); a domain panel adjudicates. Current fragile set: **sisapo, laurion, luna, proconnesus, karystos, noricum.**

## Reliability (IRR done; α pending)
A blind independent second rater scored all 28 cases from this codebook alone (see `_audit/IRR-REPORT.md`): linear-weighted Cohen's **κ = 0.87 / 0.84 / 0.82** on the three convergence axes, **genus κ = 0.85**, within-1 agreement 100% — "almost perfect" (Landis & Koch). The instrument is therefore a **criterion-explicit, reproducible** rubric, not merely single-rater coding. Outstanding: a **third** rater for a Krippendorff's-α (the 2-rater κ is solid but a 3-rater α would tighten it); and the fragile-boundary cases above remain conditional classifications, not settled ones.