The Play Readiness Index

Industrial policy has shifted to operating on subnational geographies while its diagnostic apparatus remains national. We introduce the Play Readiness Index (PRI), an integrated subnational diagnostic that combines five dimensions of readiness — a subnational specialization ratio (state vs national, not country-vs-world Balassa), relatedness density, labor-market depth, clean-energy share, industrial-park infrastructure — with a binary scope-2 grid-credibility gate and an absolute minimum-scale filter for anchor classification, operationalizing how CBAM and IRA-linked procurement actually evaluate manufacturer qualification. The state export denominator comes from INEGI's entidad-federativa series, which corrects the customs fiscal-HQ attribution bias present in standard subnational trade data.

We classify each state-play cell into one of eight archetypes (anchor, near-anchor, fragile exporter, latent reserve, stranded capacity, infrastructure orphan, peripheral, transitional) and validate the framework across an 18-specification methodological cross-product, a 1000-draw weight bootstrap, and benchmark comparisons against three alternative measures.

Removing the scope-2 binary gate flips 23 cells — every one a promotion of a fossil-grid corridor state into anchor or near-anchor — validating its construction as a binding constraint rather than a continuous scoring dimension. The archetype each cell lands in determines which intervention is in scope (supplier development, clean-PPA procurement, blended finance, or grid reinforcement) and who owns it; the companion paper applies the framework to that diagnostic question.

The diagnostic apparatus for industrial policy is at the wrong unit of analysis. National-level instruments — the Economic Complexity Index (Hidalgo & Hausmann 2009), the standard Revealed Comparative Advantage ranking (Balassa 1965), the Green Complexity Index (Mealy & Teytelboym 2022) — answer questions about countries when industrial policy has shifted to operating on states, regions, and metro areas.

Three structural shifts in the policy environment of 2024-2026 make the unit-of-analysis mismatch increasingly costly:

• The post-2020 revival of targeted state action documented by Juhász, Lane & Rodrik (2024). The IRA's Section 30D, the EU's Net-Zero Industry Act, and Mexico's PIIRCE 2024-2038 all share a granular state-by-play structure.
• Scope-2 grid credibility as a binding force under the EU CBAM (Mehling et al. 2022) and US IRA-linked procurement preferences. Both regimes evaluate manufacturer qualification on grid-sourced emissions intensity at state level — Mexico's national grid average ~430 gCO2e/kWh sits within CBAM threshold, but auto-supplier corridor states source from grids at roughly 500-800 gCO2e/kWh (Nuevo León ~500, Querétaro ~560, Coahuila ~800), at or above the CBAM-compatibility ceiling.
• Capital-allocation patterns under nearshoring. Investment committees no longer treat "Mexican auto-supplier" as a unitary thesis; they treat each state-play cell as a distinct option whose readiness depends on a multidimensional configuration.

The empirical literature has begun to address each piece of this diagnostic gap separately, but none integrate capability with scope-2 credibility at the state-by-play resolution that policy now requires.

The PRI assembles five dimensions of subnational green industrial readiness, each addressing a distinct question implementers and investors actually ask:

• Revealed comparative advantage (RCA-MX) — what does the state currently export competitively in the play's HS6 basket? A subnational specialization ratio (state vs national, not Balassa's country-vs-world): play exports from BACI 2024, state denominator from INEGI's entidad-federativa export series (Method-B rescaled), which corrects the customs fiscal-HQ attribution bias that inflates CDMX and understates producing states.
• Relatedness density — how close is the state's existing capability basket to the play's product space? (Hidalgo et al. 2007; co-occurrence proximity at HS4.)
• Labor-market depth (DENUE) — what manufacturing employment exists in SCIAN sectors concorded to the play? Location quotient relative to national.
• Clean-energy share — what proportion of state generation capacity is zero-emission? (PRODESEN 2024 cuadros.) The scope-2 gate's carbon-intensity reading comes from Climate TRACE v7 (2024 facility-level emissions).
• Industrial-park infrastructure — net-zero-readiness scoring across AMPIP-registered parks (composite of grid-proximity, clean-PPA potential, play-relevant occupier mix).

The scope-2 credibility constraint operates as a binary gate, not a sixth scoring dimension. CBAM and IRA-linked procurement evaluate qualification at thresholds (qualify or not) rather than continuously. A state with strong capability but a fossil-dominant grid is disqualified from preferential access — its PRI summarizes the production architecture but the gate determines market access.

The composite is a weighted sum of the five within-play min-max-normalized dimensions (RCA and relatedness 0.25 each, labor-market depth 0.20, clean-energy share and park infrastructure 0.15 each). Because normalization is within-play, PRI ranks each state against its peers in the same play and is not comparable across plays; the archetype rules add the absolute floors — the fossil-share scope-2 gate and a $300M minimum-scale filter for anchor status — precisely to inject non-relative anchors (Figure 6). The eight-archetype taxonomy (anchor, near-anchor, fragile exporter, latent reserve, stranded capacity, infrastructure orphan, peripheral, transitional) is rule-based, not clustered, so each cell's archetype follows from a published rule book combining within-play percentiles, absolute gates, and the scale floor (Figure 7). Transitional is the rule book's residual branch and the largest single class (37 of 96 cells): for most cells no single constraint binds, and diagnosis is play-by-play.

The fragile-exporter construct identifies state-play cells with high export specialization (RCA ≥ 3) but a thin capability basket (relatedness density ≤ 0.30) — exporters whose position would not survive a moderate platform shock. Tlaxcala in EV components is the canonical case: RCA 7.4 on relatedness 0.03. This is distinct from the standard RCA-only ranking and is what the standard literature misses.

Three validation exercises:

• Cross-product sensitivity (18 specifications). Vary the weighting scheme (equal / informed / PCA), the composite aggregation (arithmetic / geometric), and the park-score aggregation (mean / median / max) across the 18 combinations. 85 of 96 cells classify identically across every spec; the 11 that move cluster on the near-anchor / latent-reserve boundary.
• 1000-draw weight bootstrap. Perturb the informed weights by ±0.10 (uniform) and re-classify across 1000 draws. 91 of 96 cells (95%) hold their archetype in ≥90% of draws; the rest sit on the near-anchor / latent-reserve boundary.
• Benchmark comparison to three alternative measures (Balassa RCA-only, the Hidalgo-Hausmann ECI, the Mealy-Teytelboym GCI), plus a PRI-without-gate counterfactual. PRI vs ECI Spearman ρ = −0.733 for auto-supplier (n = 25 states with a defined ECI) — the negative correlation is the point: ECI rewards diversity in complex products, PRI rewards depth in green plays under credibility constraints. Different objects, by design.

The three plays in NZIPL Mexico Cycle 1 — auto-supplier upgrading, EV components, grid hardware — produce different cell-level pictures:

• Auto Supplier (the discriminating case). Aguascalientes is the play's only anchor; Zacatecas, Puebla and Tlaxcala are near-anchors (1 anchor, 3 near-anchors, 4 stranded-capacity cells). Zacatecas clears every dimension and the scope-2 gate but its ~$69M of play exports fall below the $300M minimum-scale floor, so it stays near-anchor rather than becoming a denominator-artifact anchor. The deepest-capability corridor states — Querétaro, Coahuila, San Luis Potosí — are stranded-capacity: the scope-2 gate disqualifies their fossil-dominant grids. Across the three plays the framework yields 1 anchor : 18 stranded-capacity.
• EV Components (the fragile-exporter case). Tlaxcala ships heavily (RCA 7.4) on almost no relatedness (0.03) — high specialization with no diversification cushion: the fragile-exporter archetype, a position that would not survive a platform shock. No state reaches anchor; Aguascalientes is the lone near-anchor, and the capability leaders (Chihuahua, Tamaulipas, Nuevo León) sit in stranded-capacity — capability on grids that fail the gate.
• Grid Hardware (the no-anchor case). No state qualifies as an anchor — the capability leaders (Sonora, Nuevo León, Tamaulipas) are stranded-capacity: real manufacturing depth on grids that fail the scope-2 gate. Only Aguascalientes and Zacatecas reach near-anchor, on clean-grid plus park and labor strength rather than play-specific RCA. The play exposes how few states pair manufacturing depth with the clean grid investment-grade buyers require.

• Unit of credibility evaluation. Scope-2 thresholds in CBAM/IRA evolve. The binary gate's threshold is parameterized; the framework is robust to threshold movement, but specific cell archetypes shift.
• Trade-data vintage and denominator. Play exports use BACI 2024 (the latest pre-publication vintage); the state denominator uses the INEGI entidad-federativa series. Updates revise RCA-MX scores at the margin, and the 85/96 stability ceiling is a property of this vintage.
• Coverage and small samples. RCA is observed for 65 of 96 cells (the rest are structural export zeros), relatedness for 63, parks for 60; 56 cells have all five dimensions observed. The ECI comparator is defined for 25 of 32 states and the GCI for 17-22 states per play. Archetypes for thin-coverage cells are directional, not precise.
• Relatedness method. These v3.2 results use the HS4 co-occurrence relatedness heuristic; the production platform now ships an international product-space (phi) relatedness on a different scale. Adopting phi into the PRI is a deliberate v3.3 change, not a refresh — the fragile-exporter ≤ 0.30 threshold is defined on the v3.2 scale.
• Geographic resolution. Federal-entity is the operating scale; corridor and metropolitan-area resolution is feasible (and partially built into the production deploy) but introduces stronger sample-size constraints in DENUE.
• Generalization. The methodology applies to any country with state-level trade + employment + grid data. Brazil and India are next; other LMICs depend on subnational data availability.

Full archetype rule book, composite formula and weights, reproducibility instructions (pipeline at github.com/Gilix/nzipl-fragile-geographies, private — available to Lab collaborators), computational footprint, output schema, and figure index live in the source manuscript.