| Category | SHAP % | RCA | Classification |
|---|---|---|---|
| Machinery | 33.5 | 0.47 | Critical gap |
| Chemicals | 30.3 | 0.19 | Critical gap |
| Industrial Materials | 25.1 | 0.92 | Build-up |
| Metals | 11.1 | 0.55 | Build-up |
Reading the Capability Map
A Handbook for SHAP–RCA Gap Analysis — Model Implications for Industrial Strategy
What this is. A working handbook for reading the two model-driven panels of the Atlas of the Clean Industrial Base — the Capability Radar and the PC Scatter — and turning them into industrial-strategy calls. It explains the one idea both panels encode (a gap between what a technology needs and what a country has), the seven-cell grid that names every gap, and a worked example for each cell drawn from live model output (RCA reference year 2024).
1 How to use this handbook
Read §2–3 once to learn the framework, then treat §4 as a reference: each of the seven cells has a plain-language meaning, a reading rule, a policy orientation, and a real worked case. §5 walks a full country read end to end. §6 covers what the map does not say — the limits matter as much as the signal.
The framework is deliberately coarse. It classifies clusters of capability, not individual products. As the underlying working paper puts it: precision beyond the cluster level “would be applying a false level of precision to the model.” The handbook is a guide to reasoning, not a recipe.
2 The two questions behind every panel
Both panels answer two questions at once and plot one against the other.
What does the technology need? — the SHAP weight. The competitiveness model (a Random Forest predicting whether a country reaches RCA > 1 in a clean technology) tells us which capabilities carry the predictive signal. Aggregating the model’s SHAP importances to five capability clusters — Chemicals, Electronics, Industrial Materials, Machinery, Metals — gives a weight \(w_c\) (% of the technology’s total importance) for each cluster. High weight = the model says this capability is decisive for the technology.
What does the country have? — the RCA. Revealed Comparative Advantage measures whether a country exports more of a cluster than its fair world share. \(RCA \geq 1\) means genuine specialisation; below 1 means no revealed strength. RCA is already normalised for country size and world trade, so it is comparable across countries.
| Panel | SHAP weight (need) | Country RCA (have) | Granularity |
|---|---|---|---|
| Capability Radar | gold polygon, 5 axes | green polygon, 5 axes | Category (5 clusters) |
| PC Scatter | x-axis | y-axis (bubble = trade value) | Product (HS6) |
The radar and the scatter are the same comparison at two zoom levels. The radar averages each cluster; the scatter shows the individual products inside it. This distinction is not cosmetic — it is the single most common source of misreading (§4.1).
3 The gap framework: a seven-cell grid
Crossing three SHAP levels with three RCA levels yields a 3×3 grid that collapses to seven policy-relevant cells.
SHAP level (what the technology needs)
| Level | Threshold | Meaning |
|---|---|---|
| High | \(w_c \geq 25\%\) | Critical driver — dominates the model’s signal |
| Medium | \(10\% \leq w_c < 25\%\) | Important contributor |
| Low | \(w_c < 10\%\) | Supporting role only |
RCA level (what the country has)
| Level | Threshold | Meaning |
|---|---|---|
| Strength | \(r_c \geq 1.0\) | Genuine specialisation |
| Partial | \(0.5 \leq r_c < 1.0\) | Approaching competitiveness |
| Weak | \(r_c < 0.5\) | No meaningful specialisation |
The grid
| Strength \(r\geq1\) | Partial \(0.5\leq r<1\) | Weak \(r<0.5\) | |
|---|---|---|---|
| High SHAP \(\geq25\%\) | 🟢 Leverage | 🟠 Build-up | 🔴 Critical gap |
| Medium SHAP 10–25% | 🟡 Mature strength | 🟠 Build-up | 🟥 Gap |
| Low SHAP \(<10\%\) | 🔵 Bonus strength | ⚪ Not a priority | ⚪ Not a priority |
The diagonal logic is the whole point: the top-left is where a country’s strength meets the technology’s need (act here first), and the top-right is where the technology’s most important need is unmet (the hardest, most strategic problem).
4 The seven cells, with worked examples
Each cell below pairs the reading rule with a real case from model output (reference year 2024). The three “hero” cases you will see recur — India · Solar, Brazil · Biofuel, Canada · Batteries — because a single country×technology typically spans several cells across its five clusters. That is normal and is exactly how the map is meant to be read.
4.1 A note before the cells: radar vs. scatter
On the radar, Brazil · Biofuel shows no Leverage cell — its high-need clusters (Machinery, Chemicals) sit in Build-up / Critical gap. Yet on the PC scatter, Brazil holds clear Leverage products: sugarcane-derived ethanol and processing inputs where its product-level RCA is far above 1 and the model weights the product heavily. Both readings are correct. The category average dilutes a handful of star products; the scatter resolves them. Rule: use the radar to set the strategic frame, the scatter to find the specific products to anchor.
4.2 🟢 Leverage — strength meets need
Reading rule: High SHAP (\(\geq25\%\)) and RCA \(\geq 1\). The country already specialises in a capability the technology most demands.
Policy orientation: Anchor FDI and supply-chain strategy here. This is the nucleus to build outward from — the country competes where it counts.
Worked case — Brazil · Biofuel (product level). Brazil’s sugarcane-ethanol complex is the textbook leverage position: the feedstock-processing products the biofuel model weights most are exactly the ones Brazil dominates in world trade. On the scatter these bubbles sit top-right; the radar’s Metals/Industrial-Materials clusters carry the diluted signal of the same strength.
4.3 🟠 Build-up — close enough to matter
Reading rule: High or Medium SHAP and Partial RCA (\(0.5 \leq r < 1\)). The country is just below the competitiveness line in an important capability.
Policy orientation: Highest expected return on targeted investment. A modest push to cross \(RCA = 1\) converts a near-miss into a genuine strength.
Worked case — Canada · Batteries.
| Category | SHAP % | RCA | Classification |
|---|---|---|---|
| Chemicals | 48.6 | 0.94 | Build-up |
| Machinery | 32.9 | 0.53 | Build-up |
| Metals | 9.5 | 0.49 | Not a priority |
| Electronics | 6.0 | 0.37 | Not a priority |
| Industrial Materials | 2.9 | 0.00 | Not a priority |
Canada’s two highest-need battery clusters — Chemicals and Machinery — sit in Build-up (RCA just under 1). With strong upstream minerals and existing chemical capacity, this is a country one targeted step from competitiveness in the capabilities batteries most reward.
4.4 🔴 Critical gap — the top driver is missing
Reading rule: High SHAP (\(\geq25\%\)) and Weak RCA (\(r < 0.5\)). The model’s single most important capability is absent.
Policy orientation: The hardest problem. Requires upstream industrial policy — sector-building, not just incentives. Cannot be closed by demand subsidies alone.
Worked case — India · Solar.
| Category | SHAP % | RCA | Classification |
|---|---|---|---|
| Chemicals | 38.9 | 0.50 | Critical gap |
| Metals | 23.7 | 0.91 | Build-up |
| Electronics | 22.7 | 0.17 | Gap |
| Machinery | 12.0 | 0.72 | Build-up |
| Industrial Materials | 2.7 | 0.24 | Not a priority |
India assembles modules competitively but the model’s top-weighted cluster for solar — Chemicals (the polysilicon / cell-chemistry layer) — sits in Critical gap. This is the well-known “deep upstream” vulnerability: strong at the visible end of the chain, absent where the model says competitiveness is actually decided.
4.5 🟡 Mature strength — real but not decisive
Reading rule: Medium SHAP (10–25%) and RCA \(\geq 1\). A genuine supporting capability in a cluster of secondary importance.
Policy orientation: Maintain and deepen. Useful, but not enough on its own to carry a technology strategy.
Worked case — Canada · Nuclear. The Chemicals cluster (SHAP 22%, RCA 5.71) is a maintained specialisation that supports, but does not by itself decide, competitiveness in this technology.
4.6 🟥 Gap — a secondary driver missing
Reading rule: Medium SHAP (10–25%) and Weak RCA (\(r < 0.5\)). An important-but-not-critical capability is absent.
Policy orientation: Worth addressing, but not a show-stopper alone. Sequence it after the critical gaps.
Worked case — India · Solar · Electronics. Alongside its Chemicals critical gap, India’s Electronics cluster for solar sits in Gap — a real but second-order weakness (inverters, power electronics) to address once the upstream chemistry is underway.
4.7 🔵 Bonus strength — strength the tech doesn’t reward
Reading rule: Low SHAP (\(<25%\) — under 10%) and RCA \(\geq 1\). A genuine capability in a cluster this technology weights lightly.
Policy orientation: Do not anchor tech-specific strategy here — but flag it: this strength may be decisive for a different technology. A cross-technology asset.
Worked case — Brazil · Heat Pumps. A strong Chemicals position (RCA 11.71) that the Heat Pumps model weights only 4% — real capability, but better leveraged for a technology that prizes it.
4.8 ⚪ Not a priority — neither side cares
Reading rule: Low SHAP and Partial/Weak RCA. Neither the technology nor the country emphasises this cluster.
Policy orientation: Low-salience. Note and move on — strategy capacity is better spent on the actionable cells above.
Worked case — India · Solar · Industrial Materials. Low model weight, low Indian RCA: correctly ignored in a solar strategy.
5 A full read: India · Solar, end to end
Read top to bottom by need: Chemicals is the top driver and a Critical gap (the strategic priority). Metals and Machinery are Build-up — close, high-return targets. Electronics is a secondary Gap. Industrial Materials is Not a priority. One panel, a complete sequencing of where to act and in what order.
6 What the map does not say
- Clusters, not products. A high-SHAP cluster signals what kind of capability matters, not a list of HS6 codes to build. Use the scatter to descend to products, but treat product specifics as illustration, not prescription.
- Radar dilutes; scatter resolves. A Build-up cluster can hide a Leverage product (and vice versa). Always cross-check the two panels (§4.1).
- Macro features are excluded from the radar. The model also learns from GDP, governance, patents and other country covariates; the radar shows only the five traded-capability clusters. A country can have a clean radar yet face non-trade barriers.
- RCA is a revealed, lagging measure. It reflects what a country has exported, not nascent capacity. Pair with the trade-timeline and firm-activity panels for momentum.
- Thresholds are conventions. The 25% / 10% and 1.0 / 0.5 lines are defensible but not magic. Cases near a line should be read as “between cells,” not snapped to one.
7 One-page cheat sheet
| Cell | Need (SHAP) | Have (RCA) | Do |
|---|---|---|---|
| Leverage | High | Strength | Anchor FDI; build outward |
| Build-up | High/Med | Partial | Invest — highest return; cross RCA = 1 |
| Critical gap | High | Weak | Sector-build upstream; hardest problem |
| Mature strength | Med | Strength | Maintain and deepen |
| Gap | Med | Weak | Address after critical gaps |
| Bonus strength | Low | Strength | Don’t anchor here; asset for another tech |
| Not a priority | Low | Partial/Weak | Note and move on |
Radar = category zoom (set the frame). Scatter = product zoom (find the anchor). Reference year 2024.