How we tested and what we deliberately left open.
Five lenses, three corridors, four fragility classes, eight scenarios. The threshold ladder, the corridor-edge correction that drove the strict-zero finding, the institutional adaptive-capacity floor, and the known limits this synthesis cannot resolve from publicly available data.
“Five lenses, chosen because earlier parts of the project had already assembled the evidence.”
We went with the five lenses we already collected evidence on in earlier parts of the project: AI exposure, demographics, disruption pathways, reskilling capacity, and careers data. The constraint was not which lenses might exist in theory. It was which ones we could test rigorously today, with European data, across 36 countries.
“One calibration choice changes the headline.”
This result hinges on one “dial” setting: where we draw the corridor cut-offs. Earlier drafts used cut-offs fitted to past literature (1.50 / 3.00). The published version uses theory-anchored cut-offs (1.20 / 2.80). With the strict rule (no wiggle room), the literature cut-offs produce three “Robust” markets; the theory cut-offs produce zero. That’s why we show the choice clearly and keep it reversible: readers can recreate the earlier count from the threshold ladder. We chose the theory anchor because it matches the Lens 1 framework.
“What we deliberately left open.”
Three known gaps constrain this analysis:
- Occupational detail: capability-floor breach is reported at a coarser job classification (2-digit ISCO) because the finer-grained European Social Survey microdata (3-digit) requires a multi-week application. The count here is a lower bound.
- Funding allocation: the EU’s adaptation funding through 2030 is not yet broken down by country. We can see the overall envelope, but not its distribution.
- Employer behaviour: we cannot observe live signals like hiring, retraining launches, or redundancy plans because they sit behind paywalls or in proprietary HR systems.
“We tested three candidate additions. None held up on its own.”
The EU Industrial Accelerator Act, proposed in March 2026, looked like a candidate ninth scenario. On close inspection, it runs the same absorption pathway as Climate Adaptation Boom: state-coordinated sectoral demand, with workers pivoting into new sectors. The US Inflation Reduction Act, three years post-implementation, confirms the mechanism class.
We also tested whether the wealth-fund-rich economies (Norway, Sweden, Denmark, the Netherlands, with Switzerland adjacent) deserved their own regime. The direction is empirically defensible (pension and sovereign-wealth differences are real), but the per-country threshold did not hold cleanly across all five candidates. Switzerland’s central-bank reserves do not behave like Norway’s sovereign wealth fund, so the line between “wealth-fund rich” and “post-growth” could not be drawn consistently.
A third test asked whether startup-driven absorption (Europe’s gap relative to the US in venture-scale firm formation) warranted its own scenario. ECB analyses document the gap clearly, and examples from other regions suggest the mechanism is possible. But each existing scenario in this taxonomy rests on something concrete (Cedefop’s per-country employment projections, the Net-Zero Industry Act’s committed €100 billion, the IRA’s three years of realised data), and startup-driven absorption in Europe has no equivalent anchor yet. Without a comparable anchor, it cannot be probability-weighted alongside the others.
The three rejections rest on different reasoning, but the takeaway is the same. Going forward, more than one of these mechanisms likely has to work at the same time: climate-led sectoral pivot, industrial-policy demand, wealth-fund buffering, startup formation, and the within-occupation reshape already in the taxonomy. The next part looks at what happens when several of those forces work together, and what that would cost.
The missing European countries
The 36-market panel covers the EU-27, the four EFTA states (CH, IS, LI, NO), the UK, and four EU candidate countries with sufficient data coverage to participate in the lens framework (BA, MK, RS, TR). Five other European countries are explicitly out of scope:
- Albania, Montenegro, and Moldova: EU candidates (since 2014, 2010, and 2022 respectively), but Eurostat / Cedefop / ESCO coverage is too thin to score reliably across all five lenses.
- Kosovo: partial-recognition status and a Stabilisation and Association Agreement, but the same data-coverage gap blocks inclusion.
- Belarus: under EU sanctions since 2020, no EU integration path, no comparable data access. Excluded on both grounds.
When coverage in these five improves to the resolution this framework requires, they can be added without changing the methodology.
1. Methodology Overview
Scope. 36 countries: EU-27 + EFTA-4 (CH, IS, LI, NO) + UK + 4 candidate-partial-coverage (BA, MK, RS, TR). Ukraine is carried as Class IV reference panel only: institutional bandwidth saturated, capability floor breached by definition; corridor classification (C1 / C2 / C3) does not apply.
Five lenses
Displacement velocity vs absorption capacity
The corridor-defining ratio. Absorption decomposed into a Zone-A-to-Zone-C transition rate keyed off the institutional system; regulated-absorption-friction scored 0.46–0.68 per country.
Demographic buffer
Tested against the 80% retirement-offset threshold. Per-country object: retirement_offset_pct, working_age_change_pct_to_2050, divergence tier, zone heterogeneity.
Distributional fold
Folded into the scale tag (aggregate / distributional / both) rather than carried as a separate dimension.
Compounding-crisis & jurisdictional buffering
Compounding-crisis shock count + squeeze flag with asymmetry score. EU AI-Act overlay counts: Annex III ~40, PLD ~29–31 carry the diagnostic weight.
Polycrisis drag
Composite at 2-digit ESCO-weighted ISCO with Klinger coordination-share weighting and a capability-floor breach test. Firm-level transition-vs-turnover framework carried as interpretive lens.
What the cut-offs mean
The Lens 1 ratio is displacement velocity ÷ absorption capacity: how much faster jobs are being displaced than institutions can move displaced workers into new ones. A ratio of 1.00 means the two are running at the same pace; a ratio of 2.80 means displacement is running at 2.8× absorption. Examples from the dataset: Norway 1.06 (displacement well-matched by absorption); Ireland and the UK 3.33–3.40 (displacement materially outpaces absorption). The corridor cut-offs (1.20 / 2.80) are where data and theory put the breakpoints; anchors in §2.
Three corridors (theory-anchored)
- C1 Managed Transition: Lens 1 ratio < 1.20.
- C2 Partial Absorption: 1.20–2.80; carries four within-corridor sub-clusters.
- C3 Displacement Without Absorption: ≥ 2.80; carries two within-corridor sub-clusters.
Eight scenarios
Seven routine variants (S1 New Jobs Replace Old, S2 Climate Adaptation Boom (Zone-C), S3 Jobs Transform, S4 Muddle Through, S5 Wage Cliff, S6 New Jobs Fail to Appear (Autor 2024 weakening), S7 Bandwidth Fracture), plus one conditional, S8 Polycrisis Drag, carried orthogonally to the routine grid. Probability vectors are quoted per regime with 80 % CI bands aligned to the IPCC AR6 likelihood scale.
Four fragility classes
- Class I Robust: ±1 of baseline AND no routine variant reaches C3.
- Class II Fragile: baseline stable, but at least one routine variant lands in C3 (typically S6 or S7).
- Class III Pre-Failure Risk: the Muddle-Through baseline lands in C3 after rescaling.
- Class IV Currently Failing: candidate-partial-coverage with extreme Lens 5 readings; the Ukraine reference panel calibrates the upper bound.
Three regimes
growth_baseline (24 countries, including 4 candidates); secular_stagnation_warning (EL, IT); post_growth_empirical (10 countries: AT, CH, DE, FI, FR, LI, LU, NO, SE, UK). Regime is scenario-conditional, not country-static.
2. Threshold-Locking Ladder From literature-fitted edges to theory-anchored edges.
The first pass fitted the corridor edges to historical literature: C1 < 1.50, C3 ≥ 3.00. We replaced these with the theory-anchored 1.20 / 2.80 pair, anchored on three independent sources:
- The first-pass sub-cluster boundaries themselves: the Nordic cluster ends at 1.10; the next cluster begins at 1.59. The 1.20 cap sits inside the empirically empty band between them.
- Autor et al. (QJE 2024) documenting reinstatement-effect weakening, which lowers the historical-base-rate-derived ceiling on “managed transition.”
- El-Sahli & Upward (2017) on structural lifetime-earnings deficits among displaced workers in C3-equivalent regimes.
The result was the strict-zero finding: under the corrected edges and the strict ±0 rule, the Class I count drops to zero. Even the Nordics fail. This is the structural-bias correction the synthesis is built around.
| Threshold pair | C1 cap | C3 floor | Rule | Class I |
|---|---|---|---|---|
| First-pass (literature-fitted) | 1.50 | 3.00 | Literature-fitted | varies |
| Corrected, strict | 1.20 | 2.80 | Strict ±0 | 0 |
| Corrected, relative-stable | 1.20 | 2.80 | ±1 of baseline | 16 |
| Corrected, asymmetric guard (the rule we applied) | 1.20 | 2.80 | ±1 of baseline AND no routine variant reaches C3 | 9 |
3. The corridor-edge correction Why the published rule is the corrected one.
The corridor edges were tightened during the analysis when the original literature-fitted thresholds returned a count of zero under strict robustness; the published rule is the corrected one. Concretely, the first version restricted the Class I scope to the seven routine variants and carried the parallel-cascade scenario as an orthogonal conditional, because cascade dynamics are mechanistically distinct from routine-variant displacement velocity. The second tightened the rule to a relative-stable definition (within ±1 corridor of baseline across all routine variants) once the strict ±0 rule under the corrected 1.20 / 2.80 edges returned a Class I count of zero. The third (the rule the synthesis publishes) added an asymmetric guard so that countries whose baseline sits in C2 but whose worst routine variant reaches C3 are no longer counted as “Robust” (this took the count from 16 to 9 and aligned the cluster with what “Robust” substantively means). The strict-zero finding from the second version is preserved as the structural-bias validation headline, not deleted by the third version’s restoration of Class I to 9.
4. Capability-Floor Breach Scope Ceiling 12 countries at 2-digit ISCO; the ceiling is the aggregation level, not the answer.
12 countries: BE, CH, DE, DK, IE, IS, LI, LU, NL, NO, SE, UK. DK is the marginal entrant at 2-digit. The count rose from 11 at 1-digit aggregation to 12 at 2-digit aggregation; the cascade priority distribution across the 12 is HIGH = 7, MEDIUM = 4, LOW = 1.
The scope ceiling is the 2-digit ISCO limit. A 3-digit pass would require multi-week Eurostat microdata access (flagged in the known-limits section below). The 12-country read should be treated as a lower bound at the 2-digit aggregation level; at 3-digit, DK’s status would resolve cleanly and the list could expand by 1–2 entrants in the Continental knowledge-economy band.
5. MFF Per-Country Allocation Gap A known gap, not an error.
The per-country share of the EU’s Multiannual Financial Framework allocation is intentionally left null in the underlying data. The €64.6 B mid-cycle reinforcement (Ukraine €50 B + migration €2 B + emergency €1.5 B + STEP and other components) is not disaggregated per Member State in publicly available Council documentation.
The reinforcement is a bandwidth-allocation proxy at EU aggregate level; its existence is the concurrent-crisis tax on regular spending priorities, which is the political-economy signal Lens 5 captures. Per-country disaggregation via national contribution and rebate analysis would tighten the diagnostic; this is flagged in the known-limits section.
6. Candidate-Country C2 Sub-Cluster Routing Bosnia and Herzegovina, North Macedonia, Serbia, Turkey.
The four candidate-partial-coverage countries sit at C2 baseline but lack the institutional-system tags assigned to confirmed EU / EFTA / UK cases. For C2 sub-cluster purposes they are routed to central_eastern_european_in_c2 alongside BG, LV, RO. The institutional-similarity proxy is a weighted average over Central / Eastern European plus Southern European cases, not a direct institutional-system match.
Readers should not conflate candidate-baseline-proxied sub-clustering with confirmed institutional similarity. MK and RS additionally carry the Class IV currently-failing flag; TR is both Class IV and s2-dependent.
7. Known limits: what this synthesis cannot see Five gaps that bound the analysis.
- 3-digit occupational breach scope. Eurostat microdata access at the 3-digit level requires a multi-week application; resolving it would clean up DK’s marginal entry and likely add 1–2 Continental knowledge-economy entrants to the 12-country breach list.
- MFF per-country allocation via national contribution and rebate analysis. Council documentation does not disaggregate.
- Backporting Lenses 1, 4, and 5 for Ukraine to enable corridor-map participation under a wartime-economy variant.
- Live external intelligence. Recent moves on Draghi-track competitiveness funding, the latest national budget rounds, and ALMP reform announcements are out of scope; this synthesis draws on structured artefacts (data files, methodology notes) but does not pull live external data.
- Country-level internal-transition-vs-external-turnover diagnostic. The firm-level framework is cited from Part 5 of the suite; country-level data acquisition remains an open target.
None of these block the synthesis surface as published.
8. RMS volatility as a complement to corridor classification A continuous stress-sensitivity index alongside the categorical Class I–IV assignment.
The headline corridor assignment per country (C1 / C2 / C3) and the fragility class (I–IV) are categorical. To complement those, we computed a continuous stress-sensitivity index per country: the root-mean-square of corridor placement across the seven routine scenarios (S1–S7), excluding the conditional S8 Polycrisis Drag scenario. Mapping is C1→1, C2→2, C3→3; the RMS is sqrt(mean(corridor² across S1–S7)), persisted per country as corridor_volatility_rms in the underlying data file.
RMS preferentially weights higher corridor placements (it squares the values before averaging), so a country whose scenario portfolio leans into C3 reads more stressed under RMS than under the arithmetic mean. Conversely, a country whose scenarios cluster cleanly at one corridor reads at that level under both measures. The metric is additive only and does not change any existing class or corridor assignment.
Where the RMS drifts meaningfully from the baseline phase-3 corridor (drift ≥ 0.5 with Class I or II, or drift < 0.2 with Class III), we surface those countries as candidates for review, not as automatic reclassifications. The categorical structural framework is the published call; the RMS adds a sensitivity check around it.
At the pan-European aggregate level, the RMS Lens-1 ratio is computed in both unweighted and population-weighted variants, alongside the existing population-weighted arithmetic mean. The unweighted RMS implies a heavier C3 read than the weighted arithmetic mean for both scopes; the population-weighted RMS lifts the European-36 read into C3 territory while the EU-27 read holds in C2. The variation guard remains the load-bearing pan-European statement: the class spread is the honest read, not any single aggregate scalar.
The methodology variant is presented as a complement, not a replacement, for the categorical corridor and class assignments. Full per-country tables and reclassification candidates are available in the underlying data file under corridor_volatility_rms per country and under rms_lens1_ratio_unweighted / rms_lens1_ratio_population_weighted in the pan-European aggregate block.
Einfache Sprache · coming soon
This page is being prepared in Einfache Sprache. The standard version is available; click Standard to read it.
Diese Seite wird in Einfacher Sprache vorbereitet. Die Standard-Version ist verfügbar.