Roadmap / 0.1.2.11
NATO split, member by member
v0.1.2.11, commit 02380ea. File docs/RESULTS_v0.1.2.11_nato_split.md.
Written at this version and not revised since. Later versions may change these numbers; see the roadmap.
Switch: nato.model: members (the harness baselines now use it; coalition keeps the v0.1.2.8 single draw). nato.scenario_class: S1 economic coercion, S2 limited seizure, S3 full invasion.
Options: tripwire_troops, warning_months, us_counter_moves, us_threatens_europe, russia_concurrent (draw/true/false), us_uses_european_bases, eu_directly_targeted, uk_government (current/reform), france_government (current/rn).
Runs: 3,000 paired runs per variant, every other module on. 21 tests pass; all harness checks pass.
Where the inputs come from
The deep-research report of 12 November 2026 ("How NATO Allies Would Respond if the US Coerced or Attacked Canada"), stored as data/alliances/nato_members.yaml: 30 non-US members, six partners and the EU, each with a bloc, tier probabilities per scenario, delays and correlations.
These probabilities are the report's analytical estimates, not forecasts. The report says they encode ordering and correlation more firmly than absolute levels, and it flags several inputs as secondary or unverified (the Arctic Endurance participant list, some 2026 election outcomes, Italy's disputed Sigonella denial). GDP weights are my approximations.
Its documented anchors: - Greenland, January 2026: 7 leaders signed a statement on day 1; 11 allies sent or offered token forces within about 10 days; 8 were threatened with tariffs; Washington backed down in 15 days. Czechia, Slovakia, Türkiye and (then) Hungary dissented. - Iran war, 2026: Spain closed its bases and airspace to the US; France refused overflight; Italy reportedly refused Sigonella. Washington ordered 5,000 troops out of Germany and floated suspending Spain. - Dependence: Germany, Italy and the UK host about 87% of US troops in Europe; 58% of European NATO arms imports come from the US.
How the module works
- One correlated draw per ally. Each ally's tier (words, sanctions and basing, materiel, forces) comes from a single latent draw: a common European factor, a group factor (Nordic 0.8, Baltic 0.85, Benelux 0.7, Visegrád 0.4) and its own noise. A shared band position moves all probabilities inside the report's low–high bands.
- Modifiers (sizes from the report unless marked as my estimate in
model.yaml): - US counter-moves chill second movers.
- A US threat to abandon Europe makes frontline states hedge and western states harden.
- A Russian move suppresses materiel and forces.
- A UK move to materiel pulls the core along; Poland moving beyond words pulls the Baltics and Romania.
- Allied soldiers killed in Canada harden the core.
- Reform UK or RN governments weaken the lead movers.
- EU. Collective trade measures need the EU's own draw, halved without Germany. EU members cannot sanction alone, but any ally can deny basing on its own.
- What the engine receives: three strengths from 0 to 1 instead of one level.
- Sanctions: half basing-and-access denial (US troops hosted, plus the four North Atlantic access points: UK, Lajes, Keflavík, Pituffik), half the coalition's economic weight.
- Materiel: donor weight times delivery feasibility (0.3–0.5 into an invaded Canada, per the report).
- Forces: only allies that would actually fight (10–25% of core-bloc deployers).
The engine's allied-support level became continuous for this. The golden check confirmed the coalition model's results did not change.
Results (chance the US quits; change vs NATO off, 95% interval)
| Variant | Full annexation (S3) | Alberta detachment (S2) | Economic coercion (S1) |
|---|---|---|---|
| NATO off (scenario default) | 62.6% | 48.0% | 31.8% |
| Coalition model (v0.1.2.8) | +4.1 (3.3 to 4.9) | +8.0 (6.9 to 9.0) | +8.2 (7.1 to 9.3) |
| Member model | +4.8 (4.0 to 5.6) | +6.8 (5.8 to 7.9) | −0.6 (−1.2 to 0.0) |
Changes relative to the member model:
| Variant | Full annexation | Alberta detachment | Economic coercion |
|---|---|---|---|
| 5,000 allied tripwire troops in Canada | +1.1 | +7.5 | – |
| Three months' warning (allies send token contingents) | +0.4 | +2.0 | – |
| US uses its European bases against Canada | +0.7 | +4.3 | – |
| No US pressure on Europe | +0.1 | +0.7 | +1.2 |
| Russia moves in Europe | −0.6 | −1.2 | 0.0 |
| Reform UK government | −0.4 | −1.0 | 0.0 |
| RN presidency in France | −0.6 | −1.3 | −0.6 |
| Both | −1.3 | −2.2 | −0.6 |
| All allies at the low end of their bands | −1.6 | −3.8 | −1.2 |
| All allies at the high end | +1.7 | +3.0 | +1.5 |
| EU itself targeted by the US | – | – | +0.1 |
What the allies do (means per run):
| Measure | S3 | S2 | S1 |
|---|---|---|---|
| Allies condemning the US (of 30) | 27.4 | 24.9 | 14.2 |
| Allies sanctioning or denying basing | 15.2 | 9.1 | 1.3 |
| Allies sending materiel | 4.5 | 2.8 | 0 (no war) |
| Allies deploying forces | 0.6 | 0.5 | 0.2 |
| Share of US troops in Europe hosted by sanctioning allies | 72% | 43% | 6% |
| North Atlantic access points denied (of 4) | 2.7 | 1.9 | 0.2 |
| Chance of EU trade measures | 69% | 39% | 11% |
| Chance at least one ally deploys | 32% | 31% | 15% |
| Chance any ally fights US forces | 8% | 8% | 0% |
Findings
- Under economic coercion the allies are worth nothing to Canada. The coalition model gave +8 points; the member model gives none. On average one ally sanctions or denies basing, and EU trade measures arrive in 11% of runs. The report's reason: the EU has no instrument to sanction the US on a non-member's behalf. This is the biggest correction in this version: with every module on, the chance the US gives up under coercion falls from about 40% to about 31%.
- In a shooting war the two models agree (+4.8 vs +4.1 in full annexation; +6.8 vs +8.0 in Alberta). Most of the value is basing denial and sanctions, not weapons: about 15 allies sanction or deny basing, 4–5 send materiel, and under one deploys.
- A tripwire is the strongest lever Canada and its allies control. 5,000 allied troops in Canada before a limited seizure add 7.5 points; three months' warning with token contingents adds 2. Killing allied soldiers moves the core bloc from words to basing denial.
- The US using its European bases against Canada costs it 4 points in the Alberta scenario, because basing denial becomes near-certain in the core and second tier.
- US pressure on Europe buys little (0.1–1.2 points). A Russian move, or far-right governments in both London and Paris, each cost Canada about 1–2 points.
- Uncertainty about how willing the allies are matters less than the tripwire. Moving every ally from the bottom to the top of its band is worth 3–7 points.
Where the model and the report disagree
- Deployment odds. The report expects a 60–75% chance that at least one ally deploys in a limited seizure. The model gives 31% (39% without US pressure; 62% at the top of the bands). The report's per-ally probabilities and its correlations cannot both hold with that aggregate: correlated draws make "at least one" less likely, not more. I kept the per-ally numbers.
- Combat odds. The chance that any ally fights (8%) is at the top of the report's 3–8%.
Weakest inputs
Every tier probability; common_corr; the mapping from coalition weight to model strength (sanctions_reference, materiel_saturation_share); GDP weights. Missing from the report: exports to the US as a share of GDP, US LNG dependence, Treasury holdings.
Source
Deep-research report, 12 November 2026 (in this project's conversation), sections 3.1, 4, 7 and 8, which cites SIPRI (March 2026), DMDC via Reuters (April 2026), the Kiel Ukraine Support Tracker (February 2026), ECFR and Pew (2026), and AP, CNN, CNBC and Reuters-sourced reporting on Greenland and the Iran war.