How a turn actually runs.
Eight AI specialists, an orchestrator forbidden from predicting, and one contrarian whose entire job is to argue. None of them sees the whole board. Their disagreement is part of the output.
The framework runs as a federation: eight domain specialists who aren't allowed to talk to each other directly; one orchestrator who settles disputes between them but isn't allowed to predict outcomes; and one contrarian, codenamed CASS, who fights the consensus every turn. The point of all that separation is that any single analyst, working alone, would pattern-match too fast. The federation is built to make pattern-matching expensive.
The rules below are the method, not the scenario. The same machinery would apply if you pointed the framework at a fantasy realm, a market, or an entirely invented polity — the domains might change and the personas might change, but the discipline doesn't. On this page: how one turn is staged step by step, the eight specialists in full with their declared blind spots, and the contrarian.
From here down: the protocol, in costume
The reading copy stamps this section PROTOCOL. The procedure is the same whether the run is dressed up as a declassified file or not — the costume is just easier to read.
What happens during a turn
- The orchestrator pulls the current state of the world.
- It picks how much time the turn should cover. Hours during a missile crisis. Months during a slow political standoff. The pace follows the action. (Turn 10 covered just two months, because the contrarian convinced the orchestrator at turn 9 that the actual cliff event was five weeks past the standard interval.)
- It decides which specialists need to run. Not all eight every turn — only the ones whose domain is in play.
- Each specialist gets only their own slice of the world, plus whatever they need from other specialists' outputs. Nobody sees the whole picture. This is the whole point.
- Each specialist returns a range of outcomes, not a single answer — plus the uncertainties they couldn't resolve and the assumptions they had to make.
- The orchestrator works out what happens at the seams — when a missile strike affects the power grid, or a cyber attack affects logistics, or an information operation shifts political will. Specialists aren't allowed to settle their own disputes.
- CASS runs in parallel against the orchestrator's draft. Her objections go into the official record alongside the consensus, not as an afterthought.
- The federation writes down what it thinks adversaries will do next turn, with how likely each action is and how confidently it could be attributed kept as separate numbers.
- Four things get produced at turn close: an updated world state, a delta file (what changed and why), an analyst-facing briefing, and an archive snapshot so any earlier turn is recoverable.
The roster · Eight domain specialists
Each one is codenamed after the analytic tradition their domain draws on. Each one is required to declare their own known biases when they file an estimate. The orchestrator watches the roster as a whole for the federation failure mode: when several specialists share the same blind spot, the federation can do worse than a single skeptical analyst would.
Friction. Fog. The gap between what an order says and what the engagement returns. Tracks missiles, aircraft losses, troop movements, ship attrition, and the depth of every magazine on every side. Spent the first three turns watching Israeli interceptor stocks collapse in slow motion. Reads every clean operational plan as something the other side will probably break.
Known blind spot: tends to under-estimate how much risk the other side will actually accept.
Counts days-of-supply before he counts anything else. Stockpiles, production rates, mobilization, what's actually getting through the chokepoints. When KARL says "we can sustain this," OMAR is the one who checks the spreadsheet. The Strategic Petroleum Reserve hitting effective zero around turn 7 was, in his framing, the actual headline.
Known blind spot: defense-industry models tend to overstate how long production can keep up.
Reads every coupled system as one tight binding away from cascade. Energy grids, transport, telecoms, payments, settlement. Brought the cascade language to the federation in turn zero and never let it leave. Was the first one to notice that the Compact's parallel settlement system was changing the financial topology in a way the old banking-crisis classification couldn't see.
Known blind spot: finance-sourced models assume crises stay contained when historically they don't.
Signal versus noise. What you know versus what you think you know. Refuses to file an estimate without spelling out how the federation could be wrong about what it's looking at. Tracks attribution separately from likelihood — by turn 8 the deepfake event volume had made attribution confidence on cyber operations effectively unreliable, and she flagged it.
Known blind spot: Western intelligence usually thinks it sees the picture more clearly than it does.
Treats the information environment as its own battlespace. Deepfakes, narrative collapse, cyber attacks on physical infrastructure, election-window targeting. He and KARL have been arguing about whether information operations count as kinetic since turn one. By turn 10 the answer was "it doesn't matter, the same actors are doing both."
Known blind spot: vendor-funded threat models overstate whatever the vendor sells against.
Tracks how alliances hold and how coalitions defect — in the same model. Skeptical of any equilibrium that hasn't been broken yet. Carried most of the weight on "will the coalition of the compromised survive another impeachment?" His honest answer at turn 7 was: probably yes, but only because the people with the most to lose carry the rest. That answer remained accurate through turn 10.
Known blind spot: people inside the system tend to assume the system is more durable than it is.
Civil-military relations. The speed at which decisions actually move through a chain of command. Reported on the president's decision-making with the bluntness of someone who's seen chain-of-command fail before. By turn 9 his decision-cycle-integrity metric was at 0.02, which he insisted meant the variable had stopped measuring anything useful. Turn 10 replaced it with "faction routing mode."
Known blind spot: uses peacetime data to model situations that don't look anything like peacetime.
The slow integrity of institutions. Scenario-scoped — she doesn't run every turn. She only speaks when an institution is being tested by a stress event that forces a believed-vs-actual reading. The rest of the time the federation has to make assumptions about her that may not survive the next test. Her readings on US institutional resilience were consistently higher than CASS thought they should be.
Known blind spot: it's hard to see how fragile your own institutions are while you're inside them.
The roster · The contrarian
Argues. Every turn, after the orchestrator drafts what just happened, CASS gets the draft plus the federation's known blind spots, plus a plain statement of which way the consensus is leaning — and pushes back from whatever angle the consensus is missing. Across eleven turns she got most of the big framing calls right (foreign restraint was actually patient pre-positioning; "contained" was the wrong word for the banking crisis; the Compact was past the point of no return long before the federation admitted it; turn 9 didn't close on resolution and turn 10 should be the dual-government turn). She also got specific predictions wrong (she predicted loyalist hardening would slow cabinet defections at turn 10; defections actually accelerated). At turn 10 she explicitly acknowledged her own analog-selection bias for the first time: every transition she'd cited was a transition that had ruptured.
Keep reading
The contrarian, in full
Why she runs every turn, what triggers a mandatory invocation, the seven arguments she carried across the series, and what she got wrong.
Why every number is a range
Confidence intervals, three kinds of doubt tracked separately, the modal-hold review (the protocol item that caught the federation when it got attached to a frame), and the failure modes the federation discovered about itself.