Every framework in the index is coded on the same twelve features, four to an axis, weighted equally. Each runs from 0 to 100 against published anchors, so any reader can check a score against the rule that produced it — and disagree with it precisely.
Within every axis, 100 means more of the thing the axis is named after. A framework scoring 90 on power concentration concentrates a great deal of authority; one scoring 90 on reversibility is easy to amend, exit or let lapse.
Two features on the reversibility axis are named for the property that reduces reversibility. Their raw scores are recorded as named, then inverted before entering the axis mean, so that all four features count in the same direction.
2.2 Amendment difficulty and 2.4 Path dependency enter the reversibility mean as 100 minus their raw score. A framework that is hard to amend and strongly self-reinforcing is less reversible, and the arithmetic has to say so. Raw scores are shown on the scores page alongside the value that actually enters the mean.
A high score is not a criticism and a low score is not praise. Concentrated authority buys enforceability at the cost of flexibility; loose coordination buys adaptability at the cost of bite. The index makes those trade-offs legible; it does not adjudicate them.
An axis score is the unweighted mean of its four features, rounded to the nearest whole number, with the two reverse-coded features inverted first. All four features count equally. This is a deliberate simplification: we have no defensible basis for weighting one feature above another, and equal weighting is the assumption a reader is most likely to make anyway.
The anchors are reference points, not the only permitted answers. Where a framework sits clearly between two anchors, we assign an intermediate value in steps of 5 and record why. Intermediate values are common — international frameworks rarely sit exactly on an anchor.
Each coded cell has a one-line justification and a citation to a primary source. This follows from the standard we hold ourselves to: Lincoln and Guba establish dependability and confirmability through exactly such an audit trail. A score without a recorded justification would fail our own criterion.
Anchors alone do not make a coding reproducible. These four rules decide what is being scored before anyone asks how well it scores, and they bind every framework in the sample.
Code what the text provides on its face. What happened afterwards — enforcement records, lobbying, political resistance, actual compliance, real-world adoption — is out of scope. Those measure the environment a framework landed in rather than the framework's design, and they introduce more variables than an index of this size can carry. Where the text creates a power, the power counts even if it has never been used.
The object of coding is the instrument at adoption, not as subsequently amended. Later amendments are recorded and discussed, but they do not move a score. This keeps the sample comparable: several frameworks here have never been amended, and coding one against a consolidated later text while the others stand at adoption would compare different things.
The index is a snapshot of design choices at the moment of commitment. It does not claim to describe what any framework looks like today. The EU AI Act, for one, was substantively amended in July 2026; that amendment is recorded on the scores page and excluded from the scores.
Pick the instrument that carries the operative obligations — the one that will end up being applied — name it, and code that at face value. The others are context, not inputs to the score.
Features 1.3 and 1.4 ask how unequally material and technical capability are distributed among the parties to the framework — not how much leverage the framework's sponsor holds over outsiders. This follows the middle-power and great-power bargaining literature the two features are drawn from, and it keeps the axis coherent: all four features then answer the same question, which is how power is distributed inside the arrangement.
The index cannot record the Brussels effect — the capacity to set global rules through market access rather than through membership. That omission sits alongside the absence of Lukes' third face, and both concern power exercised over those outside the room. They are the strongest candidates for what a version 3 should add.
How much authority the framework gathers into a single locus. 100 = authority maximally concentrated. The first two features follow the faces of power: who gets to vote, and who decides what is discussed in the first place. The second two follow the material and technical resources on which parties bargain.
Who gets to vote, and can the outcome bind those who voted against it?
Who decides what reaches the table in the first place?
Does material weight — economy, population, market access — convert into formal influence?
Who holds the know-how that determines what compliance actually means?
How easily a commitment can be undone — left, amended, allowed to lapse, or escaped from. 100 = maximally reversible. The axis follows constitutional design literature, in which institutional flexibility is preferred to irreversible commitment.
Can a member state leave, and at what cost?
How many veto players must agree before a core rule can be changed?
A high score here means less reversibility, so the raw score is inverted before it enters the axis mean: it contributes 100 − raw. Both values are shown on the scores page.
Do commitments face compulsory review, or expire if not renewed?
Is the framework self-reinforcing — does staying in get cheaper, and leaving dearer, over time?
A high score here means less reversibility, so the raw score is inverted before it enters the axis mean: it contributes 100 − raw.
How well the framework turns decisions into implementation. 100 = maximally efficient. Efficiency is hard to sustain in global frameworks, which is precisely why it tends to be bought by concentrating power rather than by distributing deliberation.
How long do commitment procedures take, from proposal to entry into force?
Interpolate at 25 for three to five years. The clock stops at entry into force; the gap between force and obligations actually applying is captured under 3.2, so the two are not double-counted. Per rule 1, delays caused by lobbying or political resistance are not modelled.
Do members actually have to apply it, uniformly, on the date it says?
Does it duplicate or contradict law the parties already have?
Is compliance monitored over the long term, and does anything follow from failing?
What the index cannot do — the scope of its claims, the absence of inter-coder testing, the dimensions of power it does not reach, and the version history — is set out in the methodology notes.