Data note
Capacity and Capability in the Draghi Implementation Record
Snapshot. Statuses as of the July 2026 update of the EPIC Draghi Implementation Index. Coded September 2026.
This is a dated snapshot, not a maintained series. It will go out of date when EPIC publishes again.
Download the coded dataset (CSV, 383 measures) · coding.py · analyze.py · robust.py
The EPIC Draghi Implementation Index reports uneven progress across the ten sectors of the Draghi report on European competitiveness: 38.6% of critical raw materials measures fully implemented against 6.4% of digital and advanced technology measures, and 1.2% of energy measures. One reading of that spread is that Europe moves fastest where the object of a recommendation can be bought, contracted, stockpiled or funded, and slowest where it must be accumulated over years. This note tests that reading against the index at measure level. It does not hold. Capacity measures and capability measures are implemented at statistically indistinguishable rates. What the record does discriminate against is a third category the capacity/capability pair does not contain: measures whose object is a transfer of authority. The note reports the coding, the null result, the sensitivity checks, and what the null does and does not license anyone to conclude.
I.
What this is
A single-coder classification of all 383 measures in the EPIC Draghi Implementation Index, together with the tests it was built to support.
It is published because the hypothesis failed. A failed hypothesis tested against a public dataset belongs on the record in the same form as a successful one, and the coded file is attached so that anyone who disagrees with a call can change it and re-run.
II.
The coding rule
Each measure was coded on its object — what the measure would actually secure — rather than on its difficulty, its political salience, or the legislative instrument named to deliver it.
Capacity. The thing the measure secures already exists somewhere and can be obtained by transaction: purchased, contracted, stockpiled, funded, permitted, taxed, tariffed, mandated, monitored or legislated. Delivery is an act of acquisition, allocation or authorization inside a budget cycle.
Capability. The thing the measure secures must be accumulated: know-how, skills base, engineering experience, research depth, industrial ecosystem, firm-level technological mastery. No transaction produces it. Repeated practice over years does.
A third code proved unavoidable during the pass.
Institutional. The object is neither purchasable nor accumulated technical know-how. It is a transfer of authority, a change in market structure, or the surrender of a national prerogative. Central regulatory oversight of the internal energy market, Single Market passporting for cloud services, the removal of national operational rules in rail, and reform of ESA governance all fall here. The category was separated out precisely so that the capacity/capability comparison would not silently absorb “politically hard.”
III.
What was coded
The 383 measures of the fixed Draghi universe, with implementation statuses as published in the July 2026 update: 60 implemented (15.7%), 98 partially implemented (25.6%), 160 in progress, 65 not implemented, giving 158 of 383 (41.3%) implemented or partially implemented. The coding reproduces EPIC’s sector and status counts exactly.
| Code | Measures |
|---|---|
| Capacity | 274 |
| Capability | 48 |
| Institutional | 61 |
IV.
Results
The capacity/capability split does not predict implementation.
| Code | n | Implemented | Implemented + partial |
|---|---|---|---|
| Capacity | 274 | 18.2% | 45.6% |
| Capability | 48 | 16.7% | 41.7% |
| Institutional | 61 | 3.3% | 21.3% |
Capacity against capability on strict implementation gives a Fisher exact p of 1.00. On the combined index, p = 0.64. There is no difference present to control for.
Sector size is not the confounder. The correlation between a sector’s measure count and its strict implementation rate is r = +0.08 (p = 0.84). Defence’s fourteen measures and energy’s eighty-three do not account for the spread between them.
The institutional category is where the record discriminates. Institutional measures run at 3.3% strict against 18.2% for capacity. In a logistic model with sector fixed effects the gap is significant on the combined index (odds ratio 0.38, p = 0.009) and marginal on the strict index (p = 0.086).
Sector composition explains much of the published sector table. Energy is 30.1% institutional; critical raw materials is 3.5%. That difference, rather than any capacity/capability contrast, accounts for most of the distance between a 1.2% strict rate and a 38.6% one.
The digital sector rules out the capability reading directly. Within digital and advanced technologies, capacity measures run at 4.2% strict and capability measures at 7.1%. The tractable measures are as stuck as the accumulative ones. Semiconductors are not slow because the object must be accumulated. Everything in that sector is slow, including the money.
A weaker version survives partway. Merging capability and institutional into a single “cannot be obtained by transaction” category gives 9.2% strict against 18.2% (Fisher p = 0.029). With sector fixed effects that becomes non-significant on the strict index (p = 0.48) and weakly significant on the combined index (odds ratio 1.71, p = 0.043). Since 61 of those 109 measures are institutional, any surviving effect is a statement about transfers of authority, not about capability accumulation.
V.
Sensitivity
The null result is robust to the codings most open to challenge. Recoding the eleven skills measures as capacity leaves it null (p = 0.82). Adding the research-funding measures to capability leaves it null (p = 0.70). Both changes together leave it null (p = 1.00).
It is not robust to one exclusion. Six of the eight capability measures recorded as implemented are cases where the launch of a programme was scored as delivery of a capability objective. Transport measures 362 to 365 are four one-line modal items under a parent recommendation to launch EU innovation projects, all four already implemented at the September 2025 baseline. Measure 123 is “boost European excellence in research and innovation in alternative materials.” Measure 180 sits under the launch of a new EU Semiconductor Strategy. Excluding those six yields 20.0% against 4.8%, p = 0.016.
That number is reported here and not used. The exclusion removes only implemented measures, so it cannot be checked against the data, and any test built on it would be circular.
VI.
What the index records
The reason those six score as implemented is more informative than the exclusion.
EPIC’s methodology is explicit: the framework is the pledge–enactment method adapted from the comparative election-pledge literature, applied through legal-text review, with only adopted EU legal acts counting as implementation. The unit is an adopted act.
That makes the outcome of a recommendation a function of how the recommendation was written. Where a Draghi recommendation is worded as “launch a programme,” it closes when the programme launches. Where it is worded as “establish a Single Market” or “centralize governance,” no single act ever closes it. Capability accumulation has no adopted act corresponding to it at all, which is why it appears in the record only through the programmes created in its name.
The index therefore cannot record whether capability is being built. This is not a defect. It is what a legal-text index is, and EPIC states as much. But it constrains what the index can be used to argue. The null result above is not evidence that Europe is accumulating capability, and a positive result would not have been evidence that it is not. The instrument measures enactment.
VII.
Limits
One coder, one pass, no second coder and no inter-rater reliability check. Roughly twenty of the 383 measures could be argued either way, concentrated where the instrument is money and the object is know-how — research funding, skills formation, and semiconductor support in particular.
Statuses are those of the July 2026 update and will go stale when EPIC publishes again.
Sources
Mike Edholm has spent three decades advising corporations and institutions on scenario planning and strategic foresight, with assignments spanning telecommunications, computing, industrial automation, and international development.