What DICA does not yet do well.
A method that publishes only its strengths is a marketing document, and a jurisdiction is right to discount it. This register is maintained on the same change-control footing as the standard itself: every known weakness is stated here, with what it means for a determination and what would close it. Anyone may cite this page against us.
The parameters are hypotheses, not calibration.
Part B values are reasoned starting points. They have not been fitted against a body of realised outcomes, because the asset class is young and the loss history is thin. A determination therefore reproduces the function exactly and the parameters only provisionally.
Attribution of shared infrastructure remains contestable.
Separating project-caused cost from background growth is the single most argued step, and the standard constrains the method without eliminating judgement. Two competent assessors can differ materially on the same road or the same feeder.
Evidence class is coarse.
Five classes cannot express the difference between a thorough independent study and a thin one bearing the same letterhead. The standard reserves class assignment to the accredited assessor precisely because the scale cannot carry that judgement on its own.
No accreditation body exists yet.
The standard refers throughout to an accredited assessor. Accreditation is currently self-asserted against the conduct rules in Annex B, which is weaker than it needs to be and is a legitimate objection to raise.
Digest binding proves identity, not truth.
Verification establishes that the document relied upon is the document reviewed. It says nothing about whether the document is correct. A confidently wrong engineering study binds and verifies exactly as well as a correct one.
The scenario floor is a judgement, not a model.
The systemic floor exists because bucket-wise sizing understates correlated failure. Its level is set by reasoning about plausible joint scenarios, not derived from a correlation structure that anyone has estimated.
Automated reading assists; it does not determine.
Where machine reading is used to locate and propose findings, it can miss material text and can propose findings that no cited document supports. Recall is not guaranteed by any configuration.
Instrument credit ignores issuer correlation.
Credit is assigned by instrument form and issuer rating, treating each independently. Where a single surety or guarantor stands behind several obligations, or behind several projects in one jurisdiction, the effective protection is less than the sum credited.
Out of scope by design.
DICA does not assess whether a project should be approved, does not value community benefit, does not opine on rate design, and is not an underwriting substitute. A determination sized under this standard tells a jurisdiction what to hold, not what to decide.
An assurance number is only worth what the process behind it can survive. Publishing the register is the cheapest available proof that the process is not being defended past its evidence — and it is the page a hostile expert would otherwise write for us.
- Band cliffs at coverage thresholdsclosed
- Double-counting of protectionclosed
- Discretionary “equivalent evidence”removed
Anyone — applicant, jurisdiction, reviewer, opposing expert — may submit an entry. Submissions and their disposition are published, including those rejected and the reason. An entry may not be closed by the person who raised it.
Every determination cites the register version current at its date. If an open entry bears on a material input, the determination says so on its face rather than leaving the reader to find it here.
The standard is falsifiable, and that is the point.
Each entry above names what would close it. That is a commitment against which DICA can be measured — and, over time, the honest way to tell whether the method is improving or merely being defended.