About · Governance
An index is only as good as its refusals
MICA is built to be argued with. What these rules protect is an open public benchmark initiative for consumer execution, not a vendor marketing surface. This page records who decides, what they are not allowed to decide, and how a result gets corrected when it is wrong.
Preview 0.1
6 markets · 10 task families · 3 separate outcome axes.
Standing rules
What MICA will not do
These are commitments, not preferences. Breaking one is a governance failure, not a product decision.
- MICA will not publish a score it cannot re-derive from the raw record. A task score is the product of its accuracy, speed and cost components, and all three raw axes stay published beside it.
- MICA will not accept a score, a weighting or a self-declared component from a submitter in place of a measurement.
- MICA will not accept payment for placement, for inclusion, or for the timing of a result.
- MICA will not publish a figure it cannot trace to a run record it holds.
- MICA will not treat missing coverage as a zero, or a fast failure as a fast success.
- MICA will not remove a critical safety event from a cell. The block is permanent.
- MICA will not run irreversible actions against real accounts belonging to real people.
Decisions
Who decides what
Separating editorial authority from operator relationships is a requirement MICA holds itself to, and the separation below is stated so it can be checked rather than assumed.
- Method changes
- Changes to definitions, gates or axes are versioned with the edition and published before the results computed under them. A method is never changed after seeing which system it would favour.
- Edition sign-off
- Each market edition is signed by its edition note. The note records scope choices and known gaps in that market.
- Operator relationships
- Submitters may correct facts about their own snapshot composition. They have no say over verification status, publication eligibility or interpretation.
- Conflicts of interest
- Any commercial relationship with an operator is disclosed on the affected results, and disqualifies MICA staff involved in it from verification of that system.
Independence
What is not settled yet
MICA is intended to be neutral, and it is not yet structurally independent. Stating the gap is more useful than claiming it is closed, so the open requirements are listed here and will be updated only when they are actually met.
- Initiator and challenger
- MICA was initiated by vooy, which also expects to be measured as a consumer-agent system. An initiator that may become a subject of the benchmark cannot also be its scoring authority; separating those roles is an open requirement, not a completed one.
- Decision control
- Method changes, verification outcomes and publication decisions must rest with a body that no evaluated party controls. That body has not been constituted, so no claim of independent decision control should be read into this site today.
- Funding and relationships
- Who funds MICA, on what terms, and what commercial relationships exist with any evaluated party must be disclosed publicly before results are published. No funding or operator relationship has been documented here yet.
- Public and holdout sets
- Ownership of the public task set and the holdout set must be separable, so that access to one confers no advantage on the other. The catalogue records this distinction per task; the holdout set is never published.
- Status
- Unresolved. These requirements gate the first official edition: MICA should not publish verified results until they are met and documented.
Corrections
How an error is fixed
MICA expects to be wrong sometimes. The obligation is to be wrong in public and briefly.
A correction request should identify the cell, the figure disputed, and the evidence. MICA re-derives the figure from the run records it holds; if the records are wrong, the cell is withdrawn rather than adjusted.
Corrections are published with the original value, the corrected value, the date, and the reason. Silent edits to a published figure are treated as a governance failure.
A withdrawn cell stays visible as withdrawn. Removing it entirely would let the index quietly forget its own mistakes.
Claims
Three kinds of statement
Every sentence MICA publishes is one of these, and is labelled as such.
- Measured
- A value read directly from run records. No inference.
- MICA interpretation
- MICA's reading of what the measured values mean. Contestable, and signed by the edition.
- Recommendation
- Advice for builders or buyers. Never a claim of measurement.
Provenance
Where this build's data comes from
Stated on the site itself so that a reader never has to trust a claim about the pipeline.
- Source kind
- Local illustrative records included with this preview
- Data status
- Illustrative only · no verified results
- Edition
- Preview 0.1 · methodology draft
- Disclosure
- Preview data for interface development. No verified results are published.
- Remote source
- No remote database is installed. The data-source boundary exists so a real edition can be served without any page importing a fixture directly, and a build with no credentials configured is the supported default.