charter

Version 1.0 — adopted 2026-08-19. Changes follow the RFC process.

Mission

MirrorQ exists to give the quantum computing community numbers it can trust: benchmark results that are blind (solved on fresh, private instances), reproducible (every score traces to an exact benchmark version and validator build), and governed in the open.

Scope

The service evaluates *solvers* — programs participants run on their own hardware — against *benchmarks* — families of generated task instances scored by a *validator*. The service layer is domain-neutral; every benchmark's domain logic lives in its validator behind the published validator contract. The service does not rank hardware and does not execute participant code.

Who operates it and who governs it

QudeLeap operates, funds, and hosts the service ("the operator"). Benchmarks, rules, and this charter are governed by the community through the RFC process and the Advisory Board. The operator holds one board seat and never the chair. Operator obligations are listed in OPERATOR_OBLIGATIONS.md.

What qualifies as a benchmark

A benchmark is accepted only if its instances are:

  1. Hard to precompute — there is no catalogue of all instances a participant could solve ahead of time;
  2. Cheap to verify — the validator scores a submission in far less time than solving it;
  3. Fresh per submission — every registration draws a new private seed; the seed and the answer key never leave the server;
  4. Deterministic given the seed — so a score can be replayed against the pinned validator build;
  5. Declared — metrics, direction of ranking, and task display fields are declared in the validator manifest, not inferred by the service.

These are the blind-instance principles checked by the conformance kit and by the design-validator review. An RFC proposing a benchmark must include a gaming analysis: how a participant could score well without solving the task, and why the design prevents it.

Benchmark lifecycle

Every benchmark carries the version of the validator that defines it and moves through four states:

StateMeaningWho moves it
proposedRFC open; no public leaderboardproposer
rcRelease candidate: public practice leaderboard, results not archivedboard decision on the RFC
stableResults count, snapshots archive them, version is immutableboard decision after a practice period
deprecatedNo new submissions; history stays queryable and citableRFC + board decision

A stable benchmark version is immutable: any change to its tasks, generation, or scoring is a new version, recorded with a new changelog entry, and the service refuses to load a version whose definition has changed.

How decisions are made

Archive snapshots

An archive snapshot is a frozen export of all published leaderboard entries for every stable benchmark version, together with benchmark definitions, provenance digests, baselines, and a hash manifest. The operator produces one on the first day of each quarter, starting with the first quarter after M2, under the rules in RFC-0001. The board signs off each snapshot before it is published under CC BY 4.0 with a permanent identifier. A published snapshot is a permanent scientific record; live leaderboards remain open and continue to change between snapshot dates.

Disputes

Any participant may dispute a score, a rule application, or an exclusion by opening an issue labelled dispute within 14 days of the event. The operator responds with the facts (provenance, logs) within 7 days; if the participant is not satisfied, the board decides, with any conflicted member recused. Decisions are published in the issue.

Conflicts of interest

See CONFLICT_OF_INTEREST.md. In short: the operator's own solvers may participate, flagged [operator], under identical rules; board members recuse themselves from decisions about benchmarks or submissions they have a stake in.

Amendments

This charter changes only by RFC. The current version and its history are in the repository.