UNITIQX / PUBLIC PROOF

Methodology

Evidence first.

What's currently scored

Deterministic public-page tests: local HTTP, latency, sampled links, static HTML accessibility markers, presence of SEO metadata, payload size, limited exposure probes, and sequential real Chromium browser render checks. Each predicate can FAIL. A provisional weighted sum is not a full quality certification.

Browser tests were run at 390px and 1280px, with screenshots saved privately. These tests do not certify complete keyboard gameplay, screen-reader quality, WCAG conformance or broad performance.

How scores work

Each tested component earns only its predefined weight if it passes. A failed check earns zero. An untested check also earns zero points but remains explicitly labeled UNTESTED, not FAIL. Coverage is reported separately. Any leak detected by bounded negative probes caps the provisional score. Overall two-bench composite scores are not produced until independently calibrated graders run.

Outside-provider scores, local-model subjectivity, calibration weights, disagreement and earned badges are not yet available.

Seven-criterion shared judging rubric (planned)

  1. Design — separate independent judgment is UNTESTED; no model-produced score is presented.
  2. Clarity — separate independent judgment is UNTESTED; no model-produced score is presented.
  3. Usability — separate independent judgment is UNTESTED; no model-produced score is presented.
  4. Polish — separate independent judgment is UNTESTED; no model-produced score is presented.
  5. Trust — separate independent judgment is UNTESTED; no model-produced score is presented.
  6. Originality — separate independent judgment is UNTESTED; no model-produced score is presented.
  7. Depth — separate independent judgment is UNTESTED; no model-produced score is presented.

Future judges receive only public URLs and a blind standardized format. No private system details, source files, credentials, user prompts or other judges' scores.

Integrity and correction

Each stored verdict links to a previous SHA-256 digest; chain verification catches accidental and unsanctioned row changes under ordinary storage access. A local administrator with root access could still replace the database and its chain, so this is not external attestation or tamper-proof hardware.

Automatic code edits are disabled pending isolated snapshots, scoped changes, positive and negative regression checks and tested rollback.

Recovery

PostgreSQL scratch-restore status: PASS. Scope is the Living City database from an available backup, including event-chain verification if shown. Not a full platform recovery. Read recovery receipt.