HackYeah 2026 ReviewCouncil score
Median of 3 models, weighted by the task's official criteriaA genuinely built, honestly labeled accessibility passport with computed profile-matched verdicts and real open-data pipelines, limited mainly by a sample-data demo with mocked AI on the live instance, center-city-only coverage and a deck committed to the repo but not uploaded with the submission form.
The generic 'accessible' label is replaced by measured, profile-matched verdicts backed by a source-weighted, time-decaying trust model with conflict resolution and moderator arbitration. Accessibility maps and passports are not new, which keeps the median at 8 rather than higher.
Squarely a resident-facing city tool with named beneficiary roles, real open-data ingest and a live deployment. All members agree on the fit, and the highest median in the table reflects it.
The loops are concrete and working: set a profile once, get a four-state verdict, audit a venue in five steps. Held back because the running instance shows labeled sample data only, so real-world coverage and impact are undemonstrated.
Clean, consistent, accessibility-first UI where WCAG 2.2 AA is actually built in: high contrast, text-size controls, large targets, a hyperlegible font and axe checks in E2E. Some screens still show placeholder or demo content.
For the event window the build is exceptional: 11.7k LOC, 266 tests, Playwright plus axe E2E, real OSM and GTFS import code, dual backends, magic-link auth and an admin panel, all deployed. Member C's 7 rests on the demo-mode mock, but the integration code itself exists, so the council weighs that limit under applicability and keeps the median at 9.
Members disagree here: scores range by 2.0 points.Strengths
- Verdicts are computed, not displayed: measured door widths, steps, thresholds and toilet dimensions are compared against the user's own chair profile with explicit clearance constants
- Real open-data pipelines in code: Overpass OSM import with mirror fallback, retries and tiled queries, plus ZTP Kraków GTFS tram-stop import
- Unusually mature trust model: per-source weights, time decay, conflict detection, one-voice-per-person dedup and moderator arbitration with undo
- Quality discipline and honesty: 266 tests, Playwright with axe checks, a deployed live app, clearly labeled sample data and disclosed AI use with human sign-off
Weaknesses
- The live deployment runs the in-memory seeded backend with a handful of demo venues and a mock AI extractor, so the real OSM, GTFS and AI integrations exist in code but are not demonstrated on the running instance
- The presentation deck was not uploaded with the submission form and exists only as a PDF committed to the repo, so a required deliverable was missed
- The OSM ingest bbox covers only Kraków center; city-wide coverage and the promised new cities are asserted, not demonstrated
- Value depends on cold-start adoption by venue owners and community reporters, and the free-tier and B2G plans are on paper only
claude:glm-5.3-flash84.0100% agreedots-studio/dots-3-note-preview:free83.090% agreeinclusionai/ling-3.0-flash-sante:free74.570% agreeclaude:glm-5.3–judgeThe council, the evidence pack, the prompts and the queue are all on GitHub. If a review helped you, a star helps other teams find it.