Final results
Two tracks, judged and ranked separately. Each project carries the judges' marks out of 80, plus up to 10 for Facebook reactions and 10 for GitHub stars, for a total out of 100.
Spirit of July
Marked independently by 2 judges, and averaged. Engagement scaled against the best in this track: 181 reactions and 62 stars both score 10.
| # | Project | Judge 1 /80 | Judge 2 /80 | Average /80 | Facebook /10 | GitHub /10 | Total /100 |
|---|---|---|---|---|---|---|---|
| 1st | 唳膏Δ唰嵿ΟQR (ShottoQR)TellTheTruth 路 Dhaka | 43.0 | 67.0 | 55 | 10181 reactions | 9.1957 stars | 74.19 |
| 2nd | GovwatchTeam Buraq 路 Barishal | 49.0 | 67.5 | 58.25 | 4.8688 reactions | 7.5847 stars | 70.69 |
| 3rd | Protocol36te@rGas 路 Dhaka | 59.5 | 45.0 | 52.25 | 8.40152 reactions | 1062 stars | 70.65 |
| 4 | JonomotTeam dropOuts 路 Dhaka | 56.3 | 47.0 | 51.63 | 4.6484 reactions | 5.9737 stars | 62.23 |
| 5 | 36 Days Through Their EyesByte_force 路 Dhaka | 34.5 | 63.5 | 49 | 5.2595 reactions | 4.8430 stars | 59.09 |
| 6 | Echoes of JulyBear Minimum 路 Dhaka | 37.5 | 59.0 | 48.25 | 1.6029 reactions | 4.6829 stars | 54.53 |
| 7 | IdentityFaHa 路 Moulvibazar | 54.5 | 36.5 | 45.50 | 0.7714 reactions | 0.322 stars | 46.60 |
| 8 | March to GanabhabanBhashanir_Soinik 路 Tangail | 26.0 | 55.0 | 40.50 | 1.8233 reactions | 4.1926 stars | 46.52 |
Judges' comments
唳膏Δ唰嵿ΟQR (ShottoQR)
Judge 2: Seven forensic signals with an explainable score rather than a verdict, Ed25519 seal plus SHA-256 chain, and a QR that verifies offline with no server. Court-ready PDF with custody trail. Exactly the tool the theme calls for, built to be checkable by a sceptic.
Govwatch
Judge 2: Real corpus, real infrastructure: 50,000+ contracts, Bangla-aware FTS5 fused with vector search by reciprocal rank, z-score outliers surfaced as red flags, every answer cited, no login wall. Live on its own domain. The clearest civic-accountability entry.
Protocol36
Judge 2: Right problem, victim records that cannot be quietly edited. But the solution text is a feature list, blockchain and IPFS are asserted rather than justified against a signed append-only log, and the how is missing.
Jonomot
Judge 2: Rating government offices and publishing what documents you actually need is quietly useful and well aimed at ordinary friction. Conventional build; moderation will decide whether it survives contact with reality.
36 Days Through Their Eyes
Judge 2: 26 events each cross-checked against two independent sources and cited inline; every branch converges on the documented outcome; a Historical Reality panel stops fiction replacing record; real victims named and distinguished from composites. The most rigorous handling of contested history here.
Echoes of July
Judge 2: Holds contradictory accounts open, birdshot versus rubber bullets versus gunshot, instead of resolving them, and keeps provenance separate from caption. That editorial discipline is rarer than the engineering.
Identity
Judge 2: Ties directly to the quota grievance that started the uprising, which is a strong framing. But extracting a biometric profile from a photograph of a fingerprint via a general-purpose vision model is not sound biometrics, and the system's entire claim rests on it.
March to Ganabhaban
Judge 2: Deliberately built on raylib and WebAssembly so it runs on low-end hardware, the constraint is the point, and most teams ignore it. Depicting named real individuals as game antagonists is an editorial decision organisers should look at, not a scoring matter.
How this was marked
Each judge scored every finalist out of 80 across the five criteria in Section 9 of the guideline: impact and relevance to track (25), technical execution (20), innovation and originality (15), feasibility and resilience (10), and presentation and usability (10). The judges' column is the mean of their marks, not a sum.
The remaining 20 points are public engagement, scored exactly as Section 8 described before the sprint began: proportional to the best in the track. The highest Facebook reaction count in a track scores the full 10, and every other project scores its count 梅 the highest count 脳 10. GitHub stars are scaled the same way against the highest star count in that track. Counts were frozen on 1 August at 23:59.
Rank-based and logarithmic scaling would both narrow the distance between a widely shared post and a quiet one, and there is a fair argument that either measures engagement better than raw proportion does. Neither is used here, because the proportional rule is the one teams could read before they built anything. Changing the formula after the counts were frozen would move points on a rule nobody had seen.
A component that was never counted is shown as uncounted and scores nothing, rather than being averaged away as if it were a zero somebody measured.