Event workspace
Results
published
Publication revision 1 · Evidence cutoff 2026-09-29 14:34 UTC · Digest 1d5c5abd5d9fc2121ded15d250092e73abe214ddf42a8574bb678ec2bafc6f36. Correction history.
- Method
- additive-judge-effects+shrunk-scale(kappa=2, floor=0.35, ceiling=2.5, rounds=12)
- Converged
- yes
- Grand mean
- 3.800
- Residual SD
- 0.436
- Rankings agree
- Kendall tau 0.600
- Tiers the evidence supports
- 3 - first place is separated
| # | Tier | Project | Track | Ballots | Raw mean | Adjusted | 95% range | Was | Moved |
|---|---|---|---|---|---|---|---|---|---|
| 1 | 1 | Readback | access | 3 | 4.73 | 4.762 | 4.11 – 5.41 | 1 | - |
| 2 | 2 | Lintwright | tooling | 3 | 3.80 | 3.777 | 3.13 – 4.43 | 2 | - |
| 3 | 2 | Portmatic | tooling | 3 | 3.73 | 3.749 | 3.10 – 4.40 | 3 | - |
| 4 | 2 | Highcontrast | access | 2 | 3.40 | 3.688 | 2.96 – 4.41 | 4 | - |
| 5 | 3 | Stackless | tooling | 1 | 2.00 | 2.401 | 1.28 – 3.52 | 5 | - |
How much of this order the evidence supports
Tiers
3 levels across 5 projects.
A tier holds the projects that are not separated from the one at the top of it, so awarding on a tier's leader is defensible where awarding on the printed place is not. That is a statement about the ballots and not about the work: it means the panel did not give enough information to tell those entries apart, and a place chosen inside a tier is a judgement somebody has to make rather than a number this portal produced.
- First vs second
- separated
- Separation
- 2.26
- Strata
- 3.35
- Reliability
- 0.84
- Interval
- 95%
- Ballots per project
- 2.4
Reliability is the share of the spread in these scores that is real rather than noise: run the event again with a different panel of the same size, and this is roughly how much of the ranking would survive. Separation is the same quantity as a ratio, and the tier count comes from it directly by the Wright and Masters strata formula.
Where the spread came from
Between projects - 63.0%
Between judges - 24.2%
Unexplained - 12.8%
The first share is the one that should be largest: it is disagreement about the projects, which is what a judging round is for. The second is disagreement about the judges, and it is the share normalization exists to remove - a large one means the adjusted column above is doing real work rather than decorating the raw one. The third is everything neither explains.
- Between projects
- 0.625
- Between judges
- 0.388
- Residual
- 0.436
- One score's error
- 0.304
The same three quantities as spreads on the rubric's own scale, for anybody checking the arithmetic. They are shares of their own sum rather than an exact partition, which is what any fit on an incomplete design can offer.
What qualifies these numbers
- 1 project has a single ballot, so its standard error is wide and its rank should not be trusted alone: Stackless.
- Unbounded likelihood without the prior: 1 project has no losses and 1 has no wins. Their strengths are held finite by the prior (1) and should be read as provisional.
Judges are named positionally - "Judge B" - and not by account, deliberately: these caveats are published, and an appraisal of a volunteer is not something this portal publishes. The organizer's dashboard names them, because that is where the person who can act on it reads.
Pairwise standing
A separate ranking from head-to-head comparisons, fitted by Bradley-Terry. It is published beside the rubric ranking rather than blended into it: two methods that agree are evidence, and one number that hides a disagreement is not.
| # | Project | Strength | Won | Lost | Comparisons |
|---|---|---|---|---|---|
| 1 | Readback | 2.372 | 3 | 0 | 3 |
| 2 | Highcontrast | 0.549 | 4 | 2 | 6 |
| 3 | Lintwright | 0.525 | 3 | 3 | 6 |
| 4 | Portmatic | -0.929 | 2 | 3 | 5 |
| 5 | Stackless | -2.517 | 0 | 4 | 4 |
Which of these head-to-head places are real, and which are ties - the duel order resampled a few hundred times, on its own page because it costs a second to work out.
Appeals
Teams may file a private appeal within 7 days of publication. Only organizers and the appealing team can inspect an open appeal message.
File a private appeal
Submit a private appeal for your team's submitted project. Organizers will review the evidence and may publish a corrected result revision.
The same figures as JSON: this page's API route.