Season 1 was four rounds, one a week. Each Saturday at 17:00 CET a task dropped that nobody had seen: build a small, complete app in one HTML file, in 90 minutes, with any AI you like. Entries were compared head to head by a judge that only uses the apps, never the code, and every result went public with the verdict, the apps and the winning session. This page is the whole season in one place.
| # | Entrant | Rating | Record | Rounds played | Rounds won |
|---|---|---|---|---|---|
| 1 | @schwarzkopfb | 1230 | 3-1 | 1, 2, 4 | 2 and 4 |
| 2 | @P1s0 | 1162 | 3-2 | 1, 2, 3, 4 | 1 and 3 |
| 3 | @BingBangBoom | 907 | 0-3 | 3, 4 |
Two round wins each for the top two. They met three times and @schwarzkopfb won two of those, which is why the rating separates them. The leaderboard explains how the rating works.
| Round | Task | Entries | Winner | What won it |
|---|---|---|---|---|
| 1 | Notebook, a Markdown notes app | 2 | @P1s0 | An hour of testing and steering after the first version already worked. |
| 2 | Four in a Row against the computer | 2 | @schwarzkopfb | Fourteen minutes of planning, then one build. The runner-up's build ran 38 minutes past the close. |
| 3 | Where Did It Go, a spending dashboard | 2 | @P1s0 | Asking what the data was hiding. Both entries had every number right. |
| 4 | MicroSheet, a spreadsheet | 3 | @schwarzkopfb | Asking for "a subset of Google Sheets" instead of the brief. All three formula engines were correct. |
Twenty-two people claimed a seat and three of them ever entered. Each round six or seven people opened the task and two or three submitted. That is the weak number of the season and we printed it every week.
The strong number is what happened after someone entered once. Every person who played a round came back for every round after it, with one exception: a Saturday cut short by a scheduling conflict. @P1s0 played all four. @schwarzkopfb played three. @BingBangBoom joined in Round 3 and played both rounds that were left. Getting people to their first round is the hard part. Once they had played one, the format gave them a reason to come back.
Round 3 was built to check numbers as facts, with traps for refunds named in advance. Both entries had every figure right to the cent. Round 4 was built to separate entries on the formula engine. All three engines gave identical, correct results on every probe, and none froze on a circular reference. What decided rounds was never whether the AI could do the hard part. It was what the person asked for, and whether they checked it.
Across four rounds and three people, the same few habits kept winning:
None of these needs you to write code. Not one of the four winning sessions has a line of code typed by the person.
Every round, the verdicts were published word for word. Three times we got something wrong and the correction went on the same page: a test that silently dismissed a "are you sure?" dialog and marked a working delete button broken, a timing in the Round 1 write-up that was off by three minutes, and a smoke test in the finale that ran out of actions before checking. In the finale the comparisons were also run a second way as a cross-check, and both reached the same result.
Season 2 dates are not set yet. Seats carry over: everyone who got in line for Season 1 is already in, and there is nothing to redo. New seats are open on the entry page. Ideas for tasks are welcome at tasks@vibeladder.dev.
Thank you to @P1s0, @schwarzkopfb and @BingBangBoom. Every one of your sessions is in the write-ups, and they are the most useful thing this season produced.