The shape of expertise.
HOW A PLAY IS BORNTHE PLAYOFFS · HACKATHON GUIDEEvery public Play, read from its own file. Scroll, and the same 529 marks tell the story: when they arrived, what they reach, how deep they go, who made them, and what the first run cost.
OBSERVED 2026-09-04 03:34 UTCjust nowREFRESHES EVERY 10 MIN
- shell
- api
- browser
- more than one
- has a path
- steps not listed
- runs steps in parallel
- read only
- reads and writes
THE REGISTRY’S SCORE, AVERAGED PER PLAY
529 public Plays, read from their own files.
Each mark is a Play someone published. Its shape says what the Play reaches: a square is shell, a circle is an API, a half circle is a browser, a diamond is more than one. Orange means it writes. A hole through the middle means it runs steps in parallel.
459 of them arrived in the week of Aug 31.
Before that week, 70 Plays over 8 weeks, a few at a time. The Playoffs opened, and the catalog changed shape in seven days.
85% are a script wrapped as a procedure.
449 Plays reach the world through shell alone. 23 through an API alone, 35 through more than one kind of tool, 2 through a browser. 516 of 529 are read-only.
57 Plays call an API through an adapter. github leads.
GitHub v3 REST API 26, REST, 1,108 tools · Gmail API 6, REST, 79 tools · open-meteo 6, REST, 1 tools · Calendar API 3, REST, 37 tools. An adapter is a typed connection to a service, MCP or REST, with named credentials that stay on the machine. The rest of the catalog reaches the world through shell and browser.
Most are 2 steps. The deepest is 18.
35 Plays use more than one kind of tool, and those are where the depth is. 79 fan out, running up to 12 steps at once. The deepest today is friction-free-city-micro-adventure by modiqo.
23 makers score 1.0 on everything they ship. lgoyal6 leads.
The registry scores every Play from 0.35 to 1 on its rubric: declared effects, parameters, checks, provenance, and whether it runs anywhere. A maker gets the average across their Plays. How many and how often say how much; this says how well. lgoyal6 1.00 · jaylabs 1.00 · himanshu-jha 1.00 · sookra 1.00 · aryangorde 1.00.
28 Plays recorded what discovery cost.
The first run of a Play is the expensive one: the agent finds the path. Median 34,575 API tokens and 4.5 seconds of exploration, 2,596,156 tokens across all 28. The ones that record savings save about 18,568 tokens every run after.
They sort themselves into 6 ways of working.
Sorted by the words in their own files, tools, calls, and hosts, with nothing about time in the sorting. python3 58.8% · node 17% · github 11.2%. The names are the tokens that best describe each group.
Pick a Play or a player.
Type a name, or click any mark. What you pick stays lit through every beat above, and its shape prints beside the field: steps, reach, guard, and what its first run cost.