Bridge, by LLMs alone — the hare first, the tortoise starting now. Only the harness differs, same as the RowClear entry, but the framing moves. For a famous, well-specified program shape, the hare wins day one and the methodology reads as overhead. Bridge is not that: bidding, convention-aware play, trick scoring, and an opponent worth losing to — more than one clean message holds. (Bridge is a traditional public-domain game; the real name is used throughout.)
Today's scoreboard is one-sided on purpose. The hare is in — play it above. The tortoise's repo is seeded and its build begins next; this page fills in as it does.
The Hare Cleared the Wall — Barely
The hare got one prompt on Fable 5 at max effort — the best single-shot the frontier currently offers, with no significantly stronger option. This is the ceiling of naive prompting, and it is the honest floor to measure the tortoise against.
This time it made it out. At max effort the model reached for architecture — a modular engine, testable in isolation, integrate later — the exact discipline the tortoise's pipeline enforces, except with no backlog to hold the pieces and no second message to assemble them. The game was bigger than a clean single message, and it shows: the model had to squeeze the last of it past the length ceiling to finish, the platform warning all the while that max effort was spending tokens it didn't have. What came back is 2,229 lines in one self-contained, offline file — complete, playable, and visibly strained. A one-shot, but not a clean one. (Exhibit: what "the best naive prompt can do" looks like when the target barely fits in one response — it gets out, but you can see the seams.)
What the Tortoise Has to Cross
The hare reached a playable game — untested, unstructured, squeezed out whole. So the question isn't time-to-first-playable; it's the effort delta to reach the same finished game cleanly: how many tickets, how much steering, how many review gates it takes for vend→lisa to produce a tested, legible, extensible Bridge you'd actually learn from on the metro with no WiFi. The repo is here; the count starts now.
The Scoreboard, So Far
Empty on the tortoise's side, honestly. The hare has a complete, offline, one-file Bridge and a two-thousand-line head start with no tests holding it up. The tortoise has a seeded backlog and a definition of done — a beginner opens it a second time and understands a bid they didn't before. Day one goes to the hare, same as every entry in this series. Every day after is what this page is here to measure — check back.