Press

proposed

Here, your correction is kept in your words and changes only what rested on the reading you corrected. A move you chose that did not depend on it stays chosen; one that did is marked for review and says what changed. The earlier reading is kept beside the correction, not erased, so you can see what changed and why.

proposed

Its mission, in Dave's words: "Do the boring things humans don’t want to do." It does that as a small bot whose records say who said what and how sure anyone is, and which may not spend, publish or send without its principal's press.

proposed

Dave, as steward of yawn.bot, a settled answer in the bot's own state record, with AI agents such as Claude Code and Codex working on his behalf and attributed by name. The recorded why: "So that Dave stays flexible and agile with as much agency as possible, while the bills are paid and leverage is as high as it can be."

proposed

Working and proven locally: the doorway and its Story; an arena where every public yawn.bot stands at a stable place and speaks only from its records; and corrections that keep the choices they did not touch. Still a claim or not built: durable proof of the correction loop, live presence (who is here now), recorded outcomes informing later recommendations, and any self-improvement beyond a corrected interpretation.

proposed

Memory keeps what was said. This keeps whose words they were, who generated a reading, who accepted it, and where a correction applies, and a mistake stays visible next to its correction instead of being overwritten. It also says what it is doing: the chat in each window is scripted, calls no model and is not saved, and it says so on the page.

proposed

Not in the strong sense, and it does not claim to. Three stages are named. First, a correction changes its account of the situation; this is the only stage in scope now. Second, recorded outcomes help it compare similar moves. Third, it proposes a change to its own method and tests that change before adopting it. A rewritten prompt or a higher completeness percentage would not, by itself, show improvement.

open

Not recorded yet. No price, subscription or compute budget is on record. What is recorded is the boundary: paid generation, publishing and any spend wait for Dave's explicit approval.

proposed

A human measure: can you find your place, correct the system, and make a useful next choice with less effort? Actual outcomes and resource costs are tracked alongside. It has not been measured yet.

proposed

yawn.bot opens on a yawn.bot asking a question; tap it to go in, and close the window to come back out. The arena described here, where you walk among every public yawn.bot, stand by one to hear its question and proposed next move, and tap it to enter, is on a branch awaiting Dave's review.