bets · thesis, stake, review date

Book of bets

Chapter 5 tells the reader to keep a decision ledger: thesis, stake, review date and one sentence on "what would convince me I'm wrong". The book does the same with its own claims. Every bet has a probability (P), a resolution measure and a review date; after the review the result stays here permanently — including when the bet is lost.

  1. Z1 · P 0.80 · Review · open

    MCP remains the dominant protocol on the agent–tool edge.

    Resolution measure: Support from the 3 largest model providers and an active Agentic AI Foundation registry.

  2. Z2 · P 0.85 · Review · open

    An open-weights model matches frontier models from at most 12 months earlier.

    Resolution measure: Public benchmarks: comparison with the best closed model from a year before.

    Marginalia: The licence as the third gatekeeper

  3. Z3 · P 0.75 · Review · open

    The output-token price of frontier models falls at least 5-fold.

    Resolution measure: Price lists of the 3 largest providers, September 2026 → September 2028.

    Marginalia: The token price war: two labs cut prices by 20–50 % in a single day

  4. Z4 · P 0.40 · Review · open

    An open standard for portable user memory emerges, backed by at least 2 of the 3 largest providers.

    Resolution measure: A public specification and announcements of support.

  5. Z5 · P 0.65 · Review · open

    A METR-class randomized trial shows a speed-up of more than 20% for experienced developers working with AI.

    Resolution measure: A peer-reviewed publication or preprint reporting an RCT from 2027–2028.

  6. Z6 · P 0.25 · Review · open

    The term "post-cognitive era" in this book's sense appears in at least one peer-reviewed publication citing postcognitive.pl.

    Resolution measure: Google Scholar or Semantic Scholar.

The probabilities were proposed by a model (Claude Fable 5.1, 23 September 2026) and signed off by the author — authorship is decided by the seal, not the origin. After each review we compute the Brier score: 0 is perfect calibration, 0.25 is a coin toss. (opened: )

See the changelog →