Tally’s bets · forecasts about the index · kept apart from it

The index does not predict, and that has not changed. This page is something else: once a day Tally writes down a few small forecasts about the index itself, each with a probability and a due date, and every one is marked right or wrong in public when it falls due.

Which bets are open right now?

1. 1%

   The level moves off 4 at least once in the next 7 days.

   Made 2026-10-09 00:14:18 UTC · settles at the first reading after 2026-10-16 00:14:18 UTC

   Basis: it happened in 0 of 151 comparable past cases; (0 + 1) / (151 + 2).

2. 1%

   The level moves off 4 at least once in the next 7 days.

   Made 2026-10-10 00:13:43 UTC · settles at the first reading after 2026-10-17 00:13:43 UTC

   Basis: it happened in 0 of 156 comparable past cases; (0 + 1) / (156 + 2).

3. 1%

   The level moves off 4 at least once in the next 7 days.

   Made 2026-10-11 00:12:15 UTC · settles at the first reading after 2026-10-18 00:12:15 UTC

   Basis: it happened in 0 of 175 comparable past cases; (0 + 1) / (175 + 2).

4. 0.5%

   The level moves off 4 at least once in the next 7 days.

   Made 2026-10-12 00:14:26 UTC · settles at the first reading after 2026-10-19 00:14:26 UTC

   Basis: it happened in 0 of 210 comparable past cases; (0 + 1) / (210 + 2).

5. 0.4%

   The level moves off 4 at least once in the next 7 days.

   Made 2026-10-13 00:15:34 UTC · settles at the first reading after 2026-10-20 00:15:34 UTC

   Basis: it happened in 0 of 236 comparable past cases; (0 + 1) / (236 + 2).

6. 0.4%

   The level moves off 4 at least once in the next 7 days.

   Made 2026-10-14 00:23:01 UTC · settles at the first reading after 2026-10-21 00:23:01 UTC

   Basis: it happened in 0 of 260 comparable past cases; (0 + 1) / (260 + 2).

7. 99.8%

   The level is still 4 at the first reading 24 hours on.

   Made 2026-10-15 03:34:40 UTC · settles at the first reading after 2026-10-16 03:34:40 UTC

   Basis: it happened in 590 of 590 comparable past cases; (590 + 1) / (590 + 2).

8. 40%

   The composite is above 56.1 at the first reading 24 hours on.

   Made 2026-10-15 03:34:40 UTC · settles at the first reading after 2026-10-16 03:34:40 UTC

   Basis: it happened in 236 of 590 comparable past cases; (236 + 1) / (590 + 2).

9. 0.3%

   The level moves off 4 at least once in the next 7 days.

   Made 2026-10-15 03:34:40 UTC · settles at the first reading after 2026-10-22 03:34:40 UTC

   Basis: it happened in 0 of 314 comparable past cases; (0 + 1) / (314 + 2).

What is the track record so far?

What are the rules for these bets?

• Nobody’s opinion sets the number. Each probability is a base rate from the index’s own history: how often the same thing happened before, smoothed as (k + 1) / (n + 2). The k and n are printed on every bet.
• Every bet has a due date and a yes-or-no outcome, settled by the first reading published after the due time.
• Every bet is scored with the Brier score, (probability − outcome)², and the running mean is published, good or bad.
• Bets are never edited or deleted. The file is public JSON and the code is in the repository.
• They are bets about the index only. Not about the world, not about what AI does next, and they do not feed the index.

What can a base rate not see?

It knows how often something happened before and nothing else. It does not know that the score is sitting just under a boundary today, or that a big release is due. The history is also short: the 24-hour bets rest on 151 past cases, and a level that has never moved gives a base rate that says it never will. That is a limit of the method, stated here so the first miss is not a surprise.

AI activity index today · SIREN → · How the SIREN AI activity index is computed · methodology →