A scenario's probability moves every day. Palanor keeps what it said each day, so that when the scenario settles it can be graded on its whole run.
The daily record
Once a day, after the nightly recompute, each scenario writes a record: the probability shown, the signals and weights that produced it, and the criteria it was answering. The record is never edited. If a weight changes later, the change shows up in later records; earlier ones stay as they were.
How a scenario settles
Most criteria name a filing, a disclosure or an event, so a person settles them on the Resolution review page. Some criteria settle on a series Palanor already carries, such as a credit spread or a Treasury yield. For those, Palanor drafts a rule (the series, the comparison, the threshold, how long it must hold, and the dates), checks it against the stored data, and a Palanor reviewer confirms it. A confirmed rule settles the scenario from the data and writes a note saying what the data showed. Nothing settles from a rule nobody confirmed.
Two limits keep this honest. A "did not happen" is never called before your settlement date. A "happened" is called early only when the criterion says it can happen at any time inside the window. If a rule settles something wrongly, override it on the Resolution review page; the rule's call is kept in the record.
The scores
- Brier at lock compares the probability recorded when the criteria were locked with what happened. Lower is better; a scenario that always says 50% scores 0.250.
- Time-averaged Brier scores every day from lock to the outcome, so a scenario that sat at 20% for a year and moved to 90% the week before is not credited with the move.
- Fixed leads show what the scenario said 365, 180, 90 and 30 days before the outcome.
Ambiguous and void outcomes carry no score and are still published.
Evidence for the weights
When a scenario settles on a series Palanor carries, each mapped signal is tested against that series' history: does its change lead the series' next change, in the direction your map says, and does it improve a forecast out of sample? The Signal Map shows the result and may suggest moving a signal's strength by one step (0.1). You can accept a suggestion or ignore it; every change can be reverted. A signal that leads the other way is flagged for you; Palanor never flips a direction.
Once 30 scenarios have settled, a second test opens: which signals' contributions actually improved the scores. Until then it stays closed.