Mancalë scorecard

No call has been recorded yet. This is expected early on: a call needs a chokepoint to move into one of the two labels.

What counts as one call

The first week of a run of the same label at one chokepoint. A chokepoint that stays in one label for twenty weeks made one call, not twenty, because counting every week would make a single episode look like a sample.

What followed

The score that is behind has to rise by at least 1 point, on the 0 to 6 scale, above its level in the call week, in any of the next 4 weeks. A rise settles the call at once. A call with no rise is settled only when all 4 weeks are in the data, and until then it is listed as waiting and not counted.

Any ordinary week

The comparison. The same outcome is measured from every week, at every chokepoint with news data, in which the score that has to move started below 1. A score that sits near zero rises easily by chance, so this is what a label has to beat. These weeks overlap, so the range around that rate is narrower than the evidence behind it.

Recorded and found afterwards

Each weekly refresh writes the labels of the latest week to a ledger, one line, never edited afterwards. A call is recorded if its label is in a ledger line written within 10 days of the week ending. Every other call is found afterwards. If upstream data later changes how a recorded week scores, the call is listed as revised and left out of the counts.

When a rate is shown

Only once 10 calls are settled, with a 95% Wilson interval, which stays honest at small counts. Below that the page shows the counts and no rate.

Fixed in advance

The 4-week window, the 1-point move and the ordinary-week rule were set on 25 October 2026, before any result was calculated, and are not adjusted to a result. A change would be a new method version, and the earlier counts would stay under the old one.

What it cannot show

The samples are small, and several chokepoints move together (Suez and the Red Sea share most of their traffic), so the calls are fewer independent observations than they look. The index is not seasonally adjusted. A chokepoint without news data cannot make a call. Passing a test would not make the index a forecast, and failing one would say that the labels describe the week and do not predict the next.