We found a problem with the system itself, not with the world, and corrected it by hand.
While reviewing the map we found a problem with the system itself, not with the world. Every event Brújula processes can move a thesis's probability a little. Over months, those small movements kept adding up, almost always in the same direction, and several theses ended up near the extremes without any specific fact to justify it. The system was accumulating: it recorded what happened, but it did not discount everything that didn't.
We have done two things.
First, an editorial recalibration. A person has reviewed the theses that were outside their reasonable range or far from other independent estimates, and has set a new value for 13 of them. Each correction appears in its thesis's history, marked as an editorial recalibration and signed by the Brújula editorial team. The rest have not been touched. The eight master theses are calculated from their specific theses, so they change too, even though no one touched them.
Second, a brake. From now on, no large movement applies on its own: if it takes a thesis outside its range, it waits for a person to read it.
There is nothing you need to do. The map's figures are already updated, and the previous ones remain in the public history: we are not deleting them, because they are also part of how the system got it wrong.
9 Oct 2026 · we corrected the polarity of IA-1.4 (an inverse thesis of TM-1); TM-1 goes from 62% to 69% because of that correction, not because of new evidence.
11 Oct 2026 · second editorial recalibration. On setting a reasonable range for each of the 39 specific theses, fifteen fell outside theirs —fourteen above— and have been given a new value within range. Signed by the editorial team at the owner's request. Like the first one, this is not new evidence: it is the system correcting itself, which is why these movements do not count towards the 28-day change.
12 Oct 2026 · from today, the resolution criteria are published on each thesis's page: exactly what would have to happen for the thesis to count as resolved.
19 Oct 2026 · editorial recalibration of a single thesis: TEC-4.1 («Sustained net-positive commercial fusion <2035») goes from 60% to 45%. The previous value had been built almost single-handedly by one source, arxiv_fusion: over the last 30 days it moved theses fifty times and not once downwards, forty-nine of those on this thesis. Its last downward correction is from 21 July (across its whole history, 97 movements up and 6 down). 45% is not a figure set by eye: it is where the thesis would have ended up had the per-source speed limit been active from the start, according to the simulation over its full history. Signed by the editorial team at the owner's request. Like the previous ones, this is not new evidence —it is the system correcting itself— so it does not count towards the 28-day change.
21 Oct 2026 · we are correcting a mistake of our own. An internal Brújula calculation summarised each week the press coverage we were already reading, and that summary then came back in as if it were one more piece of news: the same news counted twice. On top of that, a metric meant to be produced once a week was being produced every day, so each week entered seven times. Between May and October this pushed up 76 points on «Measurable collapse of shared public epistemology» (SEN-7.5) and 47.5 on «Terminal polarisation of OECD democracy» (SOC-6.3), across 120 movements without a single one downwards. The calculation has been switched off since 20 October. Both theses had twice reached practical certainty and both times were brought down by hand —in July and in October— without us knowing this was the cause. The new values, 40% and 25%, are not a subtraction of those pushes: they are the editorial team's judgement of what it would take for each thesis to be met, reading its resolution criteria and the external sources that measure it. Their reasonable ranges are also re-signed, because the previous ones had been set around the inflated figure. Signed by the editorial team at the owner's request.
