Method
The Joint List re-formed. Now we count one
Four days ago this page said it would fold Balad in only when a source published the reunified list under a name of its own. That happened, the rule fired by itself, and the thing that stopped it firing seven months early was machinery built for something else entirely.
On 23 August this page carried a note titled The Joint List re-formed. We still count two. It ended on a condition: Balad would be folded into the Joint List when, and only when, a source published the reunified list as a list of its own.
That has now happened. The forecast prints one row where it printed two — the Joint List with 6 seats, and neither Hadash-Ta’al nor Balad drawn any more.
What is worth writing down is not the outcome. It is that nobody decided it.
The rule fired on its own
The corpus holds a file of party lineage — which lists became which. It runs on one rule: a component folds into its successor only once the successor has itself been polled. Until then the component genuinely is the ballot, and folding it early would credit seats to something no source has measured.
On 23 August that rule was holding the fold back. Both sources were still reporting the merged figure inside the old columns, so there was nothing to fold into. The entry was written and left inert, with a note saying what would activate it.
Wikipedia’s revision of 27 August gave the Joint List a column of its own. The entry activated. No judgement was made this week that was not already made four days ago.
It reaches back a week further than we noticed
The natural assumption is that the fold begins on the day we saw it. It does not.
The revision restated two earlier readings under the new column — Kantar’s of 20 August and Yossi Tatika’s of the same day — alongside Channel 13’s poll of the 26th. So the corpus now holds eight standard readings of a Joint List column, running from 20 August, not one from the 27th.
The page was fetched on the 27th. The polls it activates on are a week older. Dating the change to the fetch would date it to us rather than to the record, which is precisely backwards.
Added 28 August 2026. The encyclopaedia moved those rows back out the next day: its standard-ballot table went from fourteen rows reaching to 19 August down to eight starting on the 25th, so the earliest reading of a Joint List column is now 26 August, not the 20th. The paragraph above stands as what the record said on the 27th, which is what a dated post is for. The fold itself is unaffected — it is conditional on the successor having been polled, and it still has been. What the episode shows is the reason snapshots are dated at all: without one for each day, this revision would have been invisible, and the corpus would simply have quietly disagreed with a post nobody could check.
How it was caught
Not by anyone reading the seat table. By a check that compares readings of the same sample from different sources.
Channel 13’s poll of 26 August arrived twice — once from each source we read — and the two agreed on every list except one pair. One filed ten seats under Hadash-Ta’al; the other filed the same ten under the Joint List. The check prints those as near-misses and labels them: three long-standing ones in this corpus are marked same document, which means two readings inside one page and usually a hypothetical ballot. This one was marked different documents, which is the label worth stopping for.
Run without the fold, the forecast put the Joint List on 7 seats beside Hadash-Ta’al on 5 — and took a seat off Likud, Yashar, Shas, United Torah Judaism, Otzma Yehudit, the Democrats and Yisrael Beiteinu to pay for it.
That pattern is worth recognising on sight. The chamber is fixed at 120, so a list that has appeared from nowhere can only be funded by every real list at once. A new row plus every other row down exactly one seat is one electorate counted twice, never a shift in opinion.
The part we did not plan
Here is the detail that makes this worth a post rather than a line in a commit.
A Joint List column has been in the 2026 corpus since 29 January — 36
readings of it, from January through 13 August, months before Hadash, Ta’al and
Balad re-formed anything.
Every one of those 36 is a scenario: a second ballot question put to one sample, asking what would happen if the three re-formed. The pollsters were modelling the merger long before it existed.
Had those been read as ordinary polls, the lineage rule would have seen a polled successor in January and folded Balad away seven months before the merger. The forecast would have spent half a year reporting a list that was not on any ballot.
What stopped it was the scenario handling — machinery built for a completely different purpose. It exists because a pollster asking two questions of one sample must not be counted as two polls, which would let one night’s fieldwork vote twice in an average. That is an aggregation concern. It turns out to be load-bearing for lineage too, and nothing in either piece of code says so.
We have written it down now, where someone changing it would look.
What actually moved
Four new polls arrived this week: Lazar for Walla, HaMadad for Channel 13 and Direct Polls for i24 on 26 August, and Filber for Channel 14 on the 24th. A fifth changed row turned out to be the compilation renumbering a footnote.
The Joint List stands at 6 seats. Likud and Yashar are level at 24 apiece, and the model gives Likud a 53% chance of finishing largest against Yashar’s 47% — which is to say the order at the top is not settled and should not be reported as though it were.
The three blocs read 52 for the Netanyahu bloc, 57 for the opposition and 11 for the Arab parties. As ever, those are sums of seats under a published bloc coding rather than a forecast of who will govern. The opposition and the Arab parties hold 68 seats between them, which is arithmetic: it is six separate lists that do not act as one, and nothing here says they would sit together.