Method
Fifty-four each, and a poll counted twice
Three polls fielded on 3 September leave the two largest blocs level at 54 seats each and the Arab parties with 12, and the headline no longer says either bloc leads. Two dating errors came out with them: one of ours, counted twice since January, and one the compilation is making this week.
The front of this page now reads Opposition bloc and Netanyahu bloc are close, at about 54 and 54 seats, and Arab parties with 12, from 17 polls. Yesterday it read 55 for the opposition, 53 for the Netanyahu bloc and 12 for the Arab parties, and it said the opposition led. One list moved one seat, and that was enough to take the verb out of the sentence.
What arrived
Three polls, all fielded 3 September, and both sources carry each of them.
- Filber for Channel 14, sample 865: Likud 31, Yashar 22, Shas 10, Together 9, United Torah Judaism 8, the Democrats 8, Religious Zionism 7, Otzma Yehudit 7, the Joint List 7, Yisrael Beiteinu 6, Ra’am 5. Amcha Israel is under the line. Blocs 63, 45 and 12.
- Kantar for Israel Hayom, sample 555: Yashar 24, Likud 21, Together 13, the Democrats 9, Yisrael Beiteinu 9, United Torah Judaism 8, Otzma Yehudit 7, Shas 7, the Joint List 7, Religious Zionism 6, Ra’am 5, Amcha Israel 4. Blocs 53, 55 and 12.
- Yossi Tatika for Zman Yisrael, sample 500: Likud 22, Yashar 22, Together 15, Yisrael Beiteinu 10, Shas 9, United Torah Judaism 8, the Democrats 8, Otzma Yehudit 6, the Joint List 6, Religious Zionism 5, Ra’am 5, Zionist Home
- Blocs 50, 59 and 11.
They disagree by a wide margin. Filber has the Netanyahu bloc 18 seats ahead; Tatika has the opposition 9 ahead. That spread is the story of the round, and the model reads all three.
One seat, and the whole table
The chamber printed here is one rule. A list takes its median across the simulations, a median below the 4 seats the threshold allows becomes nothing, and any seats left over go to the largest lists, one each, until the chamber is full.
Yesterday the surviving medians added to 119, so one seat was left over and went to Yashar, printing it at 24. Today Likud’s median rises from 22 to 23, the medians add to 120 exactly, and there is nothing left over. So Likud gains a seat and Yashar loses the one it was lent:
- Netanyahu bloc: +1 from Likud. From 53 to 54.
- Opposition: −1 as Yashar’s extra seat goes back. From 55 to 54.
- Arab parties: 12, unchanged.
Every other list prints exactly what it printed yesterday. Two seats of difference between the two largest blocs closed on a single median crossing a whole number.
Why the headline stopped saying “leads”
The headline says a bloc leads only when the simulations make it the largest in three draws of five, counting a tie as half. Yesterday the opposition was at 60.4% and cleared that by four parts in a thousand. Today it is at 57.4%, and the headline names both blocs and says they are close instead.
Underneath, the opposition finishes ahead of the Netanyahu bloc in 55% of simulations against 58% yesterday, the Netanyahu bloc is ahead in 40% against 37%, and they are level in 5%. The Netanyahu bloc’s chance of 61 went from 3.8% to 4.6% and the opposition’s from 9.1% to 8.3%. The opposition and the Arab parties reach 61 together in 92% of simulations, from 93%. That is arithmetic and nothing more: six lists that do not act as one, and nothing here says they would.
The top and the line
Yashar is still the list most likely to finish largest, at 53% against 55% yesterday, with Likud at 47% against 44%. The printed table has them level at 23 apiece. Filber has Likud ahead by 9, Kantar has Yashar ahead by 3, and Tatika has them level.
The two lists on the threshold moved apart again. All three polls seat Religious Zionism, on 7, 6 and 5, and its chance of 4 seats or more went from 50% to 58%, its chance of nothing from 33% to 26%. Only Kantar seats Amcha Israel, and its chance of 4 or more fell from 63% to 55%, its chance of nothing from 16% to 25%. Both still print 4. Zionist Home takes 4 from Tatika, the first time a poll has seated it in this window, which lifts its chance of 4 or more from under 1% to 4%.
Two dates that were wrong
Both sources have now dated a poll to the wrong year, and neither error shows in the figures. Every seat is right and only the date is wrong, so the reading does not contradict anything: it lands in the corpus as a second poll.
Ours. Wikipedia hangs a sort key on every date cell, and this site used to
prefer it, because it carries a year where the printed cell shows only a day
and a month. The year in that key comes from a parameter of the row’s template
that the rendered page never displays, so a typo in it is invisible to a reader
and reaches nothing but the key. Two rows of the 2026 article carry one: 14
Jan and 11 Feb sort as 2025 where every row above and below them sorts as
2026. Three Kantar polls for Israel Hayom were dated a year early on that, and
each then sat in the corpus beside the same reading at its real date, the same
600 respondents and the same seats for all eleven lists that won any, counted
as a separate poll.
The site’s own table of monthly averages is where a reader could have seen it. The row for 30 January 2025 read 8 polls in the window and a Netanyahu-bloc average of 51.3; it now reads 7 polls and 51.1, with the opposition at 58.9 rather than 58.7. The row for 27 February 2025 goes from 17 polls to 15. The corpus holds 503 polls for this election rather than 505. None of this touched the current forecast, because both readings are from months the 60-day window left behind long ago — but they had been in the campaign’s published history since January, and one of them twice.
The fix is that the page’s own rule decides the year and the sort key is no longer read. Across all 1,476 polls these adapters read, the key never disagreed with that rule except on those three rows.
Theirs. The 2026 compilation is making the same mistake in the other direction, and it is live as this is written. On 4 September it published the seven polls of 2 and 3 September a second time, dated September 2025 — including readings that report Amcha Israel, a list first polled in August 2026, and the Religious Zionism–Zehut technical bloc, agreed on 1 September 2026. Neither can appear in a poll taken a year earlier.
Those seven are out of every aggregate here, but by luck rather than by design: the misdated copies label the reunified Arab list “Joint List”, and before it re-formed on 19 August that label means a hypothetical ballot rather than a real one. That is a coincidence, and it ends the day the compilation relabels that column. So the check that now stands is not on the dates but on what a date implies. A list has a first day, and a poll fielded before it cannot have asked about it. Two more say that one sample is not two polls, whatever a source writes in a date cell.