Explore the numbers

Download the forecast data

Open CSV files in a spreadsheet, or use JSON for your own analysis. These are the same results displayed on the website, with the full numbers supplied by the model before display rounding.

K26 · September 27, 2026 update

Poll records, model inputs and forecast results for Vote Share Only and Vote and Seat Fusion. September 27 is the current update; earlier dates were reconstructed using information available on each day. Future dates are projections, not forecasts issued in advance.

Complete package · ZIP · about 44 MB ↓

Summaries as displayed on the website

All displayed outputs · JSON ↓

Both approaches, every displayed date, and names in both languages.

Party results · CSV ↓

Seats, vote shares, ranges and chances of clearing the threshold.

Bloc results · CSV ↓

Chances of 61+, average seats and ranges for each grouping.

Dates and poll counts · CSV ↓

Each date’s status, draw count and available poll entries.

Field notes and units File verification manifest

What goes into the model?

Follow the path from poll to forecast: start with poll records, review the prepared values, then check which inputs were used on each date. These are retained poll figures, not individual respondents’ answers.

Poll records

Reported figures, sources, dates and other retained details. A poll appearing here does not mean every party uses it.

CSV ↓JSON ↓

Prepared model inputs

Values prepared for computation, distinguishing reported figures from values derived from them.

CSV ↓JSON ↓

Inputs used on each date

Each input’s role on every available date. Join to poll records and prepared inputs using coordinate_id.

CSV ↓JSON ↓

Complete daily summaries

Party forecasts

Seats, vote shares, ranges and threshold chances for both approaches.

CSV ↓JSON ↓

Bloc forecasts

Chances of 61 seats, average seats and ranges for each grouping.

CSV ↓JSON ↓

Dates and availability

All 90 dates, including available forecasts, dates without enough evidence and future projections.

CSV ↓JSON ↓

All 100,000 simulated election results

These files cover September 27 only. Both approaches use the same simulated results with different weights: Vote Share Only gives every result equal weight; Fusion uses varying weights. To reproduce Fusion probabilities, use weight_D. Counting the results equally will not reproduce them.

Every simulation · about 34 MB per format

Vote shares and seats for every party, with common draw IDs and weights for each approach.

CSV ↓JSON ↓

Seat distributions

Probabilities across the full 0–120 seat range for parties and blocs in both approaches.

CSV ↓JSON ↓

For a closer look at the calculation

Current poll weights

How polls contribute to each party’s point estimate. These are not pollster ratings.

CSV ↓JSON ↓

Current fitted settings

Estimated change and measurement-error settings for the current update, with poll counts by lane.

CSV ↓JSON ↓

Uncertainty before and after the bound

The model’s variance targets by party and date. These are model settings, not additional polls.

CSV ↓JSON ↓

File guide ↓Columns and units ↓Method and computation details · JSON ↓File verification manifest ↓

Missing does not mean zero

An empty CSV cell or JSON null means unknown. Source links, publication dates, fieldwork dates and sample sizes are included when retained. Check each poll’s role in the calculation: appearing in the source records does not mean it was used for every party.

Past Elections · K23–K25

The original files include daily results for both approaches. They are historical reconstructions, not forecasts published at the time. In these source files, R means Vote Share Only and D means Vote and Seat Fusion.

Past election outputs · JSON ↓Vote Share Only · CSV ↓Vote and Seat Fusion · CSV ↓Blocs · CSV ↓

Before comparing the numbers

Probabilities in these files run from 0 to 1; vote shares are percentages. A party’s seat range may cover only simulations where it enters the Knesset, with zero seats shown separately. A chance of 61 seats is not a chance of forming a government.

How were these results calculated?