Poll inclusion
Country trackers aim to include publicly reported, nationwide voting-intention polls for the relevant national legislature. Party or first-preference figures are charted separately.
- Included: published fieldwork dates, pollster, sample size when available, source link, and separate party results.
- Excluded: two-party-preferred results, seat projections, approval ratings, election results presented as polls, regional polls on national charts, and hypothetical scenarios that cannot be compared with the main series.
- Missing values: left blank. They are never treated as zero.
Dates, ranges, and elections
Polls are positioned at the midpoint of their reported fieldwork period so multi-day polls use one consistent chart and filter date. Tables retain the full fieldwork range, and the detail drawer reports both endpoints. Election markers use the exact election date rather than the middle of an election year.
The election calendar records confirmed, provisional, and date-not-announced status with a public source for every tracker. A constitutional deadline, regular cycle, or polling-page title is not converted into an exact future date.
The 1Y, 5Y, 20Y, and All controls change the visible period without changing the underlying archive. When a source has no polling for a period, the chart leaves that period empty.
Inspecting and exporting individual polls
Each published poll has a stable public identity. Opening a chart point or table record keeps its evidence visible in a detail drawer and adds that identity to the country URL, so a copied link reconstructs the same poll. The drawer shows fieldwork dates, pollster, sample when reported, the currently visible party values, the row-level public source recorded for the poll, snapshot-level compiled-source context, and relevant coverage context. A recorded row source may itself be a compiled archive rather than the pollster's original report. Compiled source pages are documented for the country snapshot and are not attributed to an individual row without a row-level mapping.
The chart is one keyboard stop, not a tab stop for every point. Left/Right moves through polls in the current party series, Up/Down changes party, Home/End moves to the visible series boundaries, and Enter opens the selected poll. A visible marker and polite live announcement identify the party, value, pollster and full fieldwork dates. The cursor covers the complete filtered observation set even when a very dense chart draws a reduced number of point marks. “View in table”, “Return to chart”, Close and Escape restore focus to the corresponding chart or table location.
The pollster research view compares each party observation with a weighted average of polls by other pollsters whose midpoints fall within 14 days. Sample size and proximity affect that comparison; at least five evaluated polls are required before a pollster-party median and interquartile range appear. The result describes variation in this archive. It is not an estimate of bias, accuracy, house effects or future performance.
The weekly historical map uses only polls available on or before its selected date. A country is coloured only when a sourced local-bloc definition covers that date, every included party has a value, the poll midpoint is outside declared gaps and the poll is no more than 90 days old. Grey means local comparison unavailable. No values are interpolated, and local blocs are not internationally equivalent.
The poll table applies range, pollster, date, and text filters before dividing results into pages of 50 rows on desktop or 20 on compact screens. “View in table” locates the selected stable record. A copied citation identifies the poll and its public URL; a correction email contains only that public identity and link.
CSV downloads include every currently filtered poll, not just the visible table page. The JSON download is separate and provides the complete immutable snapshot selected for that country. SVG and PNG chart exports include the country, selected range, source-coverage date, and a notice that the chart is descriptive rather than a forecast.
Headline and trend averages
The summary headline uses polls ending within 45 days of the latest available fieldwork. Each poll is weighted by recency, using a 30-day half-life, and by the square root of its sample size. Square-root weighting reduces, but does not eliminate, the extra influence of unusually large samples. When a source does not publish a sample size, the calculation uses 1,000.
Rolling average
A local moving window that follows nearby observations. It is the default because it is relatively easy to read against the points.
Smooth average
A wider moving window for a calmer long-term line. It reduces visual noise but can respond more slowly to genuine changes.
Exponential average
An exponentially weighted series that gives progressively more influence to recent polls.
Gaps and discontinuities
Trend lines break across long polling gaps instead of drawing a straight line through years with no evidence. Loaders also exclude table types and implausible rows that do not belong in the comparable national party series.
A sudden change can still reflect a real political event, a pollster change, or a source-table boundary. The points, pollster filters, source table, and range controls remain available so visitors can inspect it.
The quality dashboard lists every declared gap alongside snapshot age, last successful source check, source-coverage date and any non-blocking validation warning.
The global map
Each published country defines a local left and right grouping for the limited purpose of showing which broad bloc is ahead. Red indicates a left-bloc lead, blue a right-bloc lead, and purple a close race.
These classifications are not relative rankings. A blue country is not necessarily further right than another blue country, and parties in different countries are not assumed to be equivalent. The country summary documents the parties included in its local grouping.
Country comparison
The country comparison places two or three national archives on the same calendar range but keeps their party lines, legends and vertical scales separate. It does not rank parties across countries or combine them into one international series.
The optional local-balance line is right-minus-left within the selected country's own navigation grouping. It is calculated only when the poll midpoint falls inside a time-bounded manifest period and every included party reports a value. Declared archive gaps break the line; a missing definition or incomplete poll is shown as unavailable. The period link identifies the public polling archive whose party set is being grouped, not an external claim that the grouping is universally or internationally equivalent.
Freshness, archive span, next-election metadata and 30-day movement are reported separately for each country. Movement is descriptive change between eligible observations in the same period, never a forecast, seat projection or interpolated estimate.
Reviewed country guides
Every published tracker links to a country guide explaining the electoral-system details needed to interpret national party polling, the local party and bloc navigation used by Bismuth, and limitations specific to that archive. These guides provide context only; they do not add forecasts, seat models or automatic political commentary.
Guide sources are maintained as restricted Markdown with a visible last-reviewed date. Publication requires every section and at least two HTTPS sources to pass validation; the build escapes the text into sanitised HTML, and the raw source is not deployed. A local bloc explanation remains country- and period-specific and does not turn the map or comparison into an international ideology ranking.
Published snapshots and data states
Visitor pages read complete, versioned Bismuth snapshots published with the site. They do not scrape MediaWiki or another polling source in the browser. Scheduled ingestion retains the original public links and source attribution, validates each country independently, and promotes only a passing snapshot.
If collection or validation fails, the previous immutable snapshot remains available and the country is marked stale or in error instead of being presented as freshly updated. The chart status distinguishes a current published snapshot from last-known-good or fallback data and states the available source-coverage date.
The data catalogue documents the stable JSON contract and links each licensed country archive to its immutable snapshot, versioned API responses and snapshot-level source attribution. Source-register entries describe the compiled snapshot and are not asserted to be the original report for every row.