Election risk screening — new tools for the 19.04.2026 vote

Published: 2026-05-13

Composite index (47/100, High risk) from five process-integrity signals, plus four context signals (Benford, neighborhood swing, electoral volatility, polling error). Per-section risk screening (10 Critical out of 12,705) and per-party Benford 2BL. Aligned with the international electoral-integrity frameworks.

Update (2026-05-20): The per-section risk score now combines seven signals — a cross-election swing signal has been added, flagging sections whose turnout and winning-party share rose abnormally versus the same section at the previous election. A new geographic risk-cluster map on /risk-analysis groups adjacent same-party elevated-risk sections — the spatial fingerprint of a controlled or corporate vote. The six-signal section-band counts in §4 below predate the swing signal.

Composite index (47/100, High risk) from five process-integrity signals, plus four context signals (Benford, neighborhood swing, electoral volatility, polling error). Per-section risk screening (10 Critical out of 12,705) and per-party Benford 2BL.

This is a follow-up to the integrity analysis published ten days earlier covering the 19.04.2026 election. Since then we have added three new aggregation layers on top of the raw metrics:

  • National composite index — five process-integrity signals (the headline) and four context signals (shown separately).
  • Per-section risk score — 12,705 sections each given a 0–100 score from six signals.
  • Per-party Benford test — statistical distribution of digits in per-section vote counts.

The 19.04.2026 election reads 47 / 100 (High risk) under the new methodology. These are screening tools, not fraud determinations. The original integrity article's bottom-line read — "genuine but historically unusual surge rather than a manipulated outcome" — is unchanged.

1. Why we split process integrity from context

A survey of the leading electoral-integrity composites (Norris PEI, V-Dem EQI, Klimek PNAS, IFES) yields three consistent principles:

  • Separate process from context: process violations (audit-trail loss) are not averaged with statistical inferences (Benford fingerprint).
  • Exclude pollster error: forecasting miss is a forecast failure, not an electoral violation.
  • Within-community dynamics: the diagnostic signal is volatility (excess swing), not the static demographic share.

2. The five process-integrity signals (headline)

These measure disagreements between votes cast and the recorded result:

  • Section screening (38/100): weighted share (1.92%) of national turnout in risk-flagged sections.
  • Machine integrity (90/100): 0.18% drift between flash memory and protocol. The strongest signal this cycle.
  • Missing flash memory (59/100): 0.59% of machine votes outside the end-to-end audit chain.
  • Concentration (29/100): 0.59% of turnout in settlements where one party took ≥80%.
  • Procedural anomalies (18/100): 0.36% of turnout from invalid ballots and additional voters above 10%.

Headline: 47 / 100 (High risk) on the scale <20 Calm, 20–40 Elevated, 40–60 High, 60+ Critical.

3. The four context signals

These describe the environment but do not contribute to the systemic risk score:

  • Benford 2nd digit (8/100): one of twelve qualifying parties shows strong deviation (MAD ≥ 0.08). The test is a prompt to look closer, not a verdict.
  • Neighborhood swing (39/100): ПрБ shows +5.8 pp excess swing inside the tracked communities versus its national performance.
  • Electoral volatility (100/100): Pedersen index 49.7 — the cycle is hyper-volatile (>30% of the vote redistributed). Typically marks a new entrant.
  • Polling error (29/100): 2.51 pp mean MAE across the seven agencies' final pre-vote polls. The miss was directionally identical across all seven houses — every one underestimated ПрБ by 6.08–14.71 pp (ML and CAM in single digits; the other five all 10+ pp short) — pointing to a late surge that broke after fieldwork closed, not a methodology failure.

4. Per-section risk screening

Combines six signals including recount adjustments and peer-outlier z-scores.

BandSectionsShare
Critical (≥80)100.08%
High (60–80)2301.81%
Elevated (30–60)1,69613.35%
Low (<30)10,76984.76%

The 10 Critical sections (3–24 votes each) are mostly driven by invalid ballots and additional voters. The full ranked table is on /risk-score.

5. Benford's law per party

We test the second-digit (2BL) distribution. Four parties show MAD ≥ 0.04, but only two (АКБ and СБ) have enough sections (n ≥ 100) to be meaningful.

  • АКБ: 0.092 (strong)
  • БМ: 0.089 (strong)
  • СБ: 0.077 (moderate)
  • ИТН: 0.061 (moderate)

Important: none of the major parliamentary parties shows meaningful deviation.

6. Bottom line for 19.04.2026

The most serious issue is the audit-chain weakness (machine integrity 90 and missing flash 59). The enormous political realignment (Pedersen 49.7) is the context for all other anomalies. The neighborhood excess swing for ПрБ (+5.8 pp) is noticeable but moderate.

Takeaway: signals of a compromised audit chain are real and warrant attention; the neighborhood realignment is noticeable but moderate; the political reshuffle is enormous but not by itself anomalous.

Methodology and sources

All data and methodology live on /risk-analysis. The composite is computed client-side in useRiskComposite.ts — every formula and threshold lives in a single file and is auditable. The page /risk-analysis/methodology remains the quick reference for what each component measures.

References:

  • Pedersen, M. (1979). The Dynamics of European Party Systems: Changing Patterns of Electoral Volatility. European Journal of Political Research 7, 1–26.
  • Norris, P. (2014). Why Electoral Integrity Matters. Cambridge UP.
  • Mebane, W. (2006). Election Forensics: The Second-Digit Benford&#39;s Law Test and Recent American Presidential Elections.
  • Klimek, P., Yegorov, Y., Hanel, R. & Thurner, S. (2012). Statistical detection of systematic election irregularities. PNAS 109(41).
  • Stokes, S., Dunning, T., Nazareno, M. & Brusco, V. (2013). Brokers, Voters, and Clientelism. Cambridge UP.
  • Cantú, F. (2019). The Fingerprints of Fraud: Evidence from Mexico&#39;s 1988 Presidential Election. American Political Science Review 113(3).
  • Mainwaring, S. & Zoco, E. (2007). Political Sequences and the Stabilization of Interparty Competition. Party Politics 13(2).

This analysis is not a new survey but a re-analysis of existing public datasets through the lens of the international election-forensics literature.