Initial Report: Chicago Board of Elections — Machine Log Assessment 2020 & 2024

Benford's Law Temporal Analysis · Ballot Scan Checksums · Preliminary Findings

Prepared by: Benford Bench Analysis Project
Data Sources:

• Basic Election Checksums
• Preliminary Findings 2020 vs. 2024

Report Date: 2026-05-27
Elections Covered: November 3, 2020 · November 5, 2024
Status: Preliminary — pending precinct-level and device-level log verification; pending by-Ward Benford re-analysis of 2024 logs

1. Executive Summary

This report integrates two independent lines of quantitative analysis applied to Chicago Board of Elections (CBOE) Election Day machine logs for the 2020 and 2024 General Elections:

1. Benford's Law temporal analysis — examination of inter-event time intervals (Δt in seconds) between consecutive scanner log entries, tested against the expected first-digit logarithmic distribution.
2. Ballot scan checksum analysis — comparison of total machine-logged scan events (successful and failed) against the certified vote totals published in the official CBOE canvass reports.

The two analytical frameworks surface different and in some respects contrasting anomaly profiles for each election year:

• The 2020 election presents a stronger Benford signal for artificial log manipulation, with approximately 10 % deviation from the expected temporal distribution — more than double the 2024 deviation of ~4 %. Under the Benford model, 2020 exhibits the hallmarks of a system whose event timestamps were externally smoothed or corrected.
• The 2024 election presents more severe hardware/operational discrepancies: a failed-scan rate of 21.75 % (vs. 6.01 % in 2020), machine-log totals exceeding the certified count by 140.7 %, and an absolute excess of 644,325 scan events over the official report — 69.2 % more excess than in 2020.

Important methodological caveat: The two Benford analyses were not conducted under equivalent conditions. The 2020 logs were processed divided by Ward (each Ward's event stream analyzed independently), while the 2024 logs were processed with all Wards pooled into a single mixed stream. This difference has a direct and potentially significant impact on the number of qualifying data points that survive the 100-second inter-event filter, which in turn affects the statistical power of the chi-square test. The 2024 deviation figure of ~4 % should therefore not be read as a direct apples-to-apples comparison with the 2020 figure of ~10 % until a by-Ward re-analysis of the 2024 data is completed. Further by-Ward testing of the 2024 logs is pending.

Taken together, neither election's machine logs cleanly reconcile with its certified totals. The nature and likely source of anomalies, however, differs between the two cycles. 2020 shows the stronger statistical signature of log-level data manipulation; 2024 shows the greater operational breakdown or record-keeping failure at the scan-event level. The Benford comparison between the two years must be treated as provisional pending methodological alignment.

2. Analytical Framework

2.1 Benford's Law — Temporal Method

Traditional applications of Benford's Law to election data examine the leading digits of vote tallies per precinct. This analysis takes a methodologically stronger approach by examining inter-event time intervals (Δt) between consecutive log entries produced by ballot-scanning machines.

Why timestamps instead of vote counts?
Human-driven processes — voters feeding ballots, poll workers handling queues, machine throughput — generate naturally irregular time gaps that organically follow Benford-type distributions. Vote tallies, by contrast, are often artificially constrained by precinct size and are more susceptible to non-Benford behavior for legitimate structural reasons. Timestamp intervals are harder to fake consistently and are less likely to be directly tampered with by someone manipulating outcomes.

Methodology applied:

• Filtered micro-events below 100-second intervals to remove machine-internal noise
• Applied chi-square goodness-of-fit testing against the canonical Benford first-digit frequencies
• Used control datasets drawn from stochastic human behaviors (phone screen-wake event intervals) as baseline comparators
• Stratified controls for precinct size and time-of-day effects to remove confounding

⚠️ Ward-level vs. pooled-stream methodological difference:
The 2020 analysis was performed on log streams divided by Ward — each Ward's scanner events were processed as an independent time series before aggregation. The 2024 analysis was performed on a single pooled stream with all Wards mixed together. This is not a trivial distinction:

• When events from dozens of Wards are merged into one chronological stream, a "consecutive" event pair in that stream often comes from two different Wards whose scanners happened to fire close together in absolute wall-clock time. The resulting Δt can be milliseconds to a few seconds — far below the 100-second filter threshold — even though both underlying Ward streams individually had longer natural voter-arrival gaps.
• The 100-second filter therefore eliminates a substantially larger proportion of event pairs from a mixed-Ward stream than from per-Ward streams, leaving a smaller surviving pool of qualifying Δt values for the chi-square test.
• Benford's Law is scale-invariant — the magnitude of the values does not determine whether the law applies — but it is not sample-invariant. Chi-square goodness-of-fit power is a direct function of pool size (n). A smaller pool reduces the test's ability to detect a given level of deviation; the same underlying manipulation signal could produce a lower measured deviation percentage simply because fewer data points survived to carry it.
• Consequence: The 2024 ~4 % deviation figure may partly or wholly reflect a smaller qualifying data pool rather than genuinely cleaner logs. By-Ward re-analysis of the 2024 data is required before the two deviation figures can be directly compared.

Expected Benford first-digit distribution:

Deviation from this distribution in machine-log timestamps suggests that the natural randomness of voter arrival and machine processing was disrupted — either by mechanical issues, artificial batching, or deliberate log manipulation.

2.2 Ballot Scan Checksum Method

Every ballot scanner logs each feed attempt as a discrete event — whether the scan succeeds or fails. These event counts form an internal checksum that should, after accounting for re-feeds, spoiled ballots, and test records, reconcile approximately with the certified count. Material unexplained divergence between total logged scan events and the certified count constitutes a checksum failure.

3. Raw Machine-Log Data

3.1 — November 3, 2020 General Election

3.2 — November 5, 2024 General Election

Device clock note (per source): A subset of scanning devices carried incorrect internal date/clock settings during logging. Records falling outside Election Day bounds were identified and flagged; their inclusion or exclusion materially affects total scan counts.

4. Benford Analysis Results

4.1 What a 10 % Deviation Means for 2020

The 2020 Benford analysis was conducted on log streams segmented by Ward, meaning each Ward's event sequence was processed independently before results were aggregated. This approach preserves the natural intra-Ward voter-arrival rhythm as the signal of interest: consecutive events belong to the same physical polling location, and their Δt values reflect real gaps between individual voters feeding ballots.

A ~10 % deviation from expected Benford first-digit frequencies in these intra-Ward inter-event intervals indicates that the natural irregularity of voter throughput was suppressed or altered across a meaningful portion of the 2020 log record. The source authors characterize this as suggesting "artificial smoothing or correction" of the log stream — consistent with post-hoc log editing, batch-injection of scan events, or automated regularization of timestamps.

This is the stronger fraud-model signal between the two elections. The per-Ward methodology maximises the pool of qualifying Δt values (because intra-Ward consecutive events have naturally longer gaps that more easily survive the 100-second filter), giving the chi-square test its full statistical power. Under those conditions, the 2020 logs show strong evidence of log-level temporal manipulation.

4.2 What a 4 % Deviation Means for 2024 — and Why It Cannot Be Taken at Face Value

The 2024 Benford analysis was conducted on a single pooled stream combining all Ward logs together, without prior Ward-level segmentation. As explained in Section 2.1, mixing Ward streams in chronological order introduces large numbers of cross-Ward event pairs with very small Δt values (events from different Wards firing in near-simultaneous wall-clock time). These are removed by the 100-second filter, systematically shrinking the qualifying data pool relative to a per-Ward analysis.

A smaller qualifying pool reduces chi-square test power directly. The same underlying manipulation signal that would produce a 10 % deviation with a full per-Ward pool could produce a much lower measured deviation when the pool is depleted by cross-Ward noise filtering. The 2024 ~4 % figure therefore has two possible interpretations that cannot currently be distinguished:

1. Interpretation A — Genuinely less manipulation: The 2024 log timestamps were not artificially smoothed at the same scale as 2020, and the lower deviation reflects a real difference in log integrity.
2. Interpretation B — Pool depletion artifact: The mixed-Ward methodology removed enough qualifying data points to suppress the detectable deviation, and the 2024 logs may carry a comparable or larger manipulation signal that the current analysis lacks the power to measure.

Until the 2024 logs are re-run using the same per-Ward segmentation applied to 2020, Interpretation A and B cannot be separated. The lower 2024 Benford figure should not be read as an exoneration of the 2024 log record.

Critically, a lower Benford deviation in 2024 does not mean 2024 was clean — it means the type of measurable anomaly present in 2024 is different and, under current methodology, better captured by the checksum analysis in Section 5.

5. Checksum Analysis

5.1 — 2020 Discrepancy Breakdown

The 2020 machine logs record 669,485 total scan events against a certified total of 288,593 — a 2.32:1 ratio. Successful scans alone exceed the certified count by 340,638 (+118.0 %).

5.2 — 2024 Discrepancy Breakdown

The 2024 machine logs record 1,102,469 total scan events against a certified total of 458,144 — a 2.41:1 ratio. Successful scans alone exceed the certified count by 404,582 (+88.3 %).

6. Year-over-Year Comparative Summary

¹ 2024 Benford figure is not directly comparable to 2020. The mixed-Ward pooling methodology reduces qualifying data-point pool size through the 100-second filter, diminishing chi-square test power. By-Ward re-analysis of 2024 logs is pending.

7. Synthesis: Two Different Anomaly Profiles

The most important finding of this combined analysis is that the two elections exhibit structurally different anomaly signatures, which carries distinct implications for what may have occurred in each cycle.

7.1 — 2020: Benford-Signal Anomaly (Log-Level)

The 2020 Benford deviation was measured using per-Ward segmented log streams, which is the methodologically sounder approach: each Ward's events are compared against themselves, preserving the natural intra-location voter-arrival signal. This methodology maximises the number of data points surviving the 100-second filter, giving the chi-square test its full power.

The resulting 10 % departure from expected Benford first-digit frequencies is statistically significant and is more consistent with artificial batch insertion, timestamp regularization, or post-hoc log editing than with any hardware failure. The per-Ward division means this signal cannot be explained away as a cross-Ward pooling artefact — the deviation is real within each Ward's own stream.

The failed-scan rate (6.0 %) is elevated above industry norms (~2–3 %) but is not extreme enough on its own to indicate equipment crisis. The overcount relative to the official report is large in both absolute (+380,892) and proportional (+132 %) terms.

Assessment: Under the Benford model, the 2020 logs present the stronger prima facie case for data-level manipulation of the machine log stream. The per-Ward methodology makes this the more analytically robust of the two Benford measurements.

7.2 — 2024: Operational/Checksum Anomaly (Hardware/Record Level)

The 2024 logs present a more complex picture. The Benford deviation appears lower at ~4 %, but this figure was derived from a pooled mixed-Ward stream. When all Wards' events are combined in chronological order, many consecutive event pairs in that stream originate from different Wards firing within seconds of each other in wall-clock time. The Δt for those cross-Ward pairs is frequently under 100 seconds and is removed by the filter, shrinking the qualifying data pool and reducing chi-square test sensitivity.

Because Benford's Law is scale-invariant but not sample-invariant, a depleted pool can suppress measured deviation even when the underlying log-level signal is present. The 2024 Benford figure therefore cannot be reliably compared to the 2020 figure until the 2024 logs are re-run with per-Ward segmentation. By-Ward testing is pending.

What the 2024 data does show clearly — in the checksum domain — is a catastrophically elevated failed-scan rate of 21.75 % (239,743 events), nearly 8–10× the acceptable industry threshold, and a total-scan-to-official ratio of 2.41:1. A genuine equipment failure of this scale across Chicago's precincts would have been visibly catastrophic and widely documented in real-time incident reports. Its apparent absence is a significant anomaly in its own right.

Assessment: The 2024 checksum anomalies are the most severe in absolute terms. The Benford comparison with 2020 must be treated as provisional: the mixed-Ward pooling methodology may have suppressed the 2024 Benford deviation by depleting the qualifying data pool through the 100-second filter. Whether a per-Ward 2024 analysis would reveal a Benford signal comparable to 2020 is an open question that the pending re-analysis will address.

8. Possible Explanatory Factors

The following are documented technical causes that election auditors investigate when machine-log totals exceed certified counts. They are listed as investigative hypotheses and are not conclusions.

9. Anomaly Assessment: Failed Scan Spike (2024)

The 2024 failed-scan total of 239,743 is the single sharpest numerical anomaly in the combined dataset:

• It represents a 495.6 % increase over the 2020 figure of 40,254.
• At 21.75 %, every 1-in-5 scan attempts in 2024 registered as a failure.
• Industry-standard acceptable scanner failure rates are below 2–3 % for properly maintained equipment.
• This rate, if real and hardware-driven, would require that approximately 239,000 ballot feed failures occurred across Chicago's precincts on a single day — a crisis that should have been visibly catastrophic and widely reported.
• The Benford deviation in 2024 (~4 %, from a pooled mixed-Ward stream) is lower than 2020 on its current measurement — but as noted in Sections 2.1 and 4.2, the mixed-Ward pooling methodology reduces the number of qualifying data points through the 100-second filter. This makes it impossible at present to determine whether the lower deviation reflects genuinely cleaner timestamps or a smaller analytical pool with reduced chi-square power.

This combination — an implausibly high failure count with a provisionally lower Benford deviation — is consistent with multiple scenarios: scanner firmware logging re-scan attempts as discrete "failed" events, test records not purged from device memory, systematic incorrect device-state flags, or a manipulation signal present in 2024 that the current pooled-Ward analysis lacks the statistical power to fully surface.

10. Checksum Integrity Verdict

¹ 2024 Benford result is provisional. Mixed-Ward pooling reduces qualifying pool size through the 100-second filter, limiting chi-square power. By-Ward re-analysis pending.

Neither election's machine logs reconcile with the certified report within any reasonable margin. The anomaly type differs: 2020's principal red flag is its Benford temporal signature measured under the more robust per-Ward method; 2024's principal red flag is its failed-scan count and total overcount magnitude. The Benford comparison between the two years is provisional pending methodological alignment.

11. Preliminary Conclusions

1. The 2020 election shows stronger evidence of log-level data manipulation under the Benford model. A ~10 % deviation from expected inter-event timestamp distributions — measured using per-Ward segmented log streams — is statistically significant and consistent with artificial modification or injection of scan event records. The per-Ward methodology is the more analytically sound of the two approaches used and gives this finding its full chi-square power.

2. The Benford comparison between 2020 and 2024 is not methodologically equivalent and must be treated as provisional. The 2020 analysis used per-Ward segmentation; the 2024 analysis used a pooled mixed-Ward stream. Because Benford's Law is scale-invariant but not sample-invariant, mixing Ward streams in chronological order increases the number of cross-Ward event pairs eliminated by the 100-second filter, shrinking the qualifying data pool and reducing chi-square sensitivity. The 2024 deviation of ~4 % may reflect a smaller qualifying pool rather than genuinely cleaner logs. By-Ward re-analysis of the 2024 logs is pending and is required before the two Benford figures can be directly compared.

3. The 2024 election shows more severe operational anomalies in the checksum data. A 21.75 % failed-scan rate, 495 % higher than 2020, and a total-scan-to-official-count ratio of 2.41:1 indicate either a catastrophic and unreported equipment failure or systematic over-logging in the scan event record.

4. The two anomaly profiles are not mutually exclusive. It is possible for 2020 logs to have been manipulated at the log structure level while 2024 suffered from different hardware or software-level failures that inflated scan counts. It is equally possible that 2024 contains a Benford-detectable manipulation signal that the current pooled-Ward analysis lacks the power to measure — a question the pending by-Ward re-analysis will help resolve.

5. Neither election's machine logs constitute a self-certifying record. In both cycles, total logged scan events exceed the official certified total by more than 2:1, a ratio that cannot be explained by re-feeds and test records alone without further documentation from CBOE.

6. These findings are preliminary. They are derived from aggregate totals under non-equivalent methodological conditions; precinct-level, device-level, and properly aligned by-Ward Benford analysis is required before any definitive conclusions about the source, geographic distribution, or systemic nature of the anomalies can be drawn.

12. Recommended Next Steps (for others: I am exhausted)

1. ⭐ Priority — By-Ward Benford re-analysis of 2024 logs — Re-run the 2024 temporal Δt analysis using the same per-Ward segmentation applied to 2020. This is the single most critical methodological step needed to make the two Benford figures comparable. Segment 2024 log events by Ward before computing consecutive Δt values, apply the 100-second filter within each Ward stream, and record qualifying pool sizes alongside deviation percentages for both years. Compare pool size (n) between the pooled-Ward and per-Ward approaches to quantify how much the current mixed-stream methodology depleted the 2024 data.
2. Pool-size sensitivity analysis — Deliberately down-sample the 2020 per-Ward pool to match the 2024 pooled-stream pool size and re-run the chi-square test. If the measured deviation drops significantly with a smaller pool, this confirms that pool depletion is a material confound in the current 2024 figure.
3. Device-level log audit — Obtain per-scanner event logs from CBOE to identify whether Benford deviations and overcount are concentrated in specific machines, precincts, or ward clusters.
4. Timestamp audit — Cross-reference all scanner timestamps against official poll-open/close windows (6:00 AM – 7:00 PM) and isolate records from devices with incorrect clock settings.
5. L&A test purge verification — Confirm that all Logic & Accuracy test feeds were zeroed on all deployed units before Election Day in both cycles; compare 2020 vs. 2024 purge logs.
6. Failed-scan incident report cross-reference — Obtain precinct-level poll inspector exception reports for 2024; match reported equipment incidents against the geographic distribution of failed scan events.
7. FOIA request — File FOIA with CBOE (Illinois FOIA, 5 ILCS 140) requesting: (a) full ballot accounting reconciliation for 2020 and 2024, (b) scanner make/model/firmware version inventory, (c) L&A test records and purge certifications, (d) any incident or exception reports filed by precinct judges on Election Day.
8. Extended Benford analysis — Once by-Ward segmentation is applied to both years, extend the analysis to the precinct level to identify geographic clustering of Benford anomalies within each Ward.
9. Cross-county comparison — Apply the same checksum and Benford methodology to Cook County suburban precincts and Illinois downstate results to test whether the anomalies are Chicago-specific or statewide.

13. Data Sources & Limitations

End of Initial Report