Ballot Information File (bif) Report (Metadata Report)

San Francisco, California, 2024 General Election

Audit Engine version:'v2.32.X'

Run Date: 2026-10-26 17:03:25.917601 Copyright (c) Citizens' Oversight

Return to Master Report Index

Ballot Metadata Report Introduction

This report is a reconciliation of all metadata captured from various sources. The number of sources included may vary, based on the workflow and availability of image archives and the CVR. This report is essential when performing public oversight workflow audits, because it allows audit conductors to examine metadata to understand ballot style assignments, repeated ballots by ballot_id and repeated images, and to compare across all sources, images, CVR and any aggregated results. Each section of this report will have a separate introductory paragraph.

Table of Contents

• High Level Reconciliation
• Metadata Summary
• Sheet Count Summary
• Ballotids in BIA but not in CVR -- Precinct Report
• Ballotids in BIA but not in CVR -- Examples
• Card code: image vs CVR
• Card code analysis (use_card_code_as_style_num)
• Small Styles Report
• Hexstyle to style_num analysis (use_hexstyle_as_cvr_style)
• Summary Report by precinct
• Summary Report by batchid
• Summary Report by tabulator
• Ballots with alignment errors or unreadable barcodes
• cvr_dups_list ballot_ids
• bia_dups_list ballot_ids
• Ballot Image Archives Summary
• Ballot Image File Size Extremes
• Sheet Sizes

High Level Reconciliation

Metadata Summary

The Unprofiled column counts all data as delivered. The CVR may cover the whole county, while the ballot images cover only part of it, so the two need not match there. The Profiled column counts only the ballots selected for this audit, the same set used in the rest of this report. Official ballots cast is summed from official results for the audited scope only, such as the precincts covered. So it is compared in the Profiled column, and shows N/A in the Unprofiled column.

Sheet Count Summary

Card code: image vs CVR

card_code is read from the ballot image. cvr_card_code is the code the CVR reports. Where they differ, the image value is used.

Ballotids in CVR but not in Image Archives

This report can provide hints about why the ballots images are missing.

Ballotids in CVR but not in BIA by cvr_file

Ballotids in CVR but not in Ballot Image Archives, Examples

First, 50 of 1470486 cases.

Created missing_images report: 'reports/CA_SanFrancisco_20241105_pilot_d05_missing_images.csv'

Card code analysis (use_card_code_as_style_num)

Important: style_nums with multiple hexstyles (needs investigation)

hexstyle is normally determined solely from the CVR (one contest-set per ballot_type_id). A style_num below has more than one distinct hexstyle among its ballots -- something other than ballot_type_id is causing the contest-set to vary (e.g. BMD vs nonBMD, or a sheet-specific vs full-ballot computation) and needs investigation before this style_num can be trusted for mapping.

Remaining style_nums (single hexstyle)

Small Styles Report

This report provides an analysis of the number of ballots that are included in "small styles" and therefore, there is a threat that the voter could be linked to their ballot. For purposes of this analysis, a "Small Style" is defined as a style with fewer than k ballots in it. The worst case is if there is only one ballot in the small style, because that voter could be linked to their ballot if other information were cross-references with the cast vote record data. AuditEngine explicitly does not link voters to their ballots.

Very commonly, the reason for small styles is due to handling the federal-only ballots, which tend to be UOCAVA (Uniformed And Overseas Citizens Absentee Voting Act) voters. The privacy of these ballots can be easily improved by NOT splitting them by precinct, but instead by including them in one large virtual precinct.

• For San Francisco, California, 2024 General Election
• anonymity_parameter_k = 10.
• There were 0 BMD ballots that did not have a style_num in the cvr and were not examined during full_bif stage.
• There were 0 ballots with no style_num specified, likey because the images or cvr records are missing.
• There were 500 small styles detected, affecting 2,192 ballots.
• There were 49 split styles with multiple precincts per style.
• There were 298 mergeable styles, affecting 1,479 ballots.
• If merged, 202 small styles, affecting 713 ballots would remain.
• Ballots anonymizable by merging = 1.6429% of all 34,541 ballots

Summary for unified table

Small Styles