Past studies of hand counts show each team averages 1/10 to 1/4 minute per vote, including time to resolve discrepancies, so 24-60 ballots per hour per team on ballots with 10 votes. Costs depend on pay and number of staff per team, plus training and summation time.

Past studies show error rates as low as 0.2% (lower in India) in experienced jurisdictions, higher for first time efforts. Tallying undervotes and overvotes as well as candidate votes, means total votes are the same in each contest, which gives a consistency check to help find errors. Consistency can also be helped by having 1 reader, 1 person watching the reader, and 2-3 people tallying on separate tally sheets. The talliers compare counts every 20-40 votes and resolve differences. Machine counts are more accurate when they work well, and require audits of all contests, to know if they work well.

Tally many contests on one page, to minimize flipping pages in big elections, labeled to avoid ambiguity

Tally about 8 contests (25 candidates) on one page, first 100 votes, then start new page

This is an 11 x 8.5 page, with spaces for write-ins, overvotes, and undervotes. The image of the tally page was created and names can be entered in a spreadsheet. When more than one write-in candidate is being counted, you can add a line for each or track them on a separate page. If you can print 11 x 17 ledger paper, you can use the extra space for more contests or more votes

When any candidate fills up all 100 tally marks, all candidates start on a new page, to minimize flipping back and forth among pages. The ballots which were tallied on the old page can be tied together in case the sheet needs to be checked later.

For contests which always appear together on a ballot sheet (e.g. state-wide and countywide contests), the space at the bottom left of each office is useful. Enter the total of all tally marks from just this page for this office. Since the tallies include undervotes, the totals in these common offices will be the same. Differences can be checked while the ballots are still close at hand.

Numbers 5, 10, etc, are at the top of each line, to remind us they reflect a full box. When there are just 2 tally marks in the box between 80 and 85, they mean 82.

Tally 16- 32 contests on one page (50-100 candidates) , first 40 votes, then start new page

This approach covers a lot of candidates, with as little flipping across pages as possible. Below is an 11 x 8.5 page or the left half of an 11 x 17 page. Each 11" x 4" column has room for about 8 contests, depending on the number of candidates in each, plus spaces for write-ins, overvotes, and undervotes. On 11 x 17 there would be 4 columns of names. The image of the tally page was created and can be changed in a spreadsheet. When more than one write-in candidate is being counted, you can add a line for each or put them on a separate page.

When any candidate fills up all 40 tally marks, all candidates start on a new page, to minimize flipping back and forth among pages. The ballots which were tallied on the old page can be tied together in case the sheet needs to be checked later.

Numbers 5, 10, etc, are on the right edge of each space, to remind us they reflect a full box, when there are just 2 tally marks in the box marked 10, they mean 7, not 12.

Tally 25 candidates per page, unlabeled, no overvotes or undervotes

“Cause of America” prefers candidates be numbered on the ballots, so reader can call out the number for talliers to enter on sheet. New sheet for each 25 candidates. Video. Manual covers how these would be copied to summary sheet and added with a calculator.

Tally few contests on one page, or many contests on many pages, labeled to avoid ambiguity

2 contests per page (12 candidates) 11 x 17, Cross out numbers San Diego, CA 2016

1% audit by county. Includes write-in total & over & undervotes. Video shows them flipping through many pages

Judicial recount handbook, 11 candidates per page, 250 votes, mark boxes which are numbered every 5, without ambiguity, Canada undated.

https://www.elections.ca/content.aspx?section=res&dir=loi/jud&document=appD&lang=e

Recount by Senate contractors, 2 contests, 5 ballots per line, total after 50 ballots, Maricopa County, AZ 2021

It is hard to add the totals accurately after 50 ballots, since the number for each candidate on each line varies.

• "Each counter will place a single line (tally) in the Ballot column indicating that 1 ballot has been viewed.

• "Counters will view the paper ballot and place a single line in the column that corresponds to the candidate selected.

• "An image of the ballot will be displayed on the monitor above the counter’s desk area that can be used to view a larger version but must be confirmed on the paper ballot.

• "All tally marks will be made in the same row as that ballot number. After 5 ballots, the tally marks will be made in the next column down on the page." (p.4)

So: for example if first ballot is for Trump and second for Biden, second tally mark can go anywhere in Biden section , not necessarily in 1^st or 2nd place within that section. If second ballot were always put in 2^nd position, and 3^rd ballot of each group in 3^rd position etc., the sheet would display the content of each ballot, as well as being a summary. However following that strict order would be error-prone. The examples farther below for Orange County or for tracking 300 candidates have formats which are less error-prone.

https://web.archive.org/web/20220202104154/https:/www.cyberninjas.com/static/20210429155650/Wake-TSI-Counting-Floor-Policies.pdf

3 write-in candidates in 1-3 contests, 100 votes, cross out numbers, Maricopa County, AZ 2020

Instructions are on pages 222-234. Forms are on pages A233-A248. There are tally sheets like this only for write-ins and accessible machines; otherwise staff just record final tallies for each candidate.

https://azsos.gov/sites/default/files/2019_ELECTIONS_PROCEDURES_MANUAL_APPROVED.pdf

1 contest (2 candidates), 100 votes, cross out numbers, Harris County, TX 2006

Recount?

1 contest (2 candidates), 825 votes, numbered groups of 5 require multiplication or lookup in a table, Indonesia 2014

Tally few contests on one page, or many contests on many pages, with ambiguous labels

1 contest (3 candidates), 280 votes, numbered in a way which caused confusion, Jeffersonville, IN 2016

Errors were 3% to 27% for various candidates in a 2016 Indiana recount in a city. Centering the 5, 10, etc. above the columns misled officials that one hash mark in the box marked 135 meant 136, while it actually meant 131. Also officials sometimes omitted absentee ballots or double-counted ballots

https://www.newsandtribune.com/news/jeffersonville-city-council-at-large-recount-tally-sheets-show-vote/article_75f432ce-cf7e-11e5-8c1a-5365ef7d3540.html

3 contests (10 candidates), 50 votes, Rhode Island test

Centering the 5, 10, etc. above the columns can mislead officials that two hash marks in the box marked 10 mean 12, while it actually means 7. This was the pattern in Jeffersonville above.

https://verifiedvoting.org/wp-content/uploads/2020/07/RI-RLA-Report-2020.pdf

2 tally pages per contest (2-6 candidates), 300 votes, Los Angeles County, CA 2020

With only 5 sections per page, and 4 sections needed for write-ins, overvotes, undervotes, and signatures, each contest other than top-two primaries and yes/no (which don't allow write-ins) needs 2 pages. Staff have to flip through many dozens of pages for the many contests in LA County. Batches of ballots in LA come from the mail or vote centers (LA does not use precincts), so each batch can include any kind of ballot from all over the city, and the tally pages must cover them all. A third page, not shown, is needed when tallies exceed 300.

Risk-limiting Audits (RLAs) which track each ballot separately

Track each ballot separately for an RLA, 5 contests (14 candidates), 20 ballots per page. Orange County, CA 2018

This was a ballot polling risk-limiting audit (RLA) of 5 primary contests with wide margins. They did not tally write-ins, overvotes or undervotes. Report is at https://verifiedvoting.org/publication/orange-county-ca-pilot-risk-limiting-audit/

Track each ballot separately for an RLA, 16-32 contests (50-100 candidates), 10 ballots per page, to minimize flipping pages in big elections

Different letter for each ballot helps staff stay in the right column.

This is an 11 x 8.5 sheet or the left half of an 11 x 17 sheet. Each 11" x 4" column has room for about 8 contests, depending on the number of candidates in each, plus spaces for write-ins, overvotes, and undervotes. On 11 x 17 there would be 4 columns of names. When more than one write-in candidate is being counted, you can add a line for each or put them on a separate sheet.

A “ballot comparison” is a type of “risk-limiting audit” (RLA), where the votes need to be recorded for each ballot separately, not just added up across all ballots. The tally sheet below provides a column to keep each ballot separate, using a different letter for each column.

The tally sheet can track many candidates, since there is a proposal in arxiv.org/abs/2012.03371 for RLAs to record all candidates. (Colorado currently records all candidates, but not on paper. They enter result of each contest into software.) The image of this tally sheet was created and can be changed in a spreadsheet.

RLAs outside Colorado usually tally only 1-2 contests and their candidates, so will use only one column. With one column of contest & candidate names there will be room for about 25 ballots per tally sheet.

The alternative of entering data directly into a computer, then printing out and checking, has a different error profile and time needs.

The above tally sheet is designed to tally ballot-comparison information for 10 ballots, with about 6 lines per inch, 60 lines in each column, 10 inches long, with two columns on 8.5 x 11 inch paper (120 lines) or five columns on 11 x 17 inch ledger paper (300 lines). If more lines are needed, tally staff are also familiar with flipping across multiple ledger size sheets when needed (https://youtu.be/lhKNS3v2fg4).

Using a distinct letter for each ballot on the page lets data entry staff use one data entry column per sheet, shown here:

Table below shows number of lines on tally sheet in selected counties & elections

Tally sheet needs a line for each candidate being tallied, plus 3 extra lines per contest, for contest title, write-ins (except on yes/no) and “no vote marked.”

Format above can handle about 300 lines on one side of one page, ledger size, in 5 columns of 60 lines each, 6 lines per inch. This means a one-sided leger page handles most counties, and allow more space

# Contested #

Lines Contests Candidates County

1407 201 804 AZ_Maric_2020g~

1063 166 565 CA_Orang_2020g~

486 48 342 CA_Orang_2020p~

518 87 257 CA_Ventu_2020g~

134 26 56 CO_Adams_2020g~

134 25 59 CO_Arapa_2020g~

178 33 79 CO_Bould_2020g~

181 35 76 CO_Denve_2020g~

100 16 52 CO_Dougl_2020g~

207 37 96 CO_El Pa_2020g~

156 29 69 CO_Jeffe_2020g~

123 24 51 CO_Larim_2020g~

195 34 93 CO_Weld__2020g~

426 49 279 MN_Henne_2021g~

1714 141 1291 PA_Alleg_2019g~

2406 139 1989 PA_Alleg_2019p~

168 24 96 PA_Alleg_2020g~

50 6 32 PA_Merce_2018g~

616 13 577 PA_Merce_2018p~

Source: “pivot” tab of spreadsheet

Sort & Stack needs batch totals without tally marks

Australia uses sort & stack with single transferable vote https://aec.gov.au/Voting/counting/hor.htm

France: one person reads to 2 talliers: https://www.interieur.gouv.fr/Archives/Archives-elections/Comment-voter/Fonctionnement-d-un-bureau-de-vote

UK uses sort & stack https://theconversation.com/explainer-how-britain-counts-its-votes-41265

Batch totals for 1 contest (3 candidates), 10 batches, from sort & stack, Georgia 2020

Georgia hand count in 2020 used "sort & stack" method, so they only kept totals for each batch. Goggin et al had found sort & stack less accurate than tally marks

Batch totals for 1 contest (3 candidates), 4 batches of 50 ballots, from sort & stack, New Hampshire 2007

New Hampshire hand counts recommend "sort & stack" method for about 50 ballots at a time, to catch discrepancies early, so they keep totals for each 50 ballots. Goggin et al had found sort & stack less accurate than tally marks, but did not break it into 50 ballots at a time

https://web.archive.org/web/20080919152131/http://www.electiondefensealliance.org/files/Hand_Count_Elections_Steps_only_Sept_6_2007.pdf

Other special purpose tally sheets

Single Transferable Vote, first two rounds, 1 contest, 5 candidates, Wisconsin 1912

"Following, is the official tally sheet upon which the first and second choice votes are entered, being called off in this case: 'For Governor, Brown first, Gray second.' Supposing the ballots of one precinct have been called off, the sheet shows as follows:--

https://digicoll.library.wisc.edu/WIReader/WER1650-Chpt4.html

Separate tallies of ambiguous and clear votes, 1 contest (3 candidates), unnumbered groups of 5, Bridgeport, CT 2010

Recount by citizens:."One individual read each ballot, while a second individual watched the ballot to make sure the reading was accurate. Two individuals completed identical hashmark sheets for each stack - if the individuals sheets did not agree, then the team resolved the issue by recounting the ballots. Hashmarks were made for each ballot, including undervotes and overvotes, such that the number of hashmarks could be cross checked to equal the number of ballots in the stacks. Stacks were clipped with metal clips, along with one of the hashmark sheets for resealing, while the other hashmark sheet was saved for the district recount paperwork for processing by the accounting team. Special stacks such as questionable ballots (those with partially filled in bubbles), write-ins, and voter intent issues were scrutinized with the assistance of recount leaders to assure accuracy and consistency." [Pages 15-17 of the Recount Report describe the process and lessons learned.]

http://www.ctelectionaudit.org/bridgeport-recount-details/