After going through the "Big Binder" and summary documents and poring over dozens of poll books, my conclusion is that the polling place ballots are an irreconcilable mess.

I've created this spreadsheet to summarize the Binder documents [see the extended entry for explanations of the columns]

My main observations and conclusions

* The starting point for the ballot/crediting discrepancy is the net 1,660 polling place and provisional ballots in excess of credited voters. (This does not include the discrepancy in the absentee votes). I can explain a number of the discrepancies, but I've yet to find any feasible explanation for "ballotless voters". So I'll concentrate on the polling places where there are more ballots than voters. Because of all of the difficulties, I can only give a rough range of about 1,300 to 2,100 illegally-cast or otherwise unexplainable polling ballots. This comes from an estimated 628 absentee voters who should have cast a provisional ballot but instead received and cast a regular ballot, and upwards of 700 provisional ballots that were issued and ended up in the Accuvote before they were verified. There are also untold other reconciliation irregularities for which there is no apparent explanation.

MORE --

Refer to the Binder documents for the rest of the discussion -- Summary Reconciliation and Notes, and also the Worksheet of the 348 Provisionals .

Some of the problem ballots were undoubtedly cast by eligible registered voters who failed to follow the rules or were given incorrect instructions by the pollworkers. It would take a careful recanvass of all 330,000+ polling place ballots to determine how many of these ballots were legitimate, how many would have been legitimate if the voter followed the rules, and how many were downright fraudulent.

* Some of the discrepancy really is attributed to post-election crediting error. For one polling place in particular, Rainier Free Methodist Church, the roughly 36 voters in the first 15 pages of the poll book were never credited for voting. (As an example, see the first page of the pollbook and go to the Sound Politics voter database to verify the voters who signed in). This large of a crediting discrepancy does not seem to be typical.

* The single biggest problem with the reconciliation is that the reports released by the county don't show the number of signatures in the polling portion of the poll books. If we had these numbers, we could do a more reasonable job of tallying true crediting discrepancies (signatures - credits) and bogus ballots (AccuVote count - signatures). Canvassers counted signatures and recorded the numbers in the actual pollbooks. If King County compiled these numbers into a single document it would be to their advantage to release it. But it's not feasible for me to go through 2,600 pollbooks to gather these numbers.

* It is also difficult to get an exact count of the "signatures". I've found a number of cases where pollworkers didn't follow the required procedures for issuing ballots. In some cases, signatures are missing; ballot serial numbers are missing; people who should have received provisional ballots signed the page for the issuance of regular ballots; in some of the latter cases, they appear to have been issued regular ballots, in other cases they appear to have received provisional ballots after all; voters with spoiled ballots were sometimes given provisional ballots when they simply should have been given a replacement ballot and the spoiled ballot set aside; It's not always possible to tell how many ballots of each type were issued and returned. But a best-effort signature count by precinct would be better than no count at all.

* The number of absentee voters who should have cast provisionals but instead cast regular ballots appears to be about 628. See the discussion of spreadsheet Columns AE and AI. We won't know whether these people also cast their absentee ballots until we get the Absentee Ballot Audit Trail.

* What does "No label" mean? A correctly cast provisional ballot would show up as an envelope in the ballot machine side pocket with an attached removable sticky label. When the provisionals are first canvassed, the labels are removed from the envelopes and placed on sheets by polling place. An example label sheet is posted here. The Binder documents give a count of "no label" provisionals by precinct. This indicates the difference between the number of provisional ballots issued and the number of labels that showed up on the label sheets. A few different scenarios could explain a "no label" -- a provisional ballot that the voters takes out of the polling place without casting it, a voter who removes the label from the envelope in the few steps between getting the envelope from the poll worker and placing the envelope in the side pocket of the ballot box; or most commonly, a provisional ballot that goes into the AccuVote without the envelope. In polling places where the number of AccuVote ballots exceeds the number of signatures and the "no label" number is close to the difference, unverified provisionals would seem to be the best explanation.

The following explains the columns in the spreadsheet --
The two leftmost columns after the polling place show the number of voters at each polling place crediting with voting at the polls and also the number of polling ballots at the polling place from the machine recount (very close to the final number). The first yellow columns are taken from the Binder Summary document; the blue columns are from the Binder Notes.

*Column L "PB348 PBAV" -- As I noted the other day, the so-called "348 unverified provisional ballots" was never 348. Adjusting just for the likely data entry errors (rows where the number of acknowledged ballots is less than the number of explained ballots, I get 376)

* Column O "CredDiffRaw" -- the sum of ballots counted + add-on ballots - credits for the polling place. [see this post for significance of "add-on"]

* Column Z "Ack PB" -- these are the unverified provisionals that King County is implicitly acknowledging -- the maximum of Column L and the "Adjusted" number they give in the Reconciliation Summary. [The Adjusted number is supposed to have included all of the unverified provisionals, but in practice it didn't). I get 718 implicitly acknowledged unverified provisionals.

* Column AE "Prov Diff" -- this is the discrepancy by polling place between provisional ballots counted and provisional voters credited. A negative discrepancy is likely to have been caused by an absentee voter who tries to vote at a polling place and is supposed to receive a provisional ballot, but is instead given a regular ballot and signs the regular place in the poll book. Their ballot is counted as a polling ballot, but in the cases I've looked at, the voter is credited as provisional.

* Column AF "AdjCredDiff" -- the ballot/credit discrepancy adjusted for the likely number of absentee voters who voted regular ballots (Column AE).

* Column AG "Capped NoLabel" -- the number of "no label" ballots identified per polling place. Many but not all of these are likely to be unverified provisionals. The number of "no labels" [column I] is capped by the Adjusted Credit Diff [Column AF]

* Column AH "Minimum Likely Unverified PB" -- If King County identified a specific number of Unverified provisionals for a polling place [Column Z], take that number, otherwise take the Capped NoLabel number [Column AG]

* Column AI "Absentee At Polls" -- Estimates the number of absentee voters who voted a regular polling ballot. The maximum of absolute value of Prov Diff [column AE] and Credit Discrepancy.

I don't guarantee perfection. Feel free to report errors and I'll do my best to correct any actual errors.