Status: 159 is the current verified working count. The remaining net difference from the original 162 has not yet been fully reconciled. This report preserves uncertainty rather than forcing the two methods to agree.
What was counted
The unit is a separate arrestee record whose ADDRESS field is counted as 391 Ellis. Charge-continuation or ditto rows are not separate arrestees. Literal source transcription is kept separate from analytical interpretation, so unusual entries such as 390 ELLIS or 394 ELLIS remain visible as written even when treated as address-error exceptions.
Monthly pattern
The current audit shows a marked increase beginning around May. January–April contain 53 records, averaging 13.25/month. May–August contain 106 records, averaging 26.5/month—exactly twice the earlier monthly average. These are counts of arrest records listing 391 Ellis, not unique people and not rates normalized by total arrests.
January
10
February
14
March
16
April
13
May
28
June
24
July
24
August
30
Month
Current verified
Original first pass
January
10
13
February
14
6
March
16
16
April
13
14
May
28
30
June
24
27
July
24
25
August
30
31
Total
159
162
Audit history
Stage
Total
What changed
Original first pass
162
Broad Jan–Aug count; not preserved as a complete itemized ledger.
Rigorous visual review
156
ADDRESS-column-focused manual review, with handwriting analysis and continuation-row controls.
Independent rescan
159
Recovered two January p.60 arrestees and one May p.101 arrestee.
Embedded CSV downloads
This HTML contains CSV data internally, so the buttons work offline. The first file is a checkpoint evidence table of key examples and exceptions; it is not falsely presented as a complete 159-row ledger. The second is the full monthly summary.
Curated handwriting & audit examples
These larger contextual crops are selected to show the specific problems encountered during the audit—not just examples of the address itself. They retain the NAME/ADDRESS row context while avoiding unnecessary columns.
Clear control sample
May p.166
A clear handwritten 391 ELLIS entry used as an easy-reference control.
Difficult handwriting
February p.79
A difficult-handwriting example confirmed during the manual review.
Difficult handwriting
July p.235
July handwriting example previously flagged as difficult but confirmed.
Address-error exception
March p.51
Literal entry appears as 394 ELLIS ST; preserved literally and counted analytically as a 391 Ellis address-error exception.
Address-error exception
April p.61
Shows literal 391 ELLIS on the preceding row and 390 ELLIS on the next row, preserving the basis for the exception decision.
Independent rescan recovery
January p.60
Two separate 391 ELLIS arrestees on the same page, both recovered during the independent rescan.
Independent rescan recovery
May p.101
Literal 391 ELLIS ST record recovered during the independent rescan.
Continuation-row control
May p.215
Illustrates the record where the following line is the same-person continuation and is not counted separately.
Multiple arrestees on one page
August p.238
Shows three separate 391 Ellis arrestees on a single arrest-log page.
The following ground-truth samples are carried forward from the earlier self-contained handwriting-recognition layer. They demonstrate why visual review was necessary and preserve the source image alongside the confirmed transcription.
#1
391 ELLIS
Page 238 · confirmed
#2
391 ELLIS ST
Page 238 · confirmed
#3
391 ELLIS ST
Page 238 · confirmed
#4
391 Ellis St
Page 1 · confirmed
#5
391 ELLIS ST / SF CA 94102
Page 3 · confirmed
How counting errors happen
Example
Classification
Why it matters
January p.60 lines 1 & 2
rescan recovery
Two separate arrestees on a page missed during the rigorous manual pass and recovered independently.
May p.101 line 1
rescan recovery
A literal 391 ELLIS ST entry recovered during the independent rescan.
March p.51 line 1
address-error exception
Appears written 394 ELLIS; literal value is preserved while the audit counts it analytically as 391 Ellis.
April p.61 line 3
address-error exception
Appears written 390 ELLIS; literal value is preserved while the audit counts it analytically as 391 Ellis.
May p.215 lines 4–5
continuation control
Line 4 is counted; line 5 is the same person/continuation and is not counted again.
August p.238 lines 1, 7 & 8
multiple arrestees
Three separate 391 Ellis arrestees appear on one page—the single-page record observed during the audit.
Current checkpoint: the full January–August corpus has undergone the rigorous visual pass and a subsequent independent rescan. The next reconciliation task is to compare surviving first-pass page-level audit artifacts against the current ledger to explain the remaining 162-versus-159 difference.
