| What a secondary stamp is | A Bates number printed on a sheet by an earlier custodian, standing above the number the Department of Justice release gives that sheet. A single sheet can carry three numbers from three different series12 |
| Series read at the page image for this page | EFTA_ and SDNY_, which appear together on the same sheets one number apart, and EFTA_R1_, a separate run marked by the R1 element in the stamp: these three are printed ink. The Department’s own EFTA number is not ink. On all twenty-three sheets read here it is set as live type in a white band the producer added beneath the scan12 |
| Sheets read at the page image | 23. The seven sheets of EFTA00134541 and the sixteen sheets of EFTA01810270, every one at the native 96 pixels per inch12 |
Where the EFTA_ run begins | At EFTA_00000001, printed at the foot of the first sheet of EFTA00134541, a Bureau of Prisons record. The same sheet carries SDNY_00000002, one number ahead, and the two series stay exactly one apart across all seven sheets1 |
| Why the stamps matter | Each series is consecutive and runs one number to one sheet, so a position no sheet can be matched to is a break in somebody’s numbering run. The breaks can be counted, and the count is a ceiling3 |
| What the released text does with them | Of the 30 earlier custodians’ stamps printed as ink on the 23 sheets read at the image, 14 come through exactly: 4 of 7 for SDNY_, 6 of 7 for EFTA_, and 4 of 16 for EFTA_R1_. The Department’s own number is in the text of all 23 sheets, but it was never scanned and so was never read12 |
| Positions unaccounted for | At most about 1.49 million across the two series that can be counted this way, re-derived over 2,925,025 sheets on 25 October 2026. A ceiling, not a count of withheld pages: every stamp the scan cannot read adds to it3 |
| What this page does not do | It does not say what stands at any missing position, it names no interview file, and it does not list the sheets carrying the sixth mark form, a litigation discovery reference, because those references index FBI interview files4 |
Page limits. This page is about numbering, not about contents. The sheets cited here carry a Bureau of Prisons register number, commissary account balances, transaction identifiers and a private individual’s street address, and none of those is reproduced, quoted or described further. No person is named from any sheet read for this page. Nothing here identifies a complainant, a witness or any subject of an FBI interview, and the one series that would make such an identification possible is excluded from this page by name and by reason, below. Where this page reports a count, the count is of marks and of numbering positions, never of documents about people.
At least two sheets in five in the Department of Justice’s Epstein release carry more than one number. A search of this project’s copy of the release found an earlier custodian’s mark on 1,173,440 of its 2,925,025 sheets, and on 4,797 of those sheets it found three such marks at once; because the search misses marks it cannot read, those are floors.3 Below the content, at the foot, sits the production number the Department assigned. Above it, often in smaller type, sit one, two or three further numbers, laid down by whoever held the paper before the Department did. Those earlier numbers are consecutive, they run one number to one sheet, and because they are consecutive they can be read against each other. Where the release holds a sheet for a number, that sheet was produced. Where no sheet can be matched to a number, the position is unaccounted for, either because nothing was produced for it or because the number printed on the sheet could not be read. This page sets out the numbering systems as they are printed on paper, shows what the automated reading of the release does with them, and explains why the arithmetic built on that reading is an upper bound on withholding rather than a count of it.
What this page covers, and what it does not
This page covers four numbering series: three of them read as ink on the paper, being the EFTA_ series and the SDNY_ series, which appear together on the same sheets, and the EFTA_R1_ series; and the Department’s own EFTA production number, which on every sheet read for this page is not ink but type the producer set beneath the scan. The prefix SDNY is the usual abbreviation for the Southern District of New York, and this page uses the stamp forms themselves wherever the distinction matters. Two further mark forms are named here and not worked: the bank production stamps, JPM-SDNY- and DB-SDNY-, which this page does not gap-count, because nothing read for it establishes how long the banks’ own runs were; and a litigation discovery reference printed on protective-order material in Dataset 9, in the form 3501.xxx-xxx. That last one is deliberately left alone. Those references index FBI case-file serials, and a serial stands for a witness or a complainant. Listing which sheets carry which reference would be an index of interview files, and this page will not build one.4
The wider statutory argument about how much the Department has released, and the floor statements by members of Congress that the release is incomplete, are set out at Epstein Files Transparency Act. Nothing on this page corroborates or contradicts those statements; the numbering is a separate question with its own evidence.
Three numbers on one sheet
EFTA00134541 is a bundle of seven sheets of Bureau of Prisons records, and every one of them was read at the page image for this page. At the foot of the first sheet, three numbers are printed one above the other:1
SDNY_00000002
EFTA_00000001
EFTA00134541
They are three different numbering systems on one piece of paper. The bottom number is the Department’s; it is also the name the release gives the whole seven-sheet file. The middle and top numbers are two earlier runs applied to the same sheets, the top one always a single number ahead of the middle one. Nothing printed on the sheet says who applied either of them or when, and this page does not supply an answer the paper does not give.
Across the seven sheets the three series advance together and stay in step. The Department’s number runs EFTA00134541 to EFTA00134547, one to each sheet. The EFTA_ series runs EFTA_00000001 to EFTA_00000007. The SDNY_ series runs SDNY_00000002 to SDNY_00000008, exactly one ahead of the EFTA_ number on every sheet. The fourteen earlier-custodian marks were read as ink on the paper; the seven Department numbers are not ink, but type the producer set in a band below each scanned sheet.1
Two things follow from that, and both are visible on the paper. The first is that the unit of this production is the sheet and not the document: a file the release calls EFTA00134541 is seven separately stamped sheets, and the name of the file is simply the stamp on its first one. The second is that the sheet stamped position 1 of the EFTA_ run is a Bureau of Prisons record. That run opens on a custody document, not an investigative one.1
The same sheet carries one more line, printed above the stamps: Page 2 of 2. The sheet at position 1 of the sequence is the second page of a two-page report, and the first page of that report is not in the bundle.1 This page does not assert that the missing page is absent from the release, because, for the reason set out below, the search that would prove it absent is not reliable enough to prove anything absent.
One number per sheet, right through a document
EFTA01810270 is a sixteen-sheet sales brochure for a business jet, filed in Dataset 10. All sixteen sheets were read at the page image. Every one of them carries an EFTA_R1_ stamp as ink, and the sixteen run consecutively from EFTA_R1_00171541 on the first sheet to EFTA_R1_00171556 on the sixteenth, with the Department’s own number running EFTA01810270 to EFTA01810285 beneath it.2
That is what the EFTA_R1_ series looks like when nothing is missing from it: an unbroken sixteen-position run, one position per sheet, printed on every sheet of the document. It is the control case. A break in such a run, read on the paper, is not an ambiguity: it is a position that the numbering run contained and the release does not fill.
What the released text does with the stamps
The release ships as scanned images with a machine reading of the words on them, and every count of these stamps that has been attempted, including this page’s own, is built on that machine reading rather than on the paper. On the twenty-three sheets read at the image for this page, that machine reading recovers fewer than half of the stamps the paper actually carries.
| series | marks on the sheet | rendered exactly in the released text |
|---|---|---|
EFTA (the Department’s own number, not ink) | 23 | 23 |
EFTA_ | 7 | 6 |
SDNY_ | 7 | 4 |
EFTA_R1_ | 16 | 4 |
The first row is not a reading at all: the Department’s number is type the producer added below the scan, so it stands in the released text by construction, and it is set out here only as the baseline the other three rows are measured against. The failures in those three rows are ordinary scanning failures and they are specific. On the first sheet of EFTA00134541 the printed SDNY_00000002 is rendered SON Y_00000002, and on the second sheet SDNY_00000003 is rendered 50NY_00000003; on the fourth, SDNY_00000005 is rendered SDNC00000005. The printed EFTA_00000001 on the first sheet is rendered EFTA_0000000 1, with a space inserted before the final digit.1 On the brochure, twelve of the sixteen printed EFTA_R1_ stamps are not rendered in their printed form at all; the first sheet’s EFTA_R1_00171541, which is the one a document-level reading would need, comes out as EFTA followed by _131_00171.2
That matters in one direction only. A stamp the machine could not read is indistinguishable, in a gap count, from a sheet that was never produced. Every stamp lost to the scan adds one position to the apparent gap. None of them can remove one. Any count of missing positions built this way is therefore a ceiling, and the real number of withheld pages is somewhere below it.
The consequence is already visible in this project’s own working files. In the cross-reference table built from an earlier reading of these stamps, the first sheet of EFTA00134541 is recorded as carrying EFTA_ number 0. The ink reads EFTA_00000001. That single misreading is the whole basis of the statement that the series begins at 0, and of a stated span one position too long.5 In the same table the SDNY_ number is recorded for four of the seven sheets and left blank for three, which at the image carry SDNY_00000002, SDNY_00000003 and SDNY_00000005.5 The sixteen-sheet brochure does not appear in the document-level cross-reference at all, because the method keys on the first sheet’s stamp and the first sheet’s stamp is the one the scan mangled; sixteen released sheets, each with its position printed on it, are missing from the map of what was released.52
Counting the gaps, and what the count is worth
An earlier working analysis inside this project put the number of pages “reviewed and numbered but not included in the public production” at 1.46 million, across the two series that can be counted this way. This page did not reproduce that figure. What it did instead was count the two series again from scratch, over the text of every sheet this project holds, with its own patterns, and the result is close to the earlier one without matching it.3
| series | range of numbers read | positions in that range | positions a stamp was found for | positions no stamp was found for |
|---|---|---|---|---|
EFTA_R1_ | 11 to 2,214,168 | 2,214,158 | 940,691 | 1,273,467 |
EFTA_ | 1 to 269,393 | 269,393 | 50,091 | 219,302 |
| both | 2,483,551 | 990,782 | 1,492,769 |
Three in five numbering positions, 60.1 per cent of the combined range, have no sheet in the release that this search could match to them, as of 25 October 2026. The unit is the sheet. The denominator is the 2,925,025 sheets this project held on that date, and the ranges are the lowest and highest values each pattern returned once values too long to be positions in an eight-digit run were discarded as scanning damage.3 The low end of the EFTA_R1_ range deserves the same suspicion: the lowest value the search returned is 11, while the lowest this page read as ink on a sheet is 171,541, and a spurious low value lengthens the range and so enlarges the gap.3
The earlier count gives 1,464,217 positions unaccounted for; this one gives 1,492,769, some 28,500 more. The difference is not a disagreement about the paper. Both counts read the same scans of the same sheets, and they differ in two conventions. The first is how forgiving each pattern was of a misread character, and a stricter pattern finds fewer stamps, which turns more positions into gaps. The second is bookkeeping: the earlier count credited a document’s whole run of positions to the stamp read on its first sheet, while this one credits only the positions it actually read a stamp for. That is the measure of how soft a figure of this kind is. Two readings of the same paper, differing only in convention, put tens of thousands of pages in different columns.
The shape of the gap is more stable than its size, and the shape is the interesting part. In the EFTA_R1_ series the 1,273,467 unaccounted positions are not one withheld block. They fall into 258,997 separate breaks, and 212,361 of those breaks, four in five, are five positions or shorter, together accounting for 434,711 positions. At the other end there is exactly one break longer than 10,000 positions, and it runs to 19,937. The earlier analysis, counting differently, found the same single longest break at the same length of 19,937. A run of numbering that stops and starts again a few positions later, hundreds of thousands of times over, is the signature of a review that pulled individual items out of a stream rather than of a block of material held back whole. It is equally the signature of a scan that loses one stamp here and another there, at the rate the page image has already shown, and this count cannot tell the two apart.3
What this count is, stated plainly. It is a ceiling, not a measurement of withholding. Every stamp the scan could not read adds a position to it and none can take one away, and the page image has already shown that reading failing on three sheets out of seven in one series and twelve out of sixteen in another. The honest form of the headline is: at most about 1.49 million numbering positions in these two series are unaccounted for in the release, and the true number of pages withheld is below that, by an amount nobody has yet measured. What is not in doubt, because it is printed on the paper, is that the runs exist, that they are consecutive, and that they break.
What was searched, and what was not found
The whole of this project’s copy of the released files was searched for the stamp forms described here: the text of 2,925,025 sheets, as this project held them on 25 October 2026, each tested against a pattern for every series.3 That is the denominator for every count on this page. The unit is the sheet, not the document; one row of the index is one sheet, and a document may be one sheet or hundreds.
Two negative results are worth stating with their limits. A search for SDNY_00000001, the position below the lowest one printed on the sheets read here, returns eight sheets that carry that string somewhere in their released text. None of the eight was read at the page image for this page, so nothing is said here about what is printed on any of them. That is not evidence that no such sheet was released: the same search shape, run for SDNY_00000002, also fails to return the sheet which this page read at the image as carrying SDNY_00000002 in ink. The control does not fire, so no absence is claimed.3 And the first page of the two-page Bureau of Prisons report discussed above was not looked for by any other route; nothing here says it is or is not in the release.
Three bounds on this page, named. Only 23 sheets were read at the page image, out of a release of millions; every statement about what the machine reading does to a stamp is a statement about those 23 sheets and is not generalised to a rate. The bank series and the discovery-reference series were counted but not read at any image, and nothing is said here about how their stamps are printed. And the figure of 1.46 million withheld pages, which an earlier working analysis in this project gives for these two series, was not reproduced; what this page offers instead is its own count, the denominator it was taken over, the date it was taken, and the direction in which it is wrong.
Footnotes
-
Bureau of Prisons records, seven sheets, Dataset 9. EFTA00134541, all seven sheets read at the page image at the native 96 pixels per inch, printed stamps EFTA00134541 to EFTA00134547. The document was fetched once, read-only, from the site’s own per-document path: HTTP 200, 323,831 bytes, 7 pages, reconciling with the seven sheets this project’s index holds for it; an invented identifier used as a negative control on the same path returned HTTP 403 with an
AccessDeniedbody. At the foot of sheet 1, three marks stand one above the other:SDNY_00000002andEFTA_00000001, both read as ink inside the scanned image, and below themEFTA00134541, which is not in the scan at all but is set as 12-point type in a white band the producer added beneath it; the same arrangement holds on all seven sheets. The same three series continue on each following sheet: sheet 2SDNY_00000003/EFTA_00000002/ stamp EFTA00134542; sheet 3SDNY_00000004/EFTA_00000003/ stamp EFTA00134543; sheet 4SDNY_00000005/EFTA_00000004/ stamp EFTA00134544; sheet 5SDNY_00000006/EFTA_00000005/ stamp EFTA00134545; sheet 6SDNY_00000007/EFTA_00000006/ stamp EFTA00134546; sheet 7SDNY_00000008/EFTA_00000007/ stamp EFTA00134547. Sheet 1 also prints the line “Page 2 of 2” above the stamps. The released text of these sheets renders theSDNY_mark asSON Y_00000002on sheet 1,50NY_00000003on sheet 2 andSDNC00000005on sheet 4, and theEFTA_mark asEFTA_0000000 1on sheet 1; the figures on this page are the image’s, and those four strings are what the released text gives. Sheets 1 to 7 carry Bureau of Prisons report headings, a register number, commissary balances and transaction identifiers, none of which is reproduced here; nothing is quoted from these sheets beyond the stamps, the “Page 2 of 2” line and the form heading. This identifier appears in no citation-restriction register held by this project. ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7 ↩8 ↩9 ↩10 -
Sales brochure for an Embraer Phenom 300 business jet, sixteen sheets, Dataset 10. EFTA01810270, all sixteen sheets read at the page image at the native 96 pixels per inch, printed stamps EFTA01810270 to EFTA01810285. Fetched once, read-only, from the site’s own per-document path: HTTP 200, 1,717,134 bytes, 16 pages, reconciling with the sixteen sheets this project’s index holds for it; the same invented-identifier negative control returned HTTP 403. Each sheet carries, as ink, an
EFTA_R1_stamp above the Department’s number, runningEFTA_R1_00171541on sheet 1 toEFTA_R1_00171556on sheet 16, one position per sheet with no break. In the released text the printed formEFTA_R1_is recovered on four of the sixteen sheets (sheets 4, 11, 12 and 16, counting from one); two further sheets give it with the underscores lost, asEFTA R1 00171553andEFTA R1 00171554; sheet 1 givesEFTAfollowed by_131_00171; and the remaining nine sheets give no form of it. The figures on this page are the image’s. The Department’s own number stands in the released text of all sixteen sheets, which is not a reading: on each of them it is type the producer set in a band below the scanned image, and the scan itself does not carry it. This identifier appears in no citation-restriction register held by this project. ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7 -
Search of this project’s copy of the released files, run on 25 October 2026 over the text of all 2,925,025 sheets then held, being 1,425,573 documents. Each sheet’s text was tested against one pattern per series:
EFTA[_ ]?R1[_ ]?followed by six to nine digits;EFTA_followed by six to nine digits;SDNY[_ ]followed by five to nine digits;SDNY[_ ]?GM[_ ]?,JPM[- ]SDNY[- ]andDB[- ]SDNY[- ]each followed by five to nine digits; and3501.followed by two or three digits, a hyphen and three digits. Counts of sheets, not of documents. The patterns match the printed form of each stamp and tolerate the loss of a separator; they do not attempt to repair a misread letter or digit, which is why every count of positions found is a floor and every count of positions missing is a ceiling. A position is credited only where a stamp was actually read on a sheet; no document’s run of positions is credited from the stamp on its first sheet, which is where this count parts company with the earlier one. Sheets matched, by series:EFTA_R1_957,833;SDNY_GM_165,776;DB-SDNY-67,076;EFTA_51,675;JPM-SDNY-50,459;SDNY_4,450; and the discovery reference on 5,073 sheets. For the gap table the values were bounded to 11 to 2,214,168 forEFTA_R1_and 1 to 269,393 forEFTA_, these being the ranges left once values too long to be positions in an eight-digit run were dropped: unbounded, the same search returns a highestEFTA_R1_value of 91,500,959 and a highestEFTA_value of 110,232,371, which are scanning damage and not positions. Gap runs in theEFTA_R1_series, by length: 212,361 runs of 1 to 5 positions (434,711 positions); 40,672 of 6 to 20 (383,013); 5,124 of 21 to 100 (190,411); 777 of 101 to 500 (137,768); 46 of 501 to 2,000 (39,520); 16 of 2,001 to 10,000 (68,107); and one of 19,937. The sheet totals are this project’s index as of the date given and are not a statement about what the Department released. ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7 ↩8 ↩9 -
The sixth mark form is a litigation discovery reference in the form
3501.xxx-xxx, printed on protective-order material in Dataset 9 alongside theEFTA_andSDNY_stamps. The3501prefix is an FBI case-file serial, and a serial stands for one witness, complainant or investigative thread. This page therefore gives no count of which sheets carry which reference, names no serial, and makes no statement about the completeness of any serial’s records. That restriction is a choice made for this page and is recorded here so that a reader knows the page is incomplete by design at exactly that point. ↩ ↩2 -
The cross-reference tables built by an earlier working analysis of these stamps inside this project,
institutional/efta_underscore_crossref.csvandinstitutional/r1_crossref.csv. They are not part of the Department’s release. The first records EFTA00134541 sheet 1 as carryingEFTA_number0, where the ink readsEFTA_00000001; it records theSDNY_number for sheets 3, 5, 6 and 7 and leaves it blank for sheets 1, 2 and 4, which carrySDNY_00000002,SDNY_00000003andSDNY_00000005in ink.0is the minimum value in that file, and the range stated for the series is taken from it. The second file, which maps documents to the device-review series by reading the first sheet only, contains no row for EFTA01810270. These are this project’s working files, not documents in the release, and they are cited here only as the artefacts whose arithmetic this page is testing. ↩ ↩2 ↩3