Revision, 2026-09-11. Two external reviews found that this chapter's close argued a claim the rest of the book refutes. Section VIII used to answer its own question with the keyhole image — that the doctrine hands a judge an aperture too small to see the record — and asserted that every restriction mechanism "is triggered by repetition visible to the court being asked." Chapter 2 disproves that on the text of the rules and chapter 8 §IV retracted it; this chapter now poses the question and hands it to chapter 2 rather than answering it wrongly first. Section VII's statement of the null result is also narrowed: the record is consistent with an institutional aggregation gap, which is not the same as establishing that any court declined to restrict because it lacked a count. No figure in this chapter has changed; its counts remain as of 2026-08-28 and are dated rather than stale.
I. A footnote
On 3 February 2026, Magistrate Judge Leo A. Latella issued a report and recommendation in Tufano v. Saladino, MD, No. 3:25-cv-01400 (M.D. Pa.). Most of it is unremarkable housekeeping. The interesting sentence is a footnote:
"Plaintiff has filed at least twenty-two separate actions in this Court within the past two years, half of which are closed."
That is a careful judicial statement. It is hedged ("at least"), bounded ("in this Court"), time-limited ("within the past two years"), and it declines to draw a conclusion from itself. It is not a vexatious-litigant determination, and no filing restriction followed it.
It is also, as far as it goes, exactly right.
Apply this project's de-duplication to the corpus — union-find over SameActionAsCaseID, so that a state case and its federal removal count once — and the number of originating actions in the Middle District of Pennsylvania on 3 February 2026 was twenty-two. Not approximately twenty-two. Twenty-two.
The number of originating actions this plaintiff had commenced by that date, across all forums, was sixty-three.
Forty-one actions existed that the footnote did not, and structurally could not, reach.
II. It happens again, five months later
On 13 July 2026, Judge Julia K. Munley dismissed Tufano v. All-Ways Forwarding, No. 3:25-cv-01558 (M.D. Pa.). Her order addressed, among other things, whether a magistrate judge had been unfair in characterising the plaintiff's litigation history. She reviewed the cited cases and concluded that the description — "a prodigious, but prodigiously unsuccessful, pro se litigant" — was not inaccurate. A footnote recorded that Clerk records showed 38 M.D. Pa. lawsuits since December 2023, many of them dismissed.
Same recomputation, same date. Middle District of Pennsylvania: thirty-eight. All forums: eighty-three.
Forty-five uncounted.
| Date | Court's own count | Same forum, recomputed | All forums | Uncounted |
|---|---|---|---|---|
| 2026-02-03 — Latella, R&R footnote (C055) | "at least twenty-two" | 22 | 63 | 41 |
| 2026-07-13 — Munley, order footnote (C057) | 38 since Dec. 2023 | 38 | 83 | 45 |
Two judges, five months apart, in the same district, each counting accurately, each arriving at between a quarter and a half of the record.
III. The observation travelled. The count did not.
The tempting reading of Part II is that two judges happened to notice the same thing independently, and that a third might eventually have noticed more. The record does not support that reading, and the way it fails is this chapter's real finding.
The phrase Munley endorsed was not hers. Its earliest appearance in this corpus is 4 November 2025, in Magistrate Judge Martin C. Carlson's report and recommendation in Tufano v. Google LLC, No. 3:25-cv-01574 (M.D. Pa.), which described the plaintiff as a prodigious but prodigiously unsuccessful pro se litigant and cited multiple earlier federal cases that had been dismissed or otherwise adversely resolved. Latella's own February footnote records that substantially the same observation appears in the report and recommendation in the companion case No. 4:25-cv-01960. Munley, in July, adopted the characterisation after reviewing the cases it rested on.
So the observation was not isolated. It was made, repeated, cross-cited and expressly ratified across the reports and orders of one district over nine months. Judges of the Middle District of Pennsylvania were, in the most ordinary sense, telling each other about this plaintiff's filing history.
And in nine months of that, the number never once moved outside the district. It went 22, then 38 — both times a count of the federal docket, both times correct, both times roughly half the record. Nothing in the mechanism by which one judge learns of another's view carries a figure across a forum boundary, because there is no such mechanism. What travels between judges is a characterisation. What does not travel is a count.
IV. Why the reproduction matters
A reader has every reason to distrust an outside count that exceeds a court's own. Courts have clerks, dockets and PACER; a research project has a database and an argument to make. The larger number is exactly the kind of figure that gets inflated by double-counting removals, by treating an appeal as a new case, or by mistaking a caption change for a filing.
That is why the exhibit is built to reproduce the judicial figures first.
The de-duplication is a union-find over the SameActionAsCaseID column: a state complaint and its federal removal collapse into a single originating action, counted once and dated from its earliest docket. Both docket records survive in the corpus for auditability, but they are one action for counting. This is the same de-duplication the public site applies, and the same one the repeat-filing coding protocol takes as its unit of analysis.
Run that de-duplication, restrict it to one forum, cut it off at one date, and it returns the judge's own number. Twice — on two dates, for two judges, at two different values. Nothing in the method was tuned to those figures, and if it stops matching after a data edit, detection_gap.py exits non-zero and says so.
The exact match is not a flourish. It is the licence for the second number. A method that reproduces a court's count when restricted to that court's forum has earned the right to report what the same method returns when the restriction is lifted. The 63 and the 83 are not a different, looser count. They are the same count with the forum filter removed.
Which reframes what the gap measures. It is not a discrepancy between a judge's rigour and a researcher's. It is the size of the filter.
V. What the filter removes
The forty-one and the forty-five are not a scatter of stray filings. They are concentrated, and the concentration is the argument.
| Forum of the uncounted action | As of 2026-02-03 (41) | As of 2026-07-13 (45) |
|---|---|---|
| Lackawanna County Court of Common Pleas | 24 | 28 |
| New York State Supreme Court | 13 | 13 |
| U.S. District Court, S.D.N.Y. | 2 | 2 |
| Superior Court of California, Los Angeles County | 2 | 2 |
No uncounted action spans two of these forums, so each is attributed to exactly one court.
The single largest invisible block, on both dates, is one state trial court. More actions sat in the Lackawanna County Court of Common Pleas, unseen, than sat on the federal docket the judges were counting.
Nor is this a case of two court systems that never touch. Ten of the ninety-two originating actions in this corpus exist because a Lackawanna complaint was removed into the Middle District of Pennsylvania; the corpus records sixteen researched Pennsylvania cross-forum relationships, eight of them confirmed removals. The federal court did not lack access to the state record in principle. It had ten pieces of it on its own docket, each arriving with a state caption and a state docket number attached.
What it lacked was any reason to count them as evidence of anything beyond the case in front of it. A removed case enters the federal docket as one case. It does not enter as a datum about the twenty-three others still in the county.
It is worth being precise about the shape of this. It is not distance — California is two actions, the smallest block on the table. It is not obscurity — every one of these dockets is a public record, and this project located them with free tools. It is adjacency without aggregation: the largest uncounted volume sits in the court procedurally nearest to the counting court, close enough to hand cases across, and there is no instrument on either side that adds them together.
VI. The honest arithmetic
The proposal for this thesis says the invisibility widens as the filing rate climbs. Over the window these two footnotes bracket, that is true of the gap and false of the share, and the distinction should be conceded before a sceptical reader finds it.
Between February and July 2026 the absolute gap grew: 41 uncounted, then 45. But the federal docket grew faster than the rest of the record over those five months — 22 to 38 in forum, against 63 to 83 overall — so the proportion of the record invisible to the counting court fell, from 65.1% to 54.2%.
The claim this chapter makes is therefore the narrow one. The absolute number of actions no court had counted rose over the period, and stood at forty-five when a judge of the Middle District of Pennsylvania last put a figure on the record. The share is volatile, moves with whichever forum happens to be busy in a given quarter, and should not be leaned on. Nothing in the argument requires it: forty-five uncounted actions is not a proportion problem, it is an aggregation problem, and it would be the same problem at any share.
For the record as it stands today — 2026-08-28 — the corpus holds 92 originating actions, de-duplicated from 102 docket records, across five trial forums in three states, naming 334 canonical defendants and grouped into 20 analytical disputes. Actions by year of filing: 3 in 2018, 1 in 2020, 1 in 2021, 1 in 2022, 3 in 2023, 10 in 2024, 40 in 2025, and 33 in the first eight months of 2026.
No court has counted them. The highest figure any court has stated is 38.
VII. The null result
If the aggregate record were merely large and otherwise unremarkable, the gap would be a curiosity. It is not.
Sixty-five court findings are recorded across these cases, involving 26 distinct judges. Their language is not gentle: frivolous claims and transparent attempts at jurisdictional manipulation; appeals expressly deemed frivolous and not taken in good faith; amendment futile; a legal theory rejected once and pleaded again. In the same document as the twenty-two-action footnote, Latella described three of these actions as a trilogy raising essentially the same claims against the same defendants, and found that one had repeated the identical factual allegations of another already dismissed.
Fourteen days later, on 17 February 2026, a fourth action against the same lead defendant was filed in the same district (No. 3:26-cv-00396). A fifth followed on 11 May 2026 (No. 3:26-cv-01266). The word "trilogy" did not survive contact with the docket.
Nor is that anecdotal. Under a coding rule published at thesis/protocols/01-refiling.md and applied to the whole corpus, 43 of the 88 actions this plaintiff commenced repeat an earlier one — 13 graded Strong, 25 Moderate, 5 Context — and the repeats track the filing rate almost exactly: 1 in 2022, 2 in 2023, 2 in 2024, 20 in 2025, 18 in the first eight months of 2026. Separately, 34 filing-restriction indicators are coded, 13 of them Strong.
Against all of that, the record contains:
- Zero pre-filing orders. No filing restriction of any kind has been entered in any tracked case, in any forum.
- Zero monetary sanctions. No Rule 11 or comparable monetary sanction has been verified anywhere in the corpus.
- Two sanctions in total — both non-monetary, both terminating, both entered by the same judge in the same department of the Los Angeles Superior Court, on that court's own order to show cause, for failing to appear and failing to file a case management statement (13 October 2025 and 14 July 2026).
- Two vexatious-litigant requests, neither ever decided. One was filed on 31 March 2026 in the action that began as No. 3:25-cv-00011 (M.D. Pa.) and moved to the Central District of California as No. 2:25-cv-00147; as of 28 August 2026 that docket had been read through entry 158 and records no ruling on it. The other was raised on 8 July 2026 in the Los Angeles action, where the court held that a vexatious-litigant designation was beyond the scope of the motion to quash before it and that a separate motion would be required. That motion was never filed.
The two sanctions sharpen the point rather than blunt it. The only judge in this record who sanctioned this plaintiff twice did so for absence — for not turning up — and the sanctions did not aggregate into anything. They did not aggregate with each other, and neither of them, nor both together, brought a ninety-two-action pattern into any court's view.
The null result is the finding, and it has to be stated at exactly the strength the record supports. A record of this size, drawing this density of adverse judicial language and producing no restriction at all, is not evidence that restriction was unwarranted. Neither is it proof that the courts declined to restrict because they lacked a cross-forum count. What it is, precisely, is consistent with an institutional aggregation gap — and the rest of this book is an argument about how much more than that can be established.
The alternatives it cannot exclude belong here rather than in a later concession. A judge may decline to restrict as a matter of discretion, restriction being "very much the exception to the general rule of free access to the courts." The evidence actually put before any particular court may have been insufficient to support an order. The procedural posture may never have presented the question. And a defendant facing one suit may have had perfectly good reasons not to ask. This book cannot rule any of those out, and does not try.
What it can say is narrower and survives all of them: on the two occasions a party did ask, neither request was decided.
VIII. Not judicial error — and therefore a question
It would be easy, and wrong, to read the preceding sections as a criticism of the judges in them.
Latella's footnote is accurate. Munley's footnote is accurate. Carlson's characterisation was reviewed by a district judge against the underlying cases and found not inaccurate. Every judge who has put a number on this record has put a correct number on it. Not one of them was careless — and the exhibit's design, reproducing the judicial figure before relaxing the filter, exists precisely so that this cannot be mistaken for a claim that they were.
That is what makes this record strange rather than scandalous, and the strangeness is worth stating exactly before anyone reaches for an explanation.
The pattern was perceived. It was perceived accurately, by more than one judge, described in strong terms, written into orders, and passed between the judges of a single district across nine months. Two parties asked a court to do something about it. Four restriction regimes were available across the forums involved, one of them written specifically for repeat pro se filers. And the outcome column is empty.
So the question this book has to answer is not why did nobody notice? Somebody did, twice, in writing. It is:
Why did an accurately-perceived pattern produce no consequence?
There is an obvious answer, and this project believed it for a long time: that the doctrine hands a judge a keyhole, that each of these judges described the keyhole accurately, and that the aperture is simply too narrow to hold the record. On the evidence assembled so far that answer is not merely plausible — it is the one the two footnotes appear to demand.
It is also wrong, and chapter 2 is where it comes apart. Read the four regimes against their own text and two of them are not docket-local at all: they name no forum, and one has already been applied by a state appellate court to a prior action in a federal one. Whatever produced the empty outcome column, it was not that the rules forbid a judge from looking further.
Chapter 2 therefore puts to each regime the narrow question this chapter has earned the right to ask. When the rule asks what the filer has done before, where is it permitted to look — and what is it able to do about what it finds?