Home/Completeness

Flight Record Completeness

How much of Epstein's flight activity appears in no released record? When the same flight universe is sampled by many independent document sets — pilot logbooks, congressional productions, estate emails, card statements, border records — the overlap structure between them lets you estimate how much none of them caught (capture–recapture, the method used to count unseen wildlife populations and census undercounts). This page presents that estimate for the 11,563 distinct flight events in our index. (Heads-up: /flights lists 16,997 records across 17 individual sources. This analysis first re-deduplicates to a stricter flight-event key (metro-area + exact date) and counts 5 independent source families — because sources inside a family transcribe the same paper and are not independent evidence. So these totals are deliberately lower, and are not the same unit as the /flights record count.)

Analytical estimate — not a documented fact. Everything on this page is a model-based inference from the overlap structure of ~20 evidentiary sources. The assumptions of capture–recapture are all violated here to some degree: the sources are not independent, they cover different eras, and an undocumented flight is precisely the kind least likely to be caught. The honest deliverable is a range with stated directions of bias, not a number — and the biases push most figures below downward, which is why the floors are quoted as floors. Full methodology & caveats →
10–40%
of Epstein's own aircraft activity (1999–2019) appears in no released record
This is the defensible number. For flights on Epstein's tail-verified fleet, four independent custodian groups (logbooks, House Oversight, government records, estate emails) could each in principle have captured every flight — so the statistical models fit well here (deviance/df 0.4–0.7). We observe 3,707 fleet flights; the estimators imply 4,100–6,600 existed, i.e. roughly 400–2,900 fleet flights appear in no released record. How this was estimated →

Why an estimate is possible: most flights rest on a single source

Capture–recapture logic in one line: if most observed flights were caught by only one source, the sources are sampling a much larger universe; if most were caught repeatedly, the universe is nearly exhausted. The 17 sources group into 5 independent source families; here is how many families attest each observed flight. Darker = more independently corroborated.
data table

The whole corpus — a softer, wider range

all travel events, 1991–2019 · less certain than the fleet figure
Extending beyond Epstein's own aircraft to all travel events in the record (guest commercial tickets, charters, staff travel) mixes populations with very different catchability, so this range is wider and should be read more cautiously. Heterogeneity-robust floor estimators put the minimum at ≈2,200–5,300 unrecorded flights; dependence-adjusted models put the plausible total at ≈15,000–25,000 (dark count ≈3,000–13,000). Models that fully parameterize source dependence permit far larger totals, but those fits are weakly identified and are not headlined. Estimator detail →

Which sources corroborate each other

Shared distinct flights between each pair of the 17 sources. The bright blocks sit almost entirely within the gold family outlines — sources in the same family are re-descriptions of the same underlying paper (three transcriptions of one logbook; emails and manifests from one congressional production). This is the dependence problem made visible: overlapping dependent sources make simple estimators read too complete, which is why the figures above are floors. Why dependence biases the estimate →
full overlap matrix (17×17)

1991–1998: unanswerable — and that is itself a finding browse this era →

The pattern of what does not exist — no financial production, no border records, no manifests for an eight-year span of heavy documented activity — is itself informative about where document discovery has and has not reached. Any flight not entered in the pilots' logbooks in the 1990s is invisible to every source we have. Full note →

Method, assumptions, and what this does not say

Unit of observation: a flight event keyed by (date, origin, destination), airports normalized to metro areas, deduplicated across sources (11,563 distinct events; 10,692–12,857 under stricter/looser merge rules). Estimators: Chao-type lower bounds, Chapman two-source pairs restricted to compatible windows, and Poisson log-linear models on the capture-pattern counts, era-stratified, with bootstrap intervals. Read these before quoting any number:
  1. Closed population — violated. Sources cover different eras; all estimates are era-stratified as mitigation.
  2. Source independence — violated, direction known. Positive dependence biases simple estimators downward — the floors are genuinely floors.
  3. Homogeneous catchability — violated, direction known. The unseen flights are precisely the low-catchability ones; also biases estimates downward.
  4. Record linkage is itself inference. Merge uncertainty bounded at roughly ±10% by the strict/loose key grid; a known linkage artifact (unparseable airport strings) is excluded from all quoted floors.
  5. The estimand is conditional. This estimates the universe of flights catchable by at least one source in principle. It says nothing about categories no source covers — the true total of all Epstein-network movement is strictly larger than any number on this page.
Full methodology writeup Flight index (the observed 11,563) Document coverage map
Home Flights Coverage Map GitHub
Estimates on this page are analytical inferences, not documented facts. Not affiliated with the U.S. Department of Justice, FBI, any government agency, or Anthropic.