# Claim audit

The basis column is the sourcing: each claim names the document, dataset or survey
it rests on. Nothing in this table is asserted without one.


11 September 2026. Every assertion the site makes, classified by how much
new evidence it would take to overturn it. The point is to stop treating a
finding drawn from a complete dataset and a finding drawn from "I haven't
found it yet" as the same kind of statement.

**Why this exists.** Over the four days of this work, several conclusions
reversed on the strength of a single newly-opened document. The cause was drawing conclusions from a fraction of a per cent of an archive without recording how sensitive each one was to new information. Reading more does not fix that by itself.

## Classes

**A — anchored.** Rests on a complete enumerated set, or is confirmed by two
independent routes. New documents can add context. They cannot reverse it.

**B — sourced and partial.** Rests on specific documents in hand. New documents can qualify or narrow it, but not make it false: the document contains what it contains.

**C — absence-based.** Rests on not having found something. Every unopened
file is a chance to overturn it. These do not appear on the site as
findings, only as statements about the state of a search, with the query
and the date it was run.

## The audit

| Claim | Class | Basis | What would overturn it |
|---|---|---|---|
| 1,079 school readings, 4 Oct 2001 – 30 Jun 2002, one above a limit | **A** | Complete series; 616 to 6 Mar 2002 matches the city's own count; 96.6% cell-by-cell agreement against two independently filed copies | Nothing short of the source table being a forgery |
| No ambient standard for asbestos existed; 0.01 f/cc is an indoor threshold | **A** | Stated by DEP in its own briefing note, and independently by the EPA Inspector General | Nothing; two independent sources, one adverse to its own interest |
| The nine exceedances, by date and corner | **A** | The city's own list, 148192 p. 351, with the denominator on the same page | A later revision of the city's own tally |
| 53 of 130 ambient samples over 0.01, none over the workplace limit, 15–23 Sep | **B** | The city's summary table, image-verified | A different city tally for the same window; the table is marked draft |
| Highest indoor reading 0.044, 11th floor, 225 Rector Place | **B** | 15 indoor samples from the daily tables | A higher indoor reading elsewhere in the release, which is likely |
| Lead above the national standard at three of ten locations | **B** | One metals table, one sampling round, column alignment eye-verified | More metals tables; 10,200 index entries mention lead |
| Benzene grab samples at 19 and 49 ppm; 24-hour samples at 0.005 | **B** | Federal draft evaluation, tabulated by location and date | A final version could revise the draft |
| Dioxin to 170 pg TEQ/m³; PCBs to 153 ng/m³; 50 landfill exceedances | **B** | Same evaluation, narrative and Table 6 | As above |
| Turnaround: median 19–27 days Oct–Jan, falling to 4 by May | **B** | 260 batches with both dates, holding assay constant | Fuller date coverage could shift the medians; the X/A decode could be wrong |
| Four sampling locations east of Broadway, none in Chinatown | **B** | 311 distinct locations extracted from the daily tables | A monitoring programme documented elsewhere in the release |
| EPA's reassurance ran ahead of its data; the White House Council on Environmental Quality shaped the wording | **A** | Inspector General finding, present in two independent copies (145345, 145510) | Only a later official retraction |
| The archive is mostly laboratory folders and building files | **A** | Structural survey of all 68 boxes; classification validated against Box 10, enumerated in full, where the sample found 16 of its 17 folders | Nothing; the survey covered every box |
| No Health Department collection exists in the release | **B** | Agency facet returns zero | A later monthly tranche |
| Whether the city concealed what it knew | **C** | Absence | Any of roughly 24,000 unread documents. Framed on the site as a state-of-search statement, not a finding |

## What this exercise found

**Absence-based claims were wrong nearly every time they were tested.**
Eight statements of the form "not searched" or "not extracted" were checked
against the archive; six turned out to be answerable from documents already
on disk, and the remaining two from documents one query away. Every one has
since been replaced with the query that was run and the date it was run.

**Class A claims did not move.** Every reversal during this work was to a
class C claim, always of the form "there is no evidence that X", overturned
by finding evidence of X. Nothing anchored in a complete series has changed. That is the argument for enumerating a frame before searching it.

**The residual risk sits in the datasets assembled as a by-product rather than built deliberately.** The box map began as an orientation exercise and became a published finding without anyone deciding it should.

## The limit that no class covers

Everything here is downstream of documents the city produced. The checks
test whether the documents were read correctly. They cannot test whether the
documents are complete. An EPA chemist alleged in 2004 that readings east of
Broadway were removed before the city published its data; the two figures
she cited are present in the internal file at the values and locations she
gave, but the published version she was comparing against is not in this
release, so the allegation cannot be settled here.

## Still unextracted, and why

**The perimeter tables.** The daily monitoring reports contain a run of
perimeter sampling tables that no figure on the site uses. An extraction
exists on disk with 594 rows covering 21 September to 4 October 2001, and it
is not published, for two reasons.

Only 91 of the 594 rows carry a numeric value; the rest are blanks or
below-limit markers. And the maximum among those 91 is 131.23 in a column
labelled PCM, which is a value on the electron-microscope scale, not the
fibre-counting one. Some rows are therefore mislabelled by method, and until
the source pages are read by eye it is not known which.

Ninety-one readings of uncertain method would weaken a dataset whose claim is
that every figure knows what it is. The tables remain the largest unextracted
source in the documents already on disk.

## What unread documents could still overturn

Testing the claims against documents already in hand found two that needed
narrowing.

**The nine.** The city's list runs to 6 March 2002 and counts electron-microscope
readings over 70 structures per square millimetre. A fibre count of 0.044 f/cc at
Vesey and River Terrace on 31 January 2002 — four times the reoccupancy threshold
— sits inside that window and is not on the list. It does not contradict it: the
two are different methods against different benchmarks. But it means "nine exceedances" is an answer about one
monitoring series measured one way. Other series with their own thresholds ran
at the same time.

**East of Broadway.** The count of three locations is drawn from the 311 in the
daily monitoring tables; the fourth, City Hall, is in the indoor readings. A second set of 21 locations, from laboratory
certificates filed in the email boxes, could not be placed. Their names put them all in Battery Park
City and the streets west of Church. That is inference from a label, not a
position.

Both counts are exact for the source they came from and say nothing about the
other series in the archive, which this project has not enumerated.