What the Keyport Lab Reports Say, and What the Summary Tables Say

I ran a detector across the public record for the Aeromarine site. I am not accusing anyone of anything. I am asking two questions that the documents raise on their own.

By Paul Samuel Guarino

New Jersey DEP's current position on the Keyport Sanitary Landfill at Aeromarine Industrial Park is that the data so far "continue to indicate that people are not being exposed to contamination from the Aeromarine/Landfill site through drinking water or breathing vapors that enter a building emanating from ground water."

Congressman Frank Pallone has demanded an urgent state-federal probe into a suspected cancer cluster at the same site. Advocates have identified more than 40 cases, with 28 of them concentrated on and around a single street.

I am not a toxicologist and I am not going to tell you what caused anyone's illness. I do not know, and neither does anyone else right now.

What I can do is read the documents.

I built a detection system for environmental exceedances. I ran it across the published Keyport record — nine source documents, 8,802 rows, 477 analytes, 65 samples, every "not detected" preserved rather than dropped. Then I checked what it found against New Jersey's own legal criteria: residential soil remediation standards, and Class II-A groundwater quality standards as published in the June 2026 GWQS compendium — current law, not a cached table. That matters for at least one analyte here: the cobalt criterion was tightened from 100 to 2 micrograms per litre in the January 2025 adoption.

Two things came out that I think are worth someone's attention. Neither requires you to trust me or my software. Both are sitting in documents that have already been published.

The first question: how much of this data can answer the question it was asked?

When a laboratory reports "not detected," that does not mean the chemical is absent. It means it was not found above the lowest level that particular test could see. That level is called the reporting limit.

Normally the reporting limit sits well below the legal limit, so "not detected" genuinely means "below the standard."

At this site, for 150 of the 848 measurements carrying a numeric legal criterion, the reporting limit is at or above the legal criterion itself.

That is 17.7 percent — roughly one in six.

For those measurements, a sample could be over the legal limit and the laboratory would still, accurately and in good faith, report "not detected." The instrument was not looking low enough to answer the question being asked of it.

This is not a criticism of the laboratory. Reporting limits are set by method, matrix and budget, and some of these criteria are extraordinarily low — the Class II-A standard for PFOA is 0.014 micrograms per litre. Meeting it requires specialised analysis.

But look at where this lands.

Benzene

Benzene is a known human carcinogen. It is one of the carcinogens a 2010 assessment already associated with this site. Its New Jersey Class II-A criterion is 0.45 micrograms per litre.

In the June 2026 sampling of the MW wells, benzene was reported at ten of the twelve wells with a reporting limit of 0.45 — exactly equal to the criterion.

At those ten wells, benzene cannot be ruled out. Not "was not present." Could not be determined either way, because the test's floor sat precisely at the legal ceiling.

At the two wells where benzene was measurable, one of them — MW-2 — came back at 0.79 micrograms per litre, above the criterion.

The same pattern holds for 1,4-dioxane, a probable human carcinogen with a 0.4 criterion: repeatedly reported at a 0.4 reporting limit. Where it was measurable, MW-5 returned 0.43 — over.

So on the two organic contaminants in this dataset that carry the clearest health weight, the analysis was largely unable to answer the question, and in the handful of instances where it could, it found exceedances.

For nearly a fifth of the criterion-bearing measurements at this site, compliance cannot be demonstrated in either direction. Not passed. Not failed. Unanswerable with the data as collected.

If a conclusion about exposure rests partly on those measurements, that seems worth stating out loud.

The second question: why are 31 exceedances in the laboratory reports but not in the summary tables?

The record contains two kinds of document. There are laboratory reports, which list every individual result. And there are summary tables in the site investigation reports, which are what most people — residents, reporters, officials — actually read.

I compared them.

The summary tables I extracted cover wells labelled AMW-1 through AMW-12, plus SESI, FB and TB. They contain no groundwater rows at all for a second set of wells, MW-1 through MW-12.

The laboratory reports cover those MW wells fully. And in them, I count 31 individual results over New Jersey Class II-A groundwater criteria at those wells, sampled 22–24 June 2026.

Before the numbers, let me get ahead of the obvious objection, because it is a fair one.

Iron and manganese occur naturally at elevated levels in coastal New Jersey groundwater, and their criteria are largely aesthetic — taste, odour, staining — rather than health-based. Nobody should read "165 times the standard" for iron as "165 times a poison." I am not going to pretend otherwise.

That is exactly why the iron figure is not the point. The point is that these rows are absent from the layer people read — and the same absent set carries the analytes that do have health weight.

By analyte: iron 11, manganese 9, aluminum 5, cobalt 4, benzene 1, 1,4-dioxane 1. The cobalt results run from 3.6 to 9.9 micrograms per litre against a criterion of 2. Benzene above criterion at MW-2. 1,4-dioxane above criterion at MW-5. Those are not aesthetic standards.

With that said, the largest single figure is iron at MW-12: 49,400 micrograms per litre against a criterion of 300 — 165 times the standard.

Four separate wells are more than 100 times over:

I want to be careful here, so let me say plainly what I am not claiming.

I am not saying anyone hid anything. There are ordinary reasons a well might sit outside a particular summary table — different investigation, different scope, different client, different reporting period. Two separate consultants submitted data to DEP for this site, working for different parties.

What I am saying is narrower and, I think, harder to dismiss:

These exceedances were measured. They are in the laboratory record. They are not in the summary layer that a reader would consult first. And some of them are more than a hundred times a legal limit.

If there is a reason for that, it is knowable from the source documents, and somebody should state it.

What my system actually did, including what it failed to prove

I am going to be straight about this, because I would rather be trusted than impressive.

What it did well. It ingested a genuinely difficult corpus — nine documents, three different formats, 477 analytes, 5,043 non-detects — mapped the correct legal criterion to each analyte, and preserved every non-detect instead of silently discarding it. Scored against the documented criteria, it caught every exceedance: 85 of 85, zero missed. It raised one flag on a sample sitting exactly at the limit rather than above it, which is a convention about where you draw the line, not an error.

That is a data-handling result on real regulatory paperwork. My earlier benchmarks — Flint lead, Delhi particulates, Bethpage TCE — were single clean sensor series. None of them tested whether the system could survive actual government documents. This one did.

What it did not prove. The part of my system I think is novel — the harmonic and coherence layers, which look for structure across time and across stations rather than simply comparing a number to a limit — never got exercised. Each well series here runs ten to thirty readings, and those layers need longer runs. The catching was done by the simplest tier: is this number above the legal limit.

A threshold comparator correctly identifying values above a threshold is arithmetic. It is not a discovery, and I am not going to present it as one.

So the honest summary is: the engine handled the data. The findings above are about the data, not about the engine. You could reproduce both of them by hand with the published PDFs and a calculator. It would just take you a very long time.

This is the third site

I have run this system against Flint, against East Palestine, and against the Bethpage TCE plume on Long Island — where a 2007 groundwater sample near the Northrop Grumman facility measured 6,300 micrograms per litre of trichloroethylene against a 5 microgram federal drinking-water limit. One thousand two hundred and sixty times over.

(Link to my Bethpage article goes here.)

I have family history with that plume. I am not neutral about this subject and I am not going to pretend to be. But being not-neutral is exactly why I publish the arithmetic and the source documents rather than the conclusion.

The ask, and I have been making it for months

Every dataset I have tested, I chose. That is the weakness in all of it, and I know it.

I need someone to hand me data I have never seen, with the answers withheld, and let me run the system against it cold. That is the only test that means anything. I have asked publicly, repeatedly, across articles that thousands of people have read, and nobody has done it.

So let me make the easiest possible version of the ask, because maybe I have been asking for too much.

Option one — replication, about an hour of your time. The full Keyport bundle exists as 30 files with a SHA-256 checksum manifest: the nine source PDFs, the assembled dataset, the scripts, the benchmark report. Verify the hashes, run three commands, and tell me whether you get the same numbers I got. Nothing is withheld from you. You do not have to prepare anything. You either reproduce 85 of 85 with one boundary flag, or you do not, and either answer is useful to me.

Option two — a blind run, which is what I actually need. If you hold environmental monitoring data with known exceedances, send me the measurements with the answers stripped out. I will run the detector and send you back what it flags. You compare. I never see the key. If it fails, I will publish that it failed, under my own name, in the same place I published this.

I am one person with no institution behind me. I do not have a laboratory, a grant, or a department. What I have is a working system and a public record.

I keep being told that independent validation is what my work lacks. That is true. It is also the one thing I cannot supply for myself.

If you are a state regulator, an environmental consultant, a university lab, a journalist with a data desk, or a resident of any town like this one who has a folder of PDFs nobody has ever properly read — you can settle this in an afternoon.

[email protected]

Sources: NJDEP Keyport Landfill current status · NJ Department of Health update, 30 April 2026 · Rep. Frank Pallone, press release · NewsNation coverage · Analysis dataset: nine published source documents, NJDEP PI G000030506. Benchmark report, scripts and checksum manifest available on request.