The 9/11 inside-job claim is two
claims.

The boundary

One phrase, two claims, and two different evidentiary bars.

The question asked what the evidence indicates about claims that the attacks were orchestrated or knowingly allowed by elements of the U.S. government. Before any search ran, the run recorded that “inside job” does not say which of those it means.

Orchestration and foreknowledge-with-inaction are not degrees of one claim. One requires participation — people placing charges, launching a missile, substituting an aircraft. The other requires only a decision not to act on something already known. They break in different places, they leave different traces, and the searches that would test them go to different archives. The run wrote that down as an ambiguity rather than quietly picking one and proceeding as though the question had been clear.

It recorded two more. Whether the question wanted the truth of the claims adjudicated or a map of what has been claimed and on what basis. And what “government” names here — the administration in office in 2001, particular agencies, or an unspecified set of institutional actors, since different theories implicate different parties. Then it stated its assumption: the comparison baseline would be the official account, al-Qaeda operatives acting without U.S. foreknowledge or complicity. A reader who rejects that baseline should discount what follows accordingly.

One exclusion matters more than the rest on a page like this. The psychology and sociology of why these theories persist was placed out of scope except where it bears on characterising the claims themselves. The run was not permitted to explain the belief instead of testing it.

Orchestration and inaction are not degrees of one claim. They break in different places.

Competing explanations

Four explanations, and the column that fills up fastest is the one to distrust.

Before any source was read the run wrote down four rival accounts and, for each, the observation that would destroy it. Ninety-five passages went on to declare a bearing on all four.

H1 is the official account. H2 is active orchestration — demolition, a missile, a substituted plane. H3 is LIHOP: specific actionable foreknowledge, deliberately unused. H4 says the public record is genuinely mixed in places without any of that amounting to orchestration.

Read H2's row first. Nine weak positives, ten strong negatives, no strong positive at all, and sixty-nine passages that bore on it not at all. Now read H3's: forty-eight positives against eighteen negatives. On the face of the table, the foreknowledge hypothesis is doing well and the orchestration hypothesis is not.

The second half of that reading does not survive chapter six. A positive bearing means the passage, if true, would support the explanation — and a passage reciting an allegation earns a positive whether or not the allegation was ever verified. Sixty of these ninety-five passages come from two compilation pages whose function is to list allegations. H1's fifty-five negatives are the same artifact seen from the other side.

The counts are unweighted by source tier on purpose. No weight vector has been argued anywhere in the system that produced this record, and an invented one would let a genre label quietly outvote the passages themselves.

A passage that recites an allegation scores the same as a passage that establishes one.

The physical channel

H2 wrote its own falsifier, and this ledger satisfies it item by item.

The orchestration hypothesis named the conditions that would destroy it before any source was read: collapses fully explained by impact and fire, no corroborated residue, no demolition seismic signature, no chain-of-custody physical evidence. The run went and collected each one.

The mechanism first, from NIST's own investigation summary. Impact severed and damaged support columns, dislodged the sprayed fireproofing and dispersed jet fuel across multiple floors. Bare steel at about 1,000 degrees Celsius softens to roughly a tenth of its room-temperature strength, and unprotected steel can reach the air temperature within the time those fires burned. Sagging floor trusses pulled the perimeter columns inward, and the potential energy released by the upper block exceeded what the structure below could absorb. Collapse initiated at the impact and fire floors and nowhere else, ran top-down, and took 56 minutes for WTC 2 and 102 for WTC 1 — intervals tracking impact damage and fire spread rather than a schedule.

Then the signatures a demolition would have to leave. No seismic signal precedes either collapse; the spikes begin about ten seconds after initiation, consistent with debris reaching the ground. No blast or explosion below the impact floors was recorded by NIST, the NYPD, the Port Authority police or the fire department. The lateral “squibs” match air compressed ahead of the falling mass. The bright molten stream is consistent with aircraft aluminium and the lingering rubble heat with oxygen-starved smouldering. USGS dust analysis found no thermite residue.

The Pentagon is a separate channel and reads the same way: hundreds of witnesses, clipped highway light poles on the approach, recovered airframe debris, DNA-identified victims, a retrieved flight recorder. The run's note on the “too small a hole” objection is that it misconstrues what a thin aluminium airliner does to a reinforced building. And the authorship of the plot is reconstructed independently of any of that — Khalid Sheikh Mohammed's proposal, the flight training, the travel, bin Laden's own acknowledgement of responsibility.

The chain below is H1's, not H2's. Two of its five conditions are the ones this ledger fully supports, and they are both physical. The rest of this dossier is about the other three.

Collapse initiated at the impact and fire floors and nowhere else.

One building

The live technical dispute is one building, and the decisive test was declined.

The disagreement among qualified people in this record is not about the Twin Towers. It is about World Trade Center 7 — and both sides of it are computer models rather than physical samples.

A four-year University of Alaska Fairbanks modelling study, released on 25 March 2020, concludes that the near-simultaneous failure of every column caused the collapse, rather than the fire-initiated local failure NIST describes. The same record notes that NIST's own simulation shows a façade distortion that does not match the observed footage. This is where H4 holds most strongly and where H2's technical falsifier has not been satisfied.

The source is worth stating plainly. It is a discussion-forum request for expert input, written by a site manager who says he is not a structural engineer and is asking for opinions he lacks. The run assessed it at the lowest tier, `unsupported`. It is in this ledger because a counter-search aimed at H2 returned it, and it is the whole of the pro-UAF case here.

NIST declined to test the WTC 7 debris for explosive or thermite residue, on the ground that its collapse hypothesis already accounted for the observations. The run treats that as a genuine methodological failure and then refuses to let it pay out. In its own words: the decision not to test is a fair criticism; it is not evidence that residue existed. So the gap supports the underdetermined reading and carries no affirmative weight for orchestration — which leaves H2 with no evidence in this ledger that is both physical and unrebutted. Only gaps.

The run also wrote down a candidate reading it did not establish: if the WTC 7 disagreement is ever settled by physical rather than computational evidence, the decisive test is residue and metallurgical analysis of retained steel and dust under documented chain of custody — and if no such material still exists in testable form, the question stays permanently open as a matter of evidence availability rather than of unresolved physics.

A test that was never run is not a test that found something.

The other claim

The warnings are not in dispute. Their specificity is.

The foreknowledge argument does not turn on whether warnings existed — this ledger establishes a great many. It turns on whether any of them named a time, a place and a method.

The stream in the record: at least twelve reports over seven years about aircraft used as weapons; the Phoenix memo in July 2001; the Moussaoui arrest in August; the President's Daily Brief of 6 August 2001, “Bin Ladin Determined to Strike in US”; a summer the Joint Inquiry described as the system blinking red; and liaison warnings from Italy, Jordan, Egypt, the United Kingdom, France, Israel, Algeria and the Taliban foreign minister.

On the single variable that decides between the two explanations, the documented warnings are generic. Both the Commission and the Joint Inquiry characterise the failure as vagueness, unrecognised significance and non-dissemination across agencies. The run's own attack on that reading is filed as selection bias and it is a fair one: the characterisation rests on executive summaries and Commission syntheses, and compartmentalised signals intelligence from the summer of 2001 could still contain something specific. The condition under which the reading fails is written down. It has not been met here.

The episodes that look most like intent are where this cuts hardest. FBI headquarters blocking the Moussaoui FISA warrant; a CIA officer blocking notification to the FBI that al-Hazmi and al-Mihdhar were inside the country; Robert Wright and John O'Neill describing shut-down al-Qaeda investigations; a curtailed inquiry into the bin Laden family; a reported change in the Attorney General's travel. Every one of those is described by the same primary witnesses in the language of negligence — Rowley on mishandling and inertia at headquarters, Grassley on the FISA application not going forward, Mueller conceding the Minneapolis warning about flying something into the World Trade Center should have been pursued more vigorously, Wright framing the shutdown as a failure to take terrorism seriously.

Blocking behaviour is compatible with both explanations, and nothing in this ledger discriminates the motive behind it. That is the run's finding, and it is not a finding for either side. Two more strands stay open on the same terms: Able Danger, where the dispute is not about interpreting a shared document but about what documents exist; and the word “solely” in the official account, which a partly declassified 2012 FBI summary and the Operation Encore reporting press on from the foreign direction while the 2004 Commission findings press back.

Blocking behaviour is compatible with both explanations. Nothing here discriminates the motive.

Where the passages came from

Sixty of the ninety-five passages came from two pages whose job is to list allegations.

Nothing in this run weights a passage by where it came from. That makes it worth seeing where they came from.

Thirty-nine of the ninety-five were read out of one Wikipedia article, “September 11 attacks advance-knowledge conspiracy theories”, whose assessment in this same record notes that it presents claims alongside official rebuttals without weighing evidentiary strength, giving equal narrative space to unsubstantiated claims and well-documented investigative conclusions. Twenty-one more came from a personal wiki page written under a pseudonym and compiled, by its own note, with AI assistance; the run assessed it at the lowest tier. Between them, sixty-three percent of everything weighed.

The technical case against demolition — the part of this record that is genuinely one-sided — rests on eight passages from a single NIST FAQ page and fifteen from a conspiracy explainer. That is why chapter two's table looks the way it does. The foreknowledge column filled up because two compilation sources enumerate dozens of allegations one at a time; the orchestration column stayed empty because the technical rebuttal is compact and the claims it answers arrive in bulk.

This is not an argument that the foreknowledge hypothesis is wrong. It is an argument that its column counts claims made rather than claims established, and that reading the bare tally as a score would invert what the run actually found. The run's own net assessment says the same thing in its own words: the foreknowledge reading is unsupported as to intent, while its factual premises about the sheer volume of warning are conceded.

The column filled up because a compilation page lists allegations one at a time.

Falsification

Ten attacks on its own explanations. Four came back holding something.

The run ran ten adversarial searches, each aimed at an explanation it had written down. Four returned a verified counter-passage — the best yield in this archive — and the six blanks are still not evidence of absence, because no probe was pre-registered.

The four that returned did real work. Against the official account: the Joint Inquiry's own findings on the withholding of information about al-Hazmi and al-Mihdhar. Against orchestration: the thread carrying the UAF Building 7 study. Against foreknowledge: the SEC's account-level trading review, which worked through 9.5 million transactions across 103 companies and six industry groups and spoke to the people responsible for the unusual trades. Against the underdetermined reading: ProPublica's account of Operation Encore and the Saudi files, which pushes back the other way.

The six blanks are pointed at things worth noticing. Independent chemical analysis of the red-grey chips — the single test that would settle whether they are nanothermite or standard primer paint — came back empty. So did the Pentagon airframe forensics, the Mineta PEOC logs, the NORAD radar reconstruction, the specificity of the foreign-service warnings, and structural-engineering consensus on the NIST collapse models.

One blank has a second life. A separate snapshot of the Department of Defense Inspector General's report on Able Danger was stored with the digest e3b0c442… — the SHA-256 of zero bytes. The run fetched it, hashed nothing, kept the record, and printed that everything about the source is absent because no text was supplied. The document that would decide whether Able Danger identified Mohamed Atta before the attacks is in this ledger as an empty file, and the page says so rather than quietly dropping it.

The report that would settle Able Danger is in the ledger as zero bytes, and the record says so.

The best-supported reading

The strongest finding, printed with the attack made on it.

The run's second finding is that every physical signature a controlled demolition would have to leave is absent or independently explained. It is typed observed rather than inferred. Then the run named the way it could be wrong, and the objection is specific enough to act on.

The attack is filed as measurement error, and it goes at the negative controls rather than at the conclusion. Ruling out thermitic material on the strength of USGS dust sampling assumes the sampling caught what mattered: the right selection of samples, enough resolution in bulk spectroscopy, and micro-chemical analysis of the distinct red-grey unreacted chips rather than of dust in aggregate. Those are real analytical limits, not rhetoric.

The condition under which the finding fails was written down in advance and is findable: re-analysis of authenticated, chain-of-custody dust samples by transmission electron microscopy and energy-dispersive X-ray spectroscopy, confirming nanostructured unreacted thermitic chips distinct from standard primer paint. That is exactly the study the counter-search went looking for and did not find. Nobody has to be dishonest for this finding to be wrong; somebody has to run the assay.

Seventeen attacks like this are on the record, at least one per asserted finding, each carrying its own failure condition. Nine of the sixteen asserted findings are typed observed and seven inferred. None of them was reviewed by a person: this run does not record whether any reviewer approved any gate, and the page says so rather than implying otherwise.

The score

Sixteen findings asserted, every one unknown, and 14 out of 100.

Sixteen findings were asserted, nine of them typed observed rather than inferred. Every one computed to a confidence of unknown, and the evidence score came to the same 14 out of 100 as the four publications before it — for the same reason.

Zero of eleven snapshots reached an upstream origin. Corroboration and provenance therefore both scored zero regardless of how good the underlying material is, because unknown provenance is unknown independence and the scoring will not let a well-argued reading buy its way past that.

Falsification scored 16 of 20 — the highest in this archive, four of ten counter-searches returning. Saturation scored zero, and this is the worst saturation measurement the series has produced: all five source families found in the second retrieval round were new, and the bearing counts on all four explanations moved when that round was added. The search was not close to exhausted when it stopped.

Two measurements belong beside the score. No source in this ledger was assessed as primary data — the technical case against demolition rests on a FAQ page summarising the NCSTAR reports rather than on the reports. And eleven of the hundred and six readable passages were refused by the verbatim gate and appear nowhere in the record, the highest refusal count in the series.