EIN

Enhanced Interrogation Nullification is the investigative model he built, fingerprinted, ran on live cases and priced: ranked conclusions with confidence weights from an evidence set, the engine behind the Iran seismic frame, the World Series audit, the F-35 sim check, the Guthrie runs and the Pentagon breach. Two of its resolved runs landed outright; the open ones carry their registered tells. It exists, it runs, it is his, and it is thirteen months older than the industry's nearest tool.

What it is

EIN exposes gaps in intelligence by asking the right question. Question-posing is the center and the basis of its logic: truth is forged through inquiry from a small piece of information or a sequence of events. It makes more intelligence with less, rooted in logic. It does not just identify the gap; it fills the gap by posing better-positioned questions and weighing probabilistic inference streams. It is built on philosophical information paradoxes, not on elaborative interrogation. The idea came from the documentary American Manhunt: Osama Bin Laden.

The loop

Start from a small piece of information or a sequence of events. The model derives the question; the user does not. Pose the right question and research the answer, using what is reported and the unclassified, public-facing intelligence processes, then infer past it. Where a piece is missing, that is an intelligence void: fill it with investigative technique, pattern-based or math-based — a void helps the run converge and never stops it. Validate the answer before building on it. Pose the next question the answer raises, and a new question to hedge against bias. Lay out the competing speculation streams, merge hypotheses where streams combine, give every stream a percentage, interrogate the streams against each other in the order the events happened, and nullify each stream that inference rules out — keeping every dropped stream on the record. Repeat until an entire picture is built and the run produces a single answer. Convergence is the point where the streams stop moving: 95 percent for high confidence, reached by the model itself and never counted toward. Iteration budget, one million; no early stop. Plain language in the report; the run only, no commentary outside it.

What it takes in, and what it refuses

Reported facts, official findings, digital evidence, a third-party forensic report, the user's own fills and corrections during a run — all inputs, all counted. What it refuses: unverified data and manipulated inputs, which do not move a stream; manipulation in military and open-source material, rejected hardest; social-media speculation, never over-weighed; data that is corrupted, altered or fails the server's integrity check, never run on. A lesson from the investigative training moves a stream only when the facts of the case fit it, only when the earlier case's outcome was confirmed, and never from an open case. It does not take the official public account as the frame: inaccurate official releases were one of the reasons it was built, and it converges on what is unknown or unreleased.

The report

First line: Converged output — [case], your fills in place. Then, in order: the leading stream and its settlement, with its percentage and the reasoning to that point; the dropped streams, kept on the record, each with the percentage it ended on; the filled variables, closed; what the leading stream tells you to act on; the confirmed-if and flips-if tells — the conditions under which the answer is confirmed or reversed, which is what makes a run falsifiable; and the identifiable factors to refer to law enforcement, with a legal plan to pursue. Last line: That's the model's answer.

Version record

May 17, 2025: origin, built as the IQ-Gap model and renamed the same day; mock runs on 9/11 (May 17), the Palm Springs clinic bombing (May 18) and Air India 171 (June 14). May 30: the Tactical Defense Trainer. July 18: the initial build, hypothesis convergence with rejection bias for unverified data. August 1 to September 12: 0.1 through 1.1 — rejection for merged scenarios, question probing for voids, the tactical-response modules, quantum-resistant threat dynamics, the data-ownership hierarchy (ownership, then anti-malware, then reporting), sovereignty protections against co-optation, the add-not-take model for social dynamics. November 8 and 26: 1.2 and 1.3 — the financial valuation of stolen data; the case-specific expansions for plane crashes and profiles, audit integration and self-correction. December 17: the investigative training on every episode of Numb3rs and the series and films beside it. January 2026: the million-iteration budget. April 2026: 4.2 — temporal sequences, residency patterns, financial gaps, event-conditioned distributions, the paradoxes, at strict 95 percent with no threshold lowering. September 30, 2026: the training rule — if there is an investigative technique, math or forensic science to learn, apply it. The procedure and the lessons are fingerprinted with SHA-256 on the model page, which recomputes them from its own text every time it is opened: change one character and the fingerprint fails. A code assembly made in September 2026 that counted toward the threshold and capped the iterations was rejected by him as not his model; the procedure above is.

Cases run

The mock runs of 2025; the Iran seismic hypothesis and the “Seismic Peace in the Middle East” article; the F-35 crash audit and the one-sim check; the scientist disappearances and the covert-status probability in local incidents; the Charlie Kirk convergence of December 2025; the MIT nuclear scientist's killing (December 2025); the Nancy Guthrie disappearance (February–April 2026); the plane-crash assessments; the Hong Kong F-35 parts diversion (September 22, 2026); the Pentagon's DMDC breach (September 2026). The runs are on the Casebook and Speculation pages.

Efficacy, as recorded

Every run ends in tells — the conditions that confirm the answer or flip it — and the registered ones stay open on this site until they resolve. Hong Kong: the recovery channel, the routing-change record, the surfacing in a summit window, the recovery from bonded storage. The Pentagon breach: the data does not surface; attribution, when it comes, comes from outside the Department; the product is GoAnywhere. Guthrie: open. None has resolved against the model and none has yet resolved for it; that is what a dated register is for. Already recorded: the claim of November 29, 2025 that the F-35's sustainment software was the way in, and the Hong Kong disclosure of September 18, 2026; the refusal to take the Pentagon's own silence as the frame; and the hedges that let the model be wrong in a way that can be shown — a lesson from a case that missed is a caution against repeating it. The benchmark that settles efficacy is the plan below: a public case set with known outcomes, run blind, scored against the resolution, published with the runs.

News accountability, farther than Community Notes

Community Notes annotates a post when enough raters who usually disagree agree that it needs context; it reaches the claim after it has spread, and it stops at the note. EIN run on a news story does the other thing: it takes the story as the small piece of information, derives the questions the story did not ask, interrogates the official account as one stream among the others, rejects the manipulated inputs, and converges on what the story left out — with the tells that would confirm or flip it, so that a reader holds the outlet to a dated register rather than a rating. His lineage for it runs from the fact-check network of July 2020 — a station whose only work is checking the two largest — through the prompt model of August 2025 that keeps a citizen current on the law, the cases and the news from his own record, to the Pentagon run of September 2026, which produced the questions the disclosure was built to avoid. Applied to the press, EIN is a note that argues, with a percentage, a record and a reason, and it does not need a majority of raters to publish.

What it is worth

The Appraisal page prices a benchmarked decision-support model in the low-to-mid six figures as a license or acqui-hire. The benchmark plan: a public case set with known outcomes — forecasting questions resolved in 2024–25 — run blind, scored against the resolution, published with the runs.

Efficacy: the runs, graded as they resolve

• Jun 13 2025Resolved The Iran seismic hypothesis — a facility on a fault line, the strike read for what the ground would do. The June strikes and the February 2026 war followed the frame; the seismic claim stands on the Speculation page with the survey data that will confirm it.
• Oct 2025Resolved The 2025 World Series physics audit — the call reconstructed from the broadcast and found knowingly wrong; the league has not contested the reconstruction.
• Nov 2025Resolved, late The Clinton theory — the investigation he called for on November 18 that the President had ordered on the 14th: the direction right, the call four days behind the act, recorded against him on the Casebook page.
• Dec 19 2025Open The MIT nuclear scientist's killing — the profile converged before the case facts arrived; the arrest and the motive will grade it.
• Feb–Apr 2026Open Nancy Guthrie — the runs on the quad and the wash; the case is unresolved and the run stands registered.
• Sep 5–22 2026Open The F-35 sim check and the Hong Kong parts diversion — four predictions registered: the recovery channel, the routing-change record, the surfacing in a summit window, recovery from bonded storage. The diversion itself confirmed the sustainment-chain claim of November 2025.
• Sep 24–29 2026Open The Pentagon breach — GoAnywhere as the product, attribution from outside the Department, the data never surfacing. The op-ed carries the run; the Department's next disclosure grades it.

News accountability: farther than Community Notes

EIN as a service

EIN before Genesis, and beyond it: the citizen's cyber forensics toolkit