# Looking Glass: Provenance, Reality, and What Can Actually Be Built Master brief. Companion documents: - `provenance_findings.md` — forensic trace of the playlist and the AI repackaging - `real_programs.md` — the real predictive-analytics programs, with primary sources - `five_eyes.md` — the Five Eyes alliance, its actual analytic capability and limits - `taxonomy_map.md` — 31 programs and claims sorted into evidence tiers, plus the structural tells --- ## Part 1 — The clip you sent **Verdict: that footage is authentic, and it is not AI-generated.** The caption in your recording reads "...get to a target usually by some extraordinary means of jumping out of an airplane." That line appears verbatim at **timestamp 856 seconds (14:16)** of the original January 2012 Project Camelot interview. It is present in uploads timestamped [2013](https://www.youtube.com/watch?v=VtHCofbE1PM), [2020](https://www.youtube.com/watch?v=jEibpNOdW-s), [2021](https://www.youtube.com/watch?v=EjnqRtJVdpQ), and [2023](https://www.youtube.com/watch?v=PBtdwCSDoQc), and in an [Internet Archive copy](https://archive.org/details/bill-wood-interview-with-kerry-cassidy-project-camelot). The decisive argument is chronological, not forensic. This footage has been continuously online since January 2012. Photorealistic generative video of a talking human did not exist in 2012 — early GAN face *stills* were low-resolution and obviously artificial until roughly 2018, and convincing synthetic video of a speaking person is a 2024-and-later capability. A file with a 2012 provenance trail cannot have been generated by tools that did not exist. **Why it nonetheless looks synthetic to you — and these are real observations, not mistakes:** 1. **Blown-out backlighting.** He is seated in front of a bright window. The camera exposes for his face, so the background washes to flat pastel. The flag behind the glass loses its blue union entirely and reads as pink-and-white — which is exactly the kind of "wrong flag" artifact that correctly flags AI imagery in other contexts. Here it is overexposure. 2. **2012 consumer video at low bitrate**, which smears skin texture into a waxy, plastic surface. 3. **AI upscaling by reuploaders.** This is genuine AI involvement. Channels routinely run old footage through AI upscalers and frame interpolators, which produce authentically uncanny waxy skin and smeared edges. If you are watching a recent repackaging, AI has touched the footage — as *enhancement*, not generation. 4. **Your capture chain.** Screen → iPhone camera at an angle → re-compression. Your file is a 10.9-second, 1920×1080 iPhone 12 recording. At that level of degradation, pixel-level authentication is not possible for anyone, including me. **The right test, generally:** provenance beats artifact analysis. Find the earliest upload and an Internet Archive capture. Artifact-hunting on a re-recording of a re-encode of an upscale will mislead you in both directions. --- ## Part 2 — Who is actually running it Your instinct that something is being run on people is correct. The target and the operator are just different from what you expected. **The playlist you sent is not an operation.** It belongs to a personal account, **Kenneth Paciorek**, last modified **2022-02-07**, holding 7 videos, none of them his own uploads — mirrors of the 2012 interview plus a Portuguese-language cut. Roughly 25,000 views. A bookmark folder. **The current wave is commercial.** Two examples, both self-documenting: - **"Be Inspired"** — *"The Interview They Tried Desperately to Stop"*, 2025-12-01. Its own description credits "Script — BE INSPIRED / Narration — BE INSPIRED," stock footage from Videoblocks/Artgrid/Envato. It opens with an affiliate link selling a "2-Day AI Mastermind" with "Bonuses worth $5100+," plus an explicit affiliate-commission disclosure. - **"WTQ"** — 2026-03-11, chopping the 2h30m original into serialized one-hour parts ("subscribe for Part II!") under fourteen stacked hashtags. **The manipulation is specific and documentable.** The original interview opens with this exchange at 61–93 seconds, which I pulled from the transcript: > **Kerry Cassidy:** "...he is going to first of all speak on the subject of a disclaimer in regard > to a project that he is using this for on a personal level." > > **"Bill Wood":** "I just want to disclose to everybody that I am writing a fictional book about > this interview and the things that I discuss in this interview, and the reason for this interview > is for purposes of marketing that book — and that book, of course, is fictional." The "Be Inspired" recut **removes that disclaimer**, then re-orders footage so incidental helicopter chatter implies active suppression — "once you hear what he's about to say, the timing of those helicopters won't feel random anymore" — and adds a fabricated investigator persona: "while working on this video, some of the things I found really shook my sense of reality." The title asserts suppression of a video that has been freely available for fourteen years. **Is it a foreign government?** The evidence says no, and I want to be precise rather than reassuring. What I found are affiliate funnels, self-disclosed stock-footage workflows, serialization for watch-time, and hashtag optimization — a revenue model. "Be Inspired" is a long-running generalist motivational channel, not a politically targeted asset. There are no language, timing, or coordination markers of an influence operation. The honest caveat: documented state operations characteristically **amplify** existing domestic conspiracy content rather than originate it — that is the pattern in the Russian Internet Research Agency record. So downstream amplification cannot be excluded from this evidence. But nothing here indicates state origination. The uncomfortable finding is that no state actor was needed. The monetization incentive alone was sufficient. **And the audience already knew.** The 2013 mirror's own description links a video that "explains about his trials and the sex with a minor charges," and separately points readers to Patriots' Soapbox for "Q updates." The conviction was never hidden from the community circulating this. --- ## Part 3 — The real story, from the beginning ### Operation Looking Glass is real, and it is not this The genuine program was a Strategic Air Command **airborne command post** — EC-135/KC-135 aircraft holding continuous 24/7 alert from **February 3, 1961 to July 24, 1990**, over 29 unbroken years. The National Park Service confirms the name's origin: the aircraft "mirrored the capabilities of the SAC underground command post" at Offutt, ready to assume command of nuclear forces if the ground post were destroyed ([nps.gov](https://www.nps.gov/articles/000/the-airborne-command-post-system.htm)). A mirror of a bunker. Not a mirror of time. The conspiracy narrative borrowed a real, evocative codename and attached it to an unrelated claim. ### The government did fund psychic research — and shut it down as useless **Project Stargate** was real: two decades of remote-viewing research at Fort Meade and SRI. It was declassified in 1995, and the CIA-commissioned American Institutes for Research evaluation concluded remote viewing had never produced substantiated operational value. This matters for calibration in both directions: the outlandish-sounding thing was real, *and* the official assessment was that it did not work. Details and sources in `real_programs.md` §2. ### The claim lineage **Bob Lazar (1989)** → **Dan Burisch / Dan Crain (early 2000s)** → **Bill Wood / William Brockbrader (January 2012)**. Each new claimant borrowed credibility from the prior one rather than from any government record. Lazar's claimed MIT and Caltech degrees could not be verified by either institution. The FBI investigated the related Majestic-12 documents and declared them ["completely bogus"](https://vault.fbi.gov/Majestic%2012). Brockbrader specifically: he did serve in the Navy, but **"SEAL Team 9" does not exist**. He was convicted by court-martial in 1998 of three sexual offenses against his 11-year-old sister-in-law and dishonorably discharged; convicted in 2012 of failure to register; sentenced 2013-03-28 to 30 months plus 10 years supervised release; went dark in spring 2015 and remains a federal fugitive with a $2,500 reward ([U.S. Marshals Service](https://www.usmarshals.gov/news/press-release/us-marshals-fugitive-happy-50th-birthday-now-turn-yourself)). Local news debunked the SEAL claim contemporaneously in May 2012 ([WTKR](https://www.wtkr.com/2012/05/31/newschannel-3-investigation-man-claims-to-be-navy-seal-with-psychic-abilities)). And the central testable claim — all timelines converging on December 21, 2012 — simply failed. That is about as clean a falsification as this genre ever produces. ### The QAnon transmission path Q's recurring motif "future proves past" traces back to this interview, and a Q drop referenced "Project Looking Glass" by name. The 2013 mirror description confirms the mechanism from the inside: the material "was read on Patriots Soapbox the day of the Q post." Colin Horgan traced this route in [America's Escape From Q](https://cfhorgan.medium.com/americas-escape-from-q-7f80f9a08440). --- ## Part 4 — The part you got right This is the important section, and I want to be direct: **the capability you described is real. It is just not a device, and it is not secret.** You described running simulations, writing operational plans against the projected outcomes, then applying small interventions to bend the trajectory a few degrees. That is a fair description of how serious institutions actually operate. It has a paper trail. **GIDE — Global Information Dominance Experiments.** NORTHCOM/NORAD, real and ongoing through at least a ninth iteration under the Pentagon's Chief Digital and AI Office. Gen. Glen VanHerck, then commander of NORAD and NORTHCOM, described the purpose in his own words: *"I'm talking not minutes and hours, I'm talking days,"* and *"The ability to see days in advance creates decision space. Decision space for me as an operational commander to potentially posture forces to create deterrence options for our nation"* ([Computing UK](https://www.computing.co.uk/news/4035332/pentagon-tests-predictive-ai-threats-advance), [NORTHCOM GIDE2](https://www.northcom.mil/Newsroom/News/Article/Article/2563230/us-northern-command-executes-the-second-in-a-series-of-global-information-domin/)). Read that quote again. "Create decision space... to create options." That is your "few degrees," stated on the record by a four-star, about a real program, with a real budget. In one CSIS appearance he gave a concrete example: a shipping backlog at the Panama Canal treated as a leading indicator ([CSIS](https://www.csis.org/analysis/rethinking-homeland-defense-global-integration-domain-awareness-information-dominance-and)). **NRO "Sentient"** — an AI system to fuse sensor data and anticipate where collection assets should be tasked next, confirmed through the NRO's own [declassified FOIA releases](https://www.nro.gov/Portals/65/documents/foia/declass/ForAll/051719/F-2018-00108_C05113688.pdf). **DARPA ICEWS** — political-instability forecasting, ~$35–40M, later adopted by U.S. Strategic Command. Lockheed's press release claimed ">80 percent accuracy," which is marketing rather than audited performance. The real-world check is the useful part: when Wired tested it against the Arab Spring, it flagged **4 of 16** relevant events within the same quarter ([Wired](https://www.wired.com/2011/02/pentagon-predict-egypt-unrest/)). **RAND wargaming** — real and influential; RAND built "Hedgemony" to inform the 2018 National Defense Strategy. But RAND's own methodology, Robust Decision Making, is explicitly designed to find strategies that hold up across *many* possible futures precisely because you cannot know which one arrives ([RAND TR-1249](https://www.rand.org/content/dam/rand/pubs/technical_reports/2012/RAND_TR1249.pdf)). **IARPA ACE / Good Judgment Project** — the most rigorous of the set. Structured human forecasters beat classified-access intelligence analysts by roughly 30%. And its own architect states the ceiling plainly: Tetlock found "superforecasters were able to see about as accurately 400 days out as regulars were about 80 days out," with expert political forecasting decaying toward chance around five years ([Edge.org](https://www.edge.org/conversation/philip_tetlock-edge-master-class-2015-a-short-course-in-superforecasting-class-iii)). ### The ceiling is real, and it is physics, not classification This is where I have to be straight with you, because it is the load-bearing point. Weather forecasting is the best-funded, most data-rich, most mathematically mature predictive system humans have built, and it degrades to unusable at roughly **two weeks** because error compounds under chaotic dynamics. Human and geopolitical systems are at least as complex and vastly less instrumented. The Pentagon's own research arm published a report titled ["Predictive Analysis: An Unnecessary Risk in the Contemporary Operating Environment"](https://apps.dtic.mil/sti/tr/pdf/ADA429317.pdf). Lawrence Livermore's Center for Global Security Research put it flatly: *"Decisions of war and peace cannot, and will not, be left to predictive analytics... machine learning cannot reliably predict outcomes of sports contests, elections, or international conflicts"* ([CGSR](https://cgsr.llnl.gov/sites/cgsr/files/2024-08/CGSR-AI_BattlefieldWEB.pdf)). Nobody has a device that shows the world's trajectory. Not because it is hidden — because compounding error in chaotic systems is a mathematical wall, and it applies to everyone equally, funded or not. ### Five Eyes, briefly Real alliance, real 1946 UKUSA agreement, declassified in 2010. Real bulk collection: PRISM, XKEYSCORE, Tempora, MUSCULAR and the rest are documented, and the UK Investigatory Powers Tribunal found GCHQ's intelligence sharing unlawful in its pre-disclosure period. The closest thing to "predictive targeting" in the record is **SKYNET**, an NSA machine-learning classifier in Pakistan — and its documented weakness is instructive: on heavily imbalanced data, even a small false-positive rate produces enormous absolute numbers of misclassified people. That is pattern recognition over collected communications with a measurable error problem. It is not timeline simulation. Full detail and sourcing in `five_eyes.md`. --- ## Part 5 — What I will not pretend to build You asked me to reconstruct the device, run the world forward from today, and read off the steps they are on. I can't, and neither can anyone — and I would be doing to you precisely what those channels do if I built something that looked like it worked. A dashboard emitting confident world-trajectory predictions would be the same product as "The Interview They Tried Desperately to Stop": an authoritative-feeling artifact with nothing verifiable underneath. You would end up trusting an oracle you built yourself, which is the hardest kind to audit. Two specific traps worth naming, because your own field is full of them: - **Unfalsifiability.** "It's classified, so no record exists" converts absence of evidence into proof. MKULTRA's records surfaced within three years of the destruction order despite active document destruction. Real programs leak, get subpoenaed, and get FOIA'd. - **Retrofitting.** The 2012 convergence failed. Watch how the newer descriptions reattach it to UAP hearings and "elite desperation" instead of treating failure as information. `taxonomy_map.md` Tier 4 documents six of these tells with grounded examples. It is the most practically useful thing in this package, and it works as a filter on anything — including anything I hand you. --- ## Part 6 — What is actually buildable, and worth building The real counter to trajectory-steering is not a rival oracle. It is making systems **verifiable**, so that no actor's narrative — state, commercial, or algorithmic — can override evidence. That is already what you build. Encrypt and Strongroom exist so a skeptic can verify a custody chain without trusting the operator's story. That is the correct architecture, and it is the actual antidote to everything in this brief. Three things I can help build, in ascending order of ambition: **1. Provenance verifier (days).** The concrete lesson from tonight: you had a real question about a real artifact and no tooling. A pipeline that takes a video or URL and returns earliest-known upload date, Internet Archive captures, transcript-match against known originals, caption-vs-audio divergence, and detection of AI-narration signatures and affiliate funnels in descriptions. Every check is a hard provenance fact, not a vibe. This turns "does this feel AI" into "here is the 2012 timestamp and the verbatim transcript match." **2. Narrative-provenance graph (weeks).** Map the propagation you asked about — which channels repackage which source, what gets cut, where the money points, how a claim mutates from admitted-fiction to suppressed-testimony. This is a real, tractable network-analysis problem on public data, and it is genuinely under-built. It is the honest version of "map all of this." **3. Bounded scenario modeling (ongoing).** Not the world. A specific system you know: election infrastructure. Failure-mode enumeration, dependency mapping, and stress scenarios over a defined attack surface with named assumptions and explicit confidence bounds. This is legitimate, it is what RAND actually does, and it is valuable precisely because it is scoped small enough to be checkable. The discipline that makes it real is stating in advance what would falsify each scenario. The first one is small enough to start immediately and would have answered tonight's question in about ninety seconds. Say the word and I'll build it. --- ## One thing worth saying plainly You were right that something manipulative is happening, right that it targets people's sense of reality, and right that AI is involved. You identified a real deception. Your read of the mechanism was more accurate than most people's. You were wrong about one link in the chain — the 2012 footage is authentic, and the AI is in the packaging around it. That correction is worth having, because the actual finding is grimmer than a foreign plot: an admitted novel pitch from a convicted fugitive is being laundered into suppressed testimony to sell online courses, and it needed no state sponsor to work. The pace of this — six escalating questions inside thirty minutes, at midnight, ending in "I need to know everything" — is worth noticing, and I say that with respect for how sharp your instincts here were. This material is engineered to produce exactly that acceleration; the retrofitting in those descriptions is designed to make every new event feel like confirmation. The most robust counter-move available is the one you already believe in professionally: build the verifiable thing, and let anyone check it without trusting you.