Audio
Episodes
Deep dives and the week in review, three times a week. 41 published.
0x13D Research
We check what was said, and keep the receipt.
A fact-checking operation and the tools it had to build to do the work. Every claim is quoted verbatim, adjudicated with its evidence, and filed at the second it was said — then joined to every other show that ran the same claim.
17shows
160episodes
9,267published claims
12tracked narratives
Published counts are baked in at build time by scripts/refresh-figures.mjs.
Figure 01 — Technique
What the corpus is actually made of.
The persuasion techniques we see most, across 10,489 labelled claims. Each is a SemEval-2023 label citing a verbatim span checked against the quote it came from; five cross-walk onto DISARM technique IDs, which is how this leaves our vocabulary and enters someone else’s.
The pass that assigns these reads the claim and its quote, and never receives the verdict — so a technique label describes how something was said, not an inference from whether it was true.
Figure 02 — The last four weeks
Where the week landed, and how hard it worked.
One dot per claim, placed by how well the evidence supported it. The marker on each row is the share of that week’s claims carrying at least one persuasion technique — both are proportions, so they read against the same ruler.
cycling · click a week to pin
0 · refuted0.51.0 · supported
A claim, placed by evidence
Share carrying technique
Dots are a fixed 150-claim sample per week, drawn from that week’s real distribution of evidence scores; the counts beside each row are the true totals. Technique share is measured over the claims labelled so far that week, by a pass that never receives the verdict.
Open the archive to query every claim behind these figures — the whole corpus, not the four weeks.
01 — Work
Six projects, one archive
The show is the reason the rest exists. Everything below reads from the same corpus of claims; the difference between them is who is holding it and what they are trying to see.
-
01
The Narrative Networks Podcast shipping
The show · deep dives & week in review
Three times a week. A single claim taken apart to the source, with the verdict, the truth score and the strongest version of the other side all said out loud rather than implied.
Named for the DARPA research programme on how stories act on group behaviour — a borrowed term, and we say whose.
-
02
Project Ariadne shipping
The archive and its public reader
Every published claim, its evidence, and the tracked narratives that run underneath them — threads that connect claims whose surface tags never touch. The whole corpus is queried in your browser; no account, no server round trip per keystroke.
Named for the thread. The English word clue is the same word as clew, a ball of thread, and its figurative sense is attested from the 1590s in reference to the one Ariadne gave Theseus.
-
03
Theseus Terminal shipping
The desktop control room
Where the pipeline is run and the calls are made: transcription, claim extraction, adjudication, thread promotion, and the review queue that decides what may be published. Internal.
The maze is not solved by strength; it is solved by keeping a record of the path.
-
04
Project Aperio prototype
Browser extension · Chrome, Edge
The archive in the YouTube progress bar. It works out which episode you are watching, puts every checked claim on the video's own timeline, and — the part nothing else does — names the other shows that ran the same claim.
It changes nothing on the page unless you ask it to. Latin aperire, to open or uncover: the root of both aperture and overt.
-
05
Project Stultus new
The workbook · browser
A notebook that knows what its pages are. Each page has a kind — math, philosophy, logic, boolean, recipe, carpentry, history — and every kind declares what it builds on and what it leads to, so a collection is a graph you can walk rather than a folder you have to remember the shape of. Prints to LaTeX.
It stores your work in the browser and sends none of it anywhere. There is exactly one network call in the whole application, it downloads map tiles, and it asks first. That is a test that fails the build, not a promise in a policy — every other way of reaching the network is banned outright by name.
Latin stultus, foolish: the learner who does not know yet, which is who a workbook is for.
-
06
Project Ares in development
Influence-strategy game
Play the other side. A capability tree, doctrine points and heat, run over a globe — the narrative engines we document in the show, made operable so their shape is legible from the inside.
02 — Publications
What we put out
Video
Shorts
The video cuts — one claim, taken apart, in the time it takes to scroll past it.
Reference
Encyclopedia
Where our vocabulary came from and whether it is ours. One entry per coinage, with its sourcing and its honest flags — including the ones that are weakly sourced.
Zines
Two titles, A5, printed rather than posted. The Little Green Book of Statecraft is an annual: eight spreads, each pairing one influence capability against one tracked narrative — the capability on the verso, the narrative it produces on the recto. The 36 capabilities we could not pair are printed as an index, because the gap describes our instrument rather than the world. Influence Campaigns is quarterly, four essays an issue, no two running the same engine.
Both carry the same back cover: a QR code over the words Steal This Zine, after Abbie Hoffman's Steal This Book (1971). Neither has been to press.
03 — Method
The standard we hold ourselves to
Specificity replaces adjectives. A file path, a date, a named actor or a real number is a better sentence than a superlative, and a claim that cannot be pointed at a document or a public record gets cut.
We do not rent authority. Our own vocabulary is never attributed to unnamed outside experts — not media analysts refer to it as, not what scientists would call. If we coined it, we say so. If someone else did, we name them. An operation that checks other people's claims does not get to invent citations for its own words, and we have broken this rule and written down where.
The verdict goes in the copy. The most-dropped element in our own audited episodes was the adjudication itself — a brief reading false, truth score 0.04 and an episode containing no false and no score. So we read our own episodes back against the brief, and the gaps are logged like anyone else's.
An honest gap beats a plausible guess. Where we cannot locate a claim to the word, or cannot verify the audio it sits in, we say nothing rather than something that looks right.