TruthFoundry 1.36: a public dataset becomes a section, and every story that runs is kept
Hand the paper a public dataset and it can run as its own section, on its own clock, with its publisher named on every figure. US Treasury outlay statements are the first one and they are live. Everything the paper's own desks write is kept in the permanent archive, and a reader can now mark a fact as corroborated or refuted, with a reason.
TruthFoundry writes a daily paper from a signed fact record and answers claims against it. The last release was about the evidence underneath a verdict. This one is about the paper itself: what it can take in, what it keeps, what it is willing to assert, how it reads, and what a reader can now add to it.
A public dataset becomes a section of the paper
Give the paper a public dataset and it can run as its own section: its own clock, its own publisher named on every figure it produces, and its readings held in the record beside everything else the paper carries.
The first one is live. The US Treasury Monthly Treasury Statement now runs on the Government Spending page, as a dated outlay statement per federal agency: net outlays for the month, the fiscal year to date, and the same point in the prior fiscal year. Every figure states the period Treasury published it for and the moment we last read it.
Two sources are never merged into one number. Where the same agency turns up in more than one source, the page shows both readings side by side, each with its own date and its own publisher, and it says how the two were matched and how strong that match is. Two true figures that disagree tell a reader something. One averaged figure tells them less.
The honest part is what the work actually is. Standing a section up is mostly not machinery. It is writing down what the file means: what each column holds, what the units are, how often it changes, and who published it. Get that description right and the section follows from it. That is the part we do with a publisher, and it is the part worth their time.
The Government Spending page itself reads as a page rather than a grid, a running account you can read straight down, where every figure states its source, its period, and when it was read.
A reader can mark a fact
On a key point in the paper, a reader can now record that they corroborate it or that they refute it. A reason is required. A mark with no reason is an opinion with a number attached, and the record does not need more of those.
What a mark is not is a verdict. A count of marks never blends into anything about whether a claim is true, it never moves a score, and it is never dressed as one. Reader marks sit beside the record, labelled as exactly what they are: readers, on the record, with reasons.
A point nobody has marked shows nothing at all. Printing a zero would assert an absence of disagreement that nobody has measured.
Every story that runs is kept
The paper has its own desks, and those desks write. Government spending, markets, science, sport. What a desk files now belongs to the permanent archive along with the rest of that day's paper, so opening a past edition gives you the paper as it actually ran rather than a part of it.
Each of those stories carries the same kind of receipt as everything else in the archive, put together the same way, so one check covers any story in the record. A reader does not need to know which desk wrote a piece, or how it was assembled, to check it.
A receipt proves who published a story and that it has not changed since. It does not make the story correct. What it buys is that our name is on it permanently, and we cannot quietly alter it, backdate it or deny having run it.
Editions that have already gone to press are left exactly as they went out. We do not reopen a finished paper to slot a story into it, because re-cutting a sealed edition to make the record look tidier is the one move this product exists to refuse. The papers that ran stand as they were printed, and the archive from here holds the whole paper.
The desks state only what they can point to
The real hazard in a newsroom is not invention. It is reasonable analysis promoted into flat assertion: a sentence that reads like a finding when it is actually a reading. Every desk now writes under one rule, and the test is whether a reader could check the sentence against the record. If they could, it is a finding and it has to be exact. If they could not, it is a reading and it has to look like one.
In practice that shows up in specific, small ways. Analysis is welcome and is marked as analysis instead of dressed as fact. Something we did not find is described as not publicly announced, never as proof that it does not exist. A judgement carries either a hedge or a fact in the same breath. What has not been established is stated as unestablished, and is never implied to be false.
The rule sits under the paper's voice rather than over it. A newsroom running its own edition can set how the paper sounds. It cannot change what the paper is allowed to assert.
A statistic runs only when it can stand up
Numbers get the same treatment. A percentile or a share runs only when the population it was cut from is stated and is large enough to mean anything: a 99th percentile does not name a point distinct from the maximum until there are at least a hundred observations under it. Where that is not true the figure is dropped rather than softened, and the story states the plain observed amount instead, which is always defensible.
A hedge on a statistic that cannot support itself is not honesty. It is the same claim in a quieter voice.
That rule now covers the section fronts as well as the stories. On the Government Spending page, a ranking against a population runs only where the population is stated and has at least a hundred observations in it. Where it does not, the plain observed amount stands in its place.
A corroboration count counts what is on the page
When a story says how many publishers carried something, that count is derived from the evidence shown on the page: distinct sources, distinct articles, the ones you can see and follow. It cannot exceed them.
A story also stays on its own subject, drawing its witnesses from reporting about that subject and nothing else. Corroboration is the thing we sell, so a count that reaches past the page for its total would be worse than no count at all.
Sections read as sections
The section strip across the top of the paper is a real nav now: pills, each with its own hand-drawn icon, seven sections in the row and the rest one press away. The section you are reading is always in the visible row. Government Spending leads it.
The icons are drawn on one grid and stroked in the text colour, so a section reads the same on screen and on the printed sheet, and they hold their shape at a phone's width.
A correction travels with the story
When a story is corrected, the reader gets both halves: the corrected text and the notice that it was corrected, on the story itself. A corrected story showing no sign of having been corrected is worse than either the original or the fix, because the record then looks clean when it is not.
The notice is compact and opens on a click, so it is always visible without taking over the story, and it links to the version it replaced. The text that ran is still reachable, next to the text that replaced it.
A shared story arrives as that story
Post a story link into a chat, a channel or a feed and the preview that unfurls is that story: its own headline, its own summary, its own card. Section fronts do the same.
The page behind the link is built and complete when it arrives rather than assembling itself while you watch, and an agent asked to read a section front is handed the section rather than a stub.
The numbers on a section front hold still
A section front states figures: what the desk covered, how much was obligated, how many sources carried it. Those figures are fixed when the page is composed and stay fixed while you read it. Nothing recalculates under you, and the number in the headline is the number in the story.
Analytics run only if you accept them
Public pages ask before measuring anything. Nothing loads until you accept, declining is remembered, and a Cookie settings link in the footer reopens the choice whenever you want it. We measure which pages are read so we can tell what is worth writing, and the privacy policy names what is collected, who processes it, and how long it is kept. A paper published under someone else's name never carries our measurement.
Read today's paper free at truthfoundry.ai, open a past edition from the archive, or check a claim at truthfoundry.ai/fact-check. Reading is free, no signup.
Alpha software under active development.
Published by
Robert Christian
Founder and CEO, DRM3 Labs Corp.
More from DRM3 Labs
DomainDrift 1.86: alerts reach your channel, and a parked domain is a reading too
Robert Christian · 3 min read
TruthFoundry 1.12: see what carries a claim and what cuts against it
Robert Christian · 2 min read
TruthFoundry 1.11: check a claim, read the verdict in seconds
Robert Christian · 2 min read
2026 DRM3 Labs Corp. All rights reserved. DRM3 Labs builds infrastructure for open protocols.
This article is for informational purposes only. Nothing here is financial, investment, or legal advice. Tokens, staking, NFTs, and blockchain protocols are described as technical mechanisms, not investment recommendations. Digital assets carry risk. Do your own research.
Many DRM3 products mentioned are in early alpha. Features, availability, and economics are subject to change. References to the Morpheus network describe the public protocol as documented at mor.org.
