GUBMENTPlain talk · policy frontier
Filings / About
Form GBMT-0 · The desk

We write our guesses down before we check them

Gubment is an independent policy-research desk. One folder at a time, we take a policy area everyone argues about, measure the thing that actually binds it, and publish the whole record — including the parts where we turned out to be wrong.

Part 1

What this is

Most policy arguments are conducted about the wrong quantity. Not dishonestly — just because some numbers are easier to see than others, and the easy ones get counted.

Four filings in, that pattern has held every single time. Childcare's fight is about the price tag; the binding constraint is three million missing educators. Housing argues about a national shortage number that turns out to be a near-linear restatement of an assumed vacancy rate. Drug policy argues about the illegal drugs; alcohol and tobacco kill several times more people annually using policy levers nobody contests. Elder care assumes Medicare pays for long-term care, and it mostly doesn't.

So the work is narrow and repetitive on purpose: take one area, find the quantity that actually constrains it, check whether the popular number survives contact with its own sources, and say plainly what would have to be true for the leading proposals to work. Every claim carries a source. Every verdict carries a stamp. Every guess we got wrong stays in the record where you can see it.

Part 2

The method

Every filing runs the same protocol, written down before the research starts. The full version lives in the repository as a document called the gubment method; this is what it means in practice.

Write the guesses down first. Before any research, every prior we're carrying goes into an anchor table — stated plainly, as a number, where it can be checked. When the research breaks one, the original guess stays on the page next to what turned out to be true. Being wrong in public is the point; a method that only shows its wins isn't showing you anything.
Two sources, of different kinds. A load-bearing claim needs corroboration from two sources that don't share a root — administrative data and peer review, say, not two articles citing the same press release. Much of what circulates as settled fact is one source wearing several coats, and tracing citations back to their origin is most of the work.
Decide what would prove us wrong, before looking. Each hypothesis gets written adjudication criteria — refuted if, supported if, indeterminate if — fixed in advance. It's much harder to move the goalposts when you've already published where they stand.
Search the literature; don't recall it. Named sources are a starting point, not the evidence base. Remembering a study is not the same as reading one, and the difference shows up as errors.
A dead premise is the headline, not a footnote. Each filing declares kill conditions up front — findings that, if they landed, would invalidate the question itself. When one fires, it leads. Elder care's central dollar figure turned out to be unreconcilable across credible sources; that became the story rather than an asterisk on it.
Argue against ourselves before publishing. Every filing gets a written red-team pass attacking its own conclusions as hard as the evidence allows. Attacks that land are absorbed into the findings and logged, not quietly resolved. Where a hypothesis leans in a fashionable direction, we owe the unfashionable side an equally serious case, built from its best evidence rather than its worst.
Stamps mean something specific. A stamp only appears where a verdict was actually adjudicated against the criteria. Nothing gets stamped VERIFIED for looking probable.
Part 3

Who makes it

Gubment is an independent research desk. It is not a think tank, a news organization, a university project, or an advocacy group. It has no clients, no institutional affiliation, and no policy program it is trying to advance.

Filings are published under the desk's name rather than individual bylines. That's a deliberate choice about what should carry the weight: the protocol, the sources, and the record are auditable in a way a résumé isn't. Everything the desk produces — protocols, raw findings, deviation logs, the scoring behind every scorecard — is public in the repository, so a reader can check the work rather than trust the author.

Part 4

Funding

There isn't any. No advertising, no sponsorship, no grants, no subscriptions, no institutional backing. Nothing on this site is paid for by anyone with an interest in what it concludes.

We state this because our own filings spend a lot of time flagging who funded other people's research — the most-cited pro-decriminalization source in American drug policy turns out to be an advocacy report paid for by a legalization group; the leading anti-legalization organization's political arm files a fully redacted donor schedule. Neither fact makes those sources worthless. Both are things a reader deserves to know. So here's ours.

Part 5

How to read a filing

Every whitepaper uses the same furniture. Four pieces are worth knowing:

The anchor table holds every prior we wrote down before researching, next to what we found and how far off we were. The honesty box is the appendix naming the filing's own weakest points — what's single-sourced, untested, or unresolved. The deviations log records every departure from the protocol as written, with its reason and its effect on the findings. The receipts strip at the foot of each filing summarizes what was actually run.

And then there are the stamps. They're not decoration — each is a specific verdict reached against pre-written criteria.

How strong is the verdict

VerifiedChecked, corroborated, and it held.
IndeterminateThe evidence genuinely can't settle it. Not a hedge — a finding.
Contested, not settledCredible researchers disagree, and the disagreement is real rather than manufactured.
Consistent, not confirmedThe evidence fits the theory but doesn't rule out rival explanations.
Directional, not finalWe're confident which way it points, not how far.

When we were wrong

Prior brokenA guess we wrote down before researching turned out to be false.
Kill condition firedSomething we declared would invalidate the question actually happened.
Corrected via red teamOur own adversarial pass caught an error before publication, and we're showing it rather than quietly fixing it.

How good is the underlying evidence

UnreplicatedOne study, never independently reproduced.
Single-rootEvery source traces back to the same origin, however many outlets repeat it.
Untested at scaleIt may work; nobody has tried it big enough to know.
Confounded natural experimentA real-world test existed, but something else changed at the same time.

What the scorecard concluded

Survives all rankingsWins regardless of which objective you weight most heavily.
Survives independent re-scoreA second scorer, working blind, reached the same result.
Clearest first moveThe best-evidenced place to start — deliberately not the same claim as "the answer."

Whether we argued fairly

Steelman appliedThe opposing case was built from its strongest evidence, not its weakest.
Mixed steelmanThe strongest version of the opposing case partly survived.
Strong comparatorThe do-nothing baseline is genuinely hard to beat, and was scored honestly rather than as a strawman.

What it would take to act

Binding constraintThe thing actually capping outcomes. Relieve it and output moves; fund around it and it won't.
Rulemaking-executableAn agency can already do this. No new statute required.
Part 6

Corrections

Corrections stay in the record. Nothing here is silently edited — when a finding changes, the original and the correction sit side by side, because a research record that quietly improves itself is indistinguishable from one that was always right, and those should not look the same.

Two real examples, both still visible in their filings and on the public corrections log. Elder care's own protocol cited a superseded figure — that 70% of Americans turning 65 will need long-term care — which the research corrected to 56%; the wrong number stays in the anchor table. The media filing's first scorecard draft claimed one architecture won under every weighting; the red-team pass found the score behind it had never actually been verified, and the claim was withdrawn rather than defended.

If you think something here is wrong, the useful thing to send is the source. See below. Full log: www.gubment.com/corrections/.

Part 7

What's coming

Nine policy areas, one folder at a time. Four are published:

GBMT-1ChildcareWhitepaper No. 1
GBMT-7Elder careWhitepaper No. 7
GBMT-8DrugsWhitepaper No. 8
GBMT-9MediaWhitepaper No. 9
GBMT-2HousingResearched, unpublished
GBMT-3Health careIn the queue
GBMT-4EnergyIn the queue
GBMT-5ImmigrationIn the queue
GBMT-6TaxesIn the queue

Housing is further along than its unpublished status suggests — most of its workstreams are complete and committed, including the finding that the national shortage figure everyone cites reproduces almost exactly from its own assumed vacancy rate. It needs its final synthesis before it ships.

Part 8

Contact

Corrections, challenges, and source tips: [email protected].

The most useful message is a specific one — the claim you think is wrong, and the source that shows it. That goes into the record the same way our own corrections do. Every filing's underlying research is public in the repository if you'd rather check the work directly.

THE DESK · Protocol written before research · every prior recorded and checked · two-source rule on load-bearing claims · adjudication criteria fixed in advance · red team before conclusions · corrections kept in the record · no funding, no clients, no affiliation.