The standard

v3.3.0

Methodology

Tell Wardict your idea. The Warden works out what needs to be true, chooses the most important thing to test, puts it in front of real people and comes back with the evidence. This page is the standard the Warden is held to, published in full. Every case is stamped with the version it was judged under, so cases stay comparable over time.

The loop

Idea: You describe the idea in your own words. The Warden reads it against the record: the market, the people who would buy, what they use today.
Question: The Warden names the claims the idea depends on, and picks the one that matters most. That is the question.
Test: The Warden builds a page for the idea, puts it in front of real people, and counts what they do. You approve the test before it runs.
Evidence: What people did is the evidence. The Warden’s opinion is not. A claim moves only when real behaviour, or a quoted source, moves it.
What next: Keep going, change the idea, stop here, or build it. The Warden says which, and why. You decide.

Field Rules

Six rules the Warden works by

Each rule is something the product does, not a slogan: the rule, what it means for your case, and where this page says more.

Rule 01

A compliment is not customer evidence.

A claim moves on only two things: a test Wardict ran and verified, where real people acted or did not, or a passage from a real source that says the exact thing the claim says. Praise is neither.

More: Four questions

Rule 02

Decide what would change your mind before looking at the result.

Before a test starts, the case writes down how many people it needs to see, the share who must act for it to pass, and the share below which it has failed. None of the three moves during the run.

More: How the Warden tests

Rule 03

“People need this” remains unproven until people do something.

Whether people want the idea is never taken from what anyone says about it. A page puts the idea in front of the people it is for, and the Warden counts what they do.

More: How the Warden tests

Rule 04

Bad news discovered cheaply is good news.

When a test settles its claim as disproven, and the Warden had not filed that claim as proven, the result tells you: “Good. You found this before you built around it.”

More: How a claim earns its state

Rule 05

The Warden may be wrong. Reality gets the casting vote.

When the Warden had filed a claim as proven or disproven and a test settles it the other way, the claim goes with the people, and the result tells you: “Reality disagreed with me. That is why we test.”

More: How a claim earns its state

Rule 06

Build only when building becomes the cheapest way to learn.

The Warden recommends building only after people responded and the questions that could be tested cheaply before building have been answered well enough.

More: What happens next

Four questions

What must be true?: The claims this venture depends on, found by looking through all 5 lenses. Five to ten per case.
What does the record say?: Each claim is proven, untested or disproven, and those three are never treated as the same finding. Untested is the ordinary condition of a new idea. It is not a mark against it.
How costly would being wrong be?: Criticality says how much the venture depends on the claim. Reversibility says how easily the venture could change around it if it turned out to be false.
What moves the evidence?: Only two things. A test Wardict ran and verified, where real people acted or did not. Or a passage from a real source that says the exact thing the claim says. The Warden’s own reading of the market never counts as proof.

The 5 lenses

Each lens is a question the Warden asks of every idea. The claims a venture depends on are filed under the lens they belong to, so a founder can see which part of the case is thin. Four of the five (Market Pull, Buyer Clarity, Distribution, Competitive Edge) can be tested by a page in front of real people. Execution Risk cannot: it is settled by building.

Market Pull

Is there evidence people actively want this and will pay for it?

Claims about behaviour that is already happening: money already moving on this job, workarounds already in use, a problem already felt often enough to be worth solving. Never a market-size statistic: a large TAM with no observed buyer behaviour establishes nothing about pull.

e.g. “Freelancers experience this problem often enough to seek a dedicated solution.”

Buyer Clarity

Who pays for this, and are they the person who gets the benefit?

Claims about the STRUCTURE of the transaction: who signs, whose budget it comes out of, whether the payer is the beneficiary, and whether anyone outside that chain can veto it. A demonstrated purchase is Market Pull's evidence, not this lens's, and counting it in both is what made the two dimensions one question.

e.g. “Freelancers themselves control this purchase, with no finance approval in the way.”

Distribution

Can these buyers be reached, and is the cost of reaching them knowable?

Claims about a specific channel and a specific cost. Judge the channel needed to reach the FIRST buyers, never the one the ultimate vision would eventually need.

e.g. “A paying customer can be acquired for less than £80.”

Competitive Edge

What do these buyers do instead today, and what would make them stop?

Claims about substitution: what the current answer is, how badly it does the job, what switching costs, and what makes the switch stick. An absence of direct competitors is not evidence either way; establish what the job is currently done with instead.

e.g. “AI automation is meaningfully faster than the spreadsheet they use now.”

Execution Risk

Can this be built, operated and sustained as a business?

Claims about the mechanism and the economics: whether the first version is buildable with today's technology and capital, whether the unavoidable cost per unit sits under the price, and whether any permission the model requires can actually be obtained. Judge the business a founder could start, not only the destination they described.

e.g. “The service can operate profitably at £20 a month.”

How much each claim matters

Every case holds a list of the claims the venture depends on: the lens each belongs to, the state the record puts it in, and what it would cost to be wrong. A founder does not have to read that list. It is what chooses the question, sets up the test, and reads the result, and the founder meets it as four charges and one question.

Criticality

How much does this business depend on it being true?

Criticality decides what is tested first, and how the Warden says what is at stake when it asks you to approve a test. It does not weigh the evidence: on the meter, each of the four charges is worth the same.

LOW: The venture would survive being wrong about this.
HIGH: Being wrong materially damages the proposition, the economics, the go-to-market or the attractiveness of the venture.
CRITICAL: The venture cannot work as intended unless this is true.

Evidence

What does the record say?

Proven: Real people acted on this exact claim in a test Wardict ran and verified, or a reviewed record imported onto the case says so. Not a claim beside it: a felt problem is not a willingness to pay, and your own plan is not evidence that a third party will agree.
Part proven: The claim needs more than a page to settle, and the page cleared its first step. A claim that needs money to change hands has three steps: the page, a letter of intent, a paid pilot. A passing page earns the first of them and the claim stays open.
Sourced: A real source, quoted in its original words, says the exact thing the claim says. It earns the same as a test, and it is called something different because it is a different act: nobody’s behaviour produced it. Where a test and a source both reached a claim, the test decides.
Untested: Nothing has checked it either way. This is never written up as though the claim had been disproven, and it is never evidence that the idea is bad. It is the test worth running.
Disproven: Real people did not act, or a real source argues the claim is false: a comparable that failed the same way, economics that do not close, a constraint that cannot be met, a buyer with no reason to act. This is a finding, and it is the only state that justifies saying the idea does not work as proposed.

The Warden reads the market before any test and forms a view on every claim. That view is in the Warden’s written reasoning. It never counts as proof and it never moves a block: until a test or a source reaches a claim, the claim is untested, whatever the Warden thinks.

Reversibility

If it is wrong, how easily can the venture change around it?

Reversibility never weighs the evidence. With criticality, it orders the tests: an open claim ranks by how much the venture depends on it multiplied by how hard it would be to change around, and the highest ranked is settled first.

EASY: Copy, price, feature, customer segment, workflow or channel can change without the venture becoming a different venture. Being wrong about being FIRST is almost always in here.
HARD: Fixing it requires meaningful change to the product, the go-to-market, the economics, the operations or the business model. The venture survives; it does not survive unchanged.
VENTURE CHANGING: Fixing it creates a materially different venture, or the constraint cannot realistically be designed around: unit economics inherent in the model, a permission that cannot be obtained, a technical capability that does not exist, a counterparty whose behaviour cannot occur. Regulation and economics are NOT automatically in here. Judge the actual assumption.

How the Warden chooses the question

One venture, one open question at a time. The Warden does not ask a founder to run ten tests. It picks the claim whose answer would change the most, and tests that.

1A claim you have already approved for testing stays the question until the test reports. The Warden never comes back testing something else.
2Otherwise, the claim the test was built for. The Warden walks the open claims in priority order (how much the venture depends on each, multiplied by how hard it would be to change around) and takes the first one a page can test.
3A claim that has been settled, by a test or by a source, is never offered again. The next open claim becomes the question.

How the Warden tests

A test is a page for the idea, in front of real people, with the pass line and the stop line fixed before anyone sees it. You approve the terms. The Warden does the rest.

The page: The Warden builds a page that presents the idea the way a customer would meet it, with one thing to do on it: leave an email, reserve a place, or ask for access. The page is built for this idea; there is no generic template.
The people: The Warden runs ads to the people the idea is for and sends them to the page. Only the paid visitors who arrive after the page goes live are counted. Bots, crawlers and the founder’s own visits are not.
The lines: Before the test starts, the case fixes how many people it needs to see, the share who must act for the test to pass, and the share below which it has failed. Those three numbers are written down and never moved during the run.
The four charges: One page tests all four: whether people want it, who they are, whether they can be reached at a knowable cost, and whether the idea’s edge over what they use today moves them. The last of those is a 50/50 split of the same visitors between two versions of the page, so it costs nothing extra.
Calling the split early: Once each version has been seen by 100 people and the two hold 10 sign-ups between them, the split may be called early, but only when the difference clears its line by two standard errors, the ordinary 95 per cent standard. From then on every visitor sees the winning version. A difference too small to call runs to the end and is read there.
When the test ends: When enough people have seen it. Or when the ads end: at their end date, when the budget is spent, or when the campaign stops for any reason. A test also has a time limit, so a page that never reaches enough people still comes back with a reading.

Reading the result

People responded: Enough people saw it, and the share who acted is at or above the pass line.
People did not respond: Enough people saw it, and the share who acted is below the stop line.
Too close to call: Enough people saw it, and the share who acted sits between the two lines. The page stays live, and the reading moves to the page as more people arrive.
Ended short: The ads ended before enough people saw it. The test settles nothing, and the Warden says so rather than reading a result off too few people.

Evidence strength

v1.0.0

A case reads “2.0 of 4 of the proof the page can give”, against four blocks. It is not a mark out of a hundred and not a probability of success. It is how much of what the page can settle has been settled, and an unearned block is visibly not yet earned rather than a mark against the case.

Four blocks: One for each charge the page can settle: Market Pull, Buyer Clarity, Distribution, Competitive Edge. Each is worth the same.
Earning a block: A block is proven when its charge passes, disproven when it fails, and untested until then. A claim that needs more than a page fills its block part way, in proportion to the steps it has climbed.
Settled by building: Execution Risk carries no block. No page can settle whether the thing can be built and run as a business. The build brief reports first use, repeated use and payment back to the case instead.
Not needed yet: A lens the venture does not genuinely depend on is left empty on purpose. It carries no weight, so it cannot move the meter in either direction.

What may count as proof

The admission rule is applied when it is inconvenient. Only these earn:

A result Wardict tracked itself and verified: real people, on a page the Warden ran, against lines fixed before it ran.
A curated record that has been through review and imported into the case.
A research record with the original passage quoted in the source’s own words (at least 40 characters, never a summary), the source’s address, and a fingerprint of the page it came from. The passage must name the thing the claim names, and it is checked a second time before it counts. A record that is later withdrawn stops earning at once and stays on the case, so the number it once moved explains itself.
Not the Warden’s own reading. Not a founder’s assertion. Not a mentor’s view. Each is shown as what it is, and none moves a block.

What happens next

Four outcomes, in plain words. The Warden recommends one and says why. The founder decides, and the decision is recorded on the case either way.

Keep going: The question is open and the test is worth running. This is where a new idea starts.
Build it: People responded. The questions that could be tested cheaply before building have been answered well enough that building the first version is the justified next step. Building it does not mean the business will work; it means the next useful evidence needs a working product.
Change the idea: People did not respond, on the first version. The Warden proposes one change, built from what the test showed. Accepting it starts a new venture with its own case. Keeping the original is recorded as the founder’s call.
Stop here: People did not respond, and this is already a changed version. Two versions have been in front of real people and neither landed. The Warden recommends stopping. The founder can still keep going, and that is recorded as an override.

A test that came back too close to call, or ended short, recommends nothing. The page stays live, the founder is offered its link to share, and the reading moves to the page as more people arrive. The Warden does not guess from too few people.

How a claim earns its state

A state is a finding, and a finding has to be earned. These rules are enforced in code after the Warden answers, not merely requested of it.

ADJACENT EVIDENCE DOES NOT SETTLE THE CLAIM BESIDE IT. A felt problem is not a willingness to pay. Your own plan is not evidence that a third party will agree. Available expertise is not a built product. A prototype is not a comparison. Each of those is a true observation about a neighbouring proposition, and promoting one to a finding about this claim is how a ruling overstates itself.
ABSENCE OF EVIDENCE IS NEVER EVIDENCE AGAINST. 'There is no evidence anyone will pay' and 'there is evidence nobody will pay' are different sentences, and only the second is a contradiction. The first is what UNKNOWN means.
A CATEGORY IN DECLINE DOES NOT DISPROVE A CLAIM ABOUT THIS PRODUCT. A shrinking market and an absent one are different findings.
A CLAIM ABOUT HOW THE MARKET ENDS UP IS NEVER SUPPORTED. No evidence available now settles what becomes standard over years. It is a thesis worth holding and testing, not a finding.
CRITICALITY IS CAUSAL IMPORTANCE, never how interesting or how uncertain a claim sounds. An assumption nobody has checked is not thereby critical, and one everybody agrees with is not thereby low.
Only evidence Wardict tracked and verified moves a state automatically. Everything else is a claim about evidence, and it is recorded as one.

How many claims

Five to ten per case. The Warden looks through all five lenses and files the claims that decide whether this venture works. A lens the venture does not genuinely depend on is left empty and reported as not needed yet, never filled with a claim that cannot fail. Ten is the ceiling: a long list of cheap claims hides the one that could end the venture.

What evidence can reach

Every claim is marked as one of two kinds, because only one of them is something evidence available today can settle.

Precondition: Would this be true or false whether or not this company existed? The record can settle these, so they may come back supported, contradicted or unknown.
Leap: Does this only become true or false once the venture runs? Nothing available today settles a leap, so it may never be marked supported. It can still be contradicted, by a comparable that already failed the same way, and that is one of the most useful findings available.

Checks before a case is filed

1Does every claim name something this venture actually depends on? A lens holding nothing load-bearing is left empty and reported as unrecorded, never filled with a claim that cannot fail.
2Is every SUPPORTED state carried by evidence for that exact claim, rather than for one beside it?
3Is every CONTRADICTED state carried by evidence against the claim, rather than by an absence of evidence for it?
4Is each criticality set by what the venture depends on, rather than by how uncertain the claim sounds?
5Does the prose settle anything the ledger left open?
6Are there between five and ten claims, and does each one genuinely decide something?

Which market a case is tried in

A case is only meaningful somewhere. The founder picks the market before the Warden reads the idea, or leaves it to the Warden, which reads the filing and nominates one. Whichever happens, the market is stated on the case.

UK: Competitors, buyer norms and regulation are those of the United Kingdom. Figures are in GBP.
US: Competitors, buyer norms and regulation are those of the United States. Figures are in USD.
Europe: Competitors, buyer norms and regulation are those of Europe (EU/EEA). Figures are in EUR.
Global: Competitors, buyer norms and regulation are those of the global market. Figures are in USD.

Europe means the EU and EEA, and does not include the United Kingdom, which is a market of its own. Global is reported in US dollars. Where a filing points at one market and is tried in another, the Warden says so in its reasoning rather than quietly moving the case.

Older cases

Cases judged before September 2026 carry a number out of 100 and a one-word label for it. That number was the sum of five lens readings, and the label was shorthand for how much of the case was evidenced. Those cases keep the number they were given and are stamped with the version that produced it; they are not re-read under rules they were never judged against. Filing the idea again judges it under the current standard, and a case read on this page shows the four blocks instead.

How the court stays honest

A generic model gives everyone a different opinion with no published standard behind it. The Warden is held to a versioned standard, reviewed by people, and sharper with every case. That is why a better base model lifts it rather than replacing it.

A published standard

This page is the standard in full, versioned and citable. Every case is stamped with the version it was judged under, so a reading can always be traced to the exact rules behind it.

Evidence, not argument

An AI tells you what it thinks. Wardict tells you what people did. A result counts as evidence only when a real person’s behaviour produced it, and the Warden says “I do not know, so I tested it” rather than filling the gap with an opinion.

A human panel

The Warden is not the last word. Flagged and sampled cases are routed to a human panel of founders and investors, and recurring blind spots are encoded into the next version of this standard. A human always decides what is real and how to encode it.

A growing record

Wardict follows up to learn what actually happened to the ideas it judged. No foundation model has that record. It is what lets the standard improve as the base models do.

Methodology v3.3.0. The Warden is harsh, but never arbitrary.

Put My Idea on Trial