Every claim in a WorldbyFlow scan carries a grade for how it is actually known — not how confident it sounds. The grades are assigned by the same rules on every scan, the rollup numbers you see are computed from the graded rows rather than asserted, and a source link renders only when the record genuinely carries one.
This page is the complete vocabulary, rendered from the same definitions the product enforces — it cannot drift from what you see on a scan, because it is generated from the identical code.
Every scan carries a Facts ledger — each discrete claim with the honest basis it rests on. These are the nine words that can appear beside a fact, from strongest to weakest.
The strongest thing this product will say about a claim. The scan did not just find it — it went back out and checked it against a source it names, and that source is attached to the row. Adversarial verification passes and the Inquiry ledger both land here, as do verbatim quotes and document excerpts that carry a real URL.
The claim traces to something the scan fetched and read during the run — reporting, a filing, a dataset, a document — rather than to model memory. A source is named. What is missing is a second, independent check, which is the only thing separating this from verified.
Real support exists but it does not close the question. In practice this is one of three situations: a single source with nothing corroborating it, a figure the interested party reported about itself, or a result too recent for anyone else to have tested. Usable, but say where it came from when you repeat it.
The evidence does not agree with itself. Either two credible sources state it differently, or the record directly contradicts the claim as stated. This is not a weaker fact — it is a finding in its own right, and the disagreement is usually the thing worth acting on.
Someone said it and the scan carried it forward, but nothing in the record supports it. Treat it as a claim attributable to whoever made it, never as an established fact. Analyses that grade their own evidence mark these "asserted, not evidenced".
The claim appears in the source material but arrives without a source marker of its own, so there is nothing to grade. Some analyses call this "untestable from the record". It is reported as-is, with no claim made about how well supported it is.
This came from what you typed into the scan — a premise, a figure, a piece of context. The analysis built on it without checking it, because you asserted it. If it turns out to be wrong, everything downstream of it is worth re-reading.
Background the model already knew rather than something it read during this run. It is kept because it is usually load-bearing context, and it is labelled because nothing about the run confirms it. Where the schema requires one, a ready-made search rides along so you can check it in one tap.
Not a claim at all — the opposite. The run states a hole in the record: a figure nobody publishes, a party that would not comment, a document that blocks automated access. It is listed because knowing what could not be established is part of the finding, and because a gap left silent reads as a subject with nothing to say.
Where a scan summarises its ledger in a line, the nine grades collapse into three groups that always sum to the total. Named gaps sit outside the count — an absence is not a claim.
Combines the two strongest grades. These are the claims you can carry forward without qualifying them, though grounded still lacks the second independent check that verified has.
Everything between the two ends: partial support, claims attributable to whoever asserted them, the model's own background knowledge, and figures you supplied yourself. Usable, but repeat them with attribution rather than as established fact — which is precisely why the group is named rather than left as a remainder.
Kept as its own group because a disagreement is a finding, not a weaker fact — and it is usually the part of a scan worth acting on first.
Facts persist across everything you run. When two scans speak to the same claim, the corpus says which of four things happened — and never merges the rows to hide it.
One scan established it. Most of the corpus looks like this, and that is fine — it just means nothing else you have run has spoken to it yet.
Two or more of your scans, run separately, arrived at the same claim with the same figures. Neither scan could know that; the corpus is the only place it shows up.
The same claim, a different number, and enough time between them that the figure moved rather than the record disagreeing with itself. This is a trend, not a problem — the newest reading is the one to use, and the older ones are the shape of the change.
Two or more of your scans state the same claim with different numbers over the same period, so one of them is wrong. This is the single most valuable thing this page produces — worth settling before you act on either figure.
Saying a moment was pivotal, a lesson transfers, or a failure cascades is a claim about causation — so it carries its own grade for how the link is known. "Inferred" exists so the analysis has somewhere truthful to put its own reasoning instead of dressing it up.
A source in the record states this causal link — a finding, a filing, an official account.
A named participant, analyst or institution asserts the link on the record.
The scan's own causal reading from the sequence or structure, labelled as its own.
Credible sources genuinely disagree about whether this link holds.
A Literature Review grades every finding by the convergence of the published record behind it. One study is not a literature — and three papers by one team over one dataset are one measurement published three times, which is what the independence check exists to catch.
At least two INDEPENDENT named studies — different teams or datasets — point the same direction. The only grade that earns "the literature shows".
One named study only. A real measurement, but one measurement — never "the literature".
A preprint, pilot, working paper or small-n result that has not been replicated or peer-validated. Marked as what it is.
Named studies genuinely disagree. The split is the finding; the sides are named, never averaged.
A published critique, failed replication, or retraction lead targets the supporting record. A finding, not noise.
The claim circulates — in coverage, marketing or folklore — without a study behind it that this review could find.
Before any finding is reported, the review states whether the literature studied the question as asked — because most applied questions were not, and evidence launders across that boundary constantly.
The literature studied this question as asked.
The literature studied related questions — proxies, narrower populations, different timescales. Every finding names what was actually studied.
No study addresses the question. Saying so IS the review.
Different teams and different data. Real convergence.
The supporting studies share authors — one group repeating its own result, not convergence.
The supporting studies analyze the same underlying data — one measurement, several papers.
Everything traces to one study or report.
The record does not let independence be established — treated as NOT independent.
When the subject is a method — a framework, a technique, a protocol — every claim of what it achieves is graded by the evidence behind it. Adoption is not evidence: "widely used" is a popularity fact, and a vendor case study reads identically to an independent evaluation in coverage.
A NAMED independent study, trial, evaluation or dataset measured the outcome. The strongest grade there is.
A NAMED record documents the result in a specific real adoption — not a controlled study, but a checkable record.
Named practitioners report it from their own use. Real testimony, uncontrolled — experience, not evidence.
The claim originates with the method's own originator, vendor, certifier or advocates, and the record does not independently support it.
The record actively disagrees — a failed replication, a contrary evaluation, or named critics with evidence. A finding, not noise.
No basis found either way. The claim circulates without a record behind it.
Each scan family carries one absolute rule that its output is structurally unable to break — the grade ladders above are how those rules reach the page.
A Literature Review may call a finding convergent only when independent teams on independent data point the same direction — and it says so when they do not.
A Method Study grades "widely used" as a popularity fact. A claim of what a method achieves needs a named study or a named record behind it.
A response read never counts an announcement as an arrival. Nine commitment stages separate promised from actually there.
Before a statistic’s move is attributed to anything in the world, the scan checks whether the counting itself changed — a redefinition, a backlog, a base effect.
When sources or studies genuinely disagree, the disagreement is the finding. Both sides are named; no synthetic middle is invented.
An appraisal scores your alternatives on your criteria, fixed before any option is examined — a yardstick fitted after the fact is the tell of a picked answer.
A source link renders only from a URL the analysis genuinely carries. A citation the record does not contain is never reconstructed from similarity.
Every grading ladder ends in an honest floor. A claim with nothing behind it is reported as exactly that — never rounded up to fill a section.
The grades are not a promise of correctness — they are a promise of honesty about how each claim is known. The rest is on the page, where you can check it.