🚩 The verification banner
Every STORM report ends Phase 4 by stamping a verification banner at the top. It’s public proof that the pipeline actually ran — a short scorecard showing how many citations were checked and what changed in the process. Without this genuine banner, what you have isn’t a STORM briefing; it’s a well-formatted opinion.
📊 What the banner measures
- •N/N checked — how many citations passed an independent checker (ideally all of them).
- •X fabricated — claims whose number/study didn’t exist and were cut.
- •Y corrected — number, author, year, or characterization that was wrong and has been corrected.
- •Z downgraded — weak evidence that lost points or was moved to the contested signal bar.
🟡 New here? — fabricated / corrected / downgraded
- Fabricated: the primary source doesn’t exist or doesn’t say that. It can’t be "corrected"—it’s removed from the report.
- Corrected: the source exists, but some detail (the value, the year, who signed it) was wrong and has been corrected.
- Demoted: the source is real and accurate, but weak (preprint, single study). The claim remains, with less weight.
VERIFICAÇÃO · 14/14 checadas 1 fabricada · 3 corrigidas · 2 rebaixadas
✓ Honest banner
- ✓Reflects what actually happened in Phase 4
- ✓Admits when something was fabricated or downgraded
- ✓The score matches the reference tags
✗ Cosmetic banner
- ✗"Everything verified" without anyone having checked
- ✗Set fabricated/downgraded to zero to make it look clean
- ✗Doesn't match the tags below
🏷️ Status tags by citation
The banner is the summary; the citation tags are the details. Each reference in the final section carries a visible verdict, so you know where each claim stands, item by item. An independent verifier returns one of these six statuses:
The primary source supports the claim.
Part checks out; one nuance or caveat was missed.
Number/year/author adjusted to the actual value.
Weak evidence: lost points or went in the sidebar.
Couldn't find the primary source.
The source doesn’t support it—removed from the body.
Why so many categories? Because “true/false” hides the most common case: the claim is almost right, but the number was wrong. See how a single citation moves along until it gets its tag:
The claim arrives
"Market X grew 40% in 2024, according to report Y." It goes to a verifier along with the number and named source.
Find the primary source
The agent goes to the original report — not a blog that summarized it — and checks the title, author, year, number, and method.
Returns the verdict
Finds 34%, not 40%, and that the data comes from a single commissioned survey. Verdict: CORRECTED + weak evidence signal.
Turns into a tag + score
The reference gets the CORRECTED tag, the text now says 34%, and the finding's reliability rating drops.
[CONFIRMADO] Doe et al. (2023), Nature — RCT, n=820 [CORRIGIDO] Relatório Y (2024) — 34%, não 40% [REBAIXADO] Preprint Z (2024) — não revisado por pares [FALSO] "Estudo" sem fonte primária localizável
📊 The reliability hierarchy in practice
In Module 1.3, you saw the principle: reliability = evidence quality, not subjective confidence. Now it becomes practice. The score from 1 to 10 for each finding doesn't come from how “right” the claim seems—it comes from where the evidence falls on the ladder below. The higher the step, the higher the score.
The ladder rises from the weakest evidence (preprint, at the bottom) to the strongest (peer-reviewed causal study, highlighted at the top). The rung’s position determines the score range: a number doesn’t become a “10” because it’s convincing—it becomes a 10 because it comes from the highest rung.
🟡 New here? — “peer-reviewed causal”
Peer reviewed = other experts evaluated the study before publication. Causal = the diagram shows that A cause B (e.g., a randomized controlled trial), not just that they occur together. Both things combined are the highest level. One analogy historical, however elegant, ranks lower: it suggests, but does not prove.
Reliability ≠ subjective confidence
A claim that "sounds obvious" may be on the lower rung (just an analogy), while a counterintuitive one may be at the top (a robust RCT). The score follows the source, not your gut feeling—so the most comfortable finding sometimes gets the lowest score in the report.
📄 Preprint × published / single study
Two cases require a firm hand in Phase 4: the preprint (study not yet peer-reviewed) and the single commissioned survey (a number from a single survey, often paid for by someone who benefits from it). Both go into the report, but with the appropriate weight.
📄 Preprint
Result shared before review. It may be completely right—or it may not hold up under review. The rule is downgrade: the claim stays, marked as "preprint, not peer-reviewed," and the note cannot appear at the top.
Typical status: DOWNGRADED + contested signal.
📐 Commissioned single study
A number from a single source, sometimes commissioned by an interested party. Don’t turn it into "the market grew X"— honestly reassign: "according to research by Y, commissioned by Z".
Typical status: PARTIALLY CONFIRMED / DOWNGRADED.
✓ Correct handling
- ✓State the status: "preprint," "commissioned research"
- ✓Attribute the number to the actual source, not to the "consensus"
- ✓Lowers the finding’s reliability score
✗ Common mistake
- ✗Treat a preprint as an established study
- ✗Present a commissioned number as a neutral fact
- ✗Give it a high rating because “everyone cites it”
⚖️ The contested-signal sidebar
Not all weak evidence should disappear — it just can’t be mixed in with strong findings. That’s what the disputed signal: a sidebar for disputed claims and preprints, always accompanied by the strongest opposing source. It's the honest way to say "this is live" without pretending it's resolved.
Claim: "Technology X reduces costs by 30%." — source: 2024 preprint, not yet peer-reviewed.
Strongest opposing source: an official industry analysis (2023) reports a gain of ~8% in real-world settings. Until the contradiction is resolved, the figure stays here — outside the ranked findings.
💡 Why the opposing source comes along
Showing only the disputed claim still gives it a platform. Pairing it with the strongest evidence from the other side turns the sidebar into a fair mini-debate—and makes it clear to the reader that there’s no winner yet. It’s Phase 2’s “empirical question that would resolve it,” made visible.
🧐 Read critically even when verified
Phase 4 greatly reduces the risk—but it doesn’t relieve you of the need to think. Remember the method’s central safeguard: the panel is built by the author. The five lenses came from the same model, so convergence among them is a strong hypothesis, not independent consensus across the field. For what really matters in your decision, go to the primary source yourself.
✓ Critical reading
- ✓You check the 2–3 numbers that could change your decision
- ✓Reads the banner and tags before citing the report
- ✓Treats convergence as a hypothesis, not proof
✗ Blind trust
- ✗Copies numbers without checking the status tag
- ✗Reads “5 lenses agree” as the truth of the field
- ✗Ignores the contested-signal bar
Self-recovery (optional): which source supports the HIGHEST reliability score?
📌 Module summary
Next module:
3.3 — Customize and Extend