The briefing unfolds in sections (cyan)—summary, ranked findings, hidden connection, claim safety, and verification—and they all support a decision you can trust. It’s the mental map for Trail 3.
Course path map
Detailed content
📋 HTML report anatomy
From the 60-second summary to verified references — the briefing sections and what each one delivers.
The first section of the briefing. It starts with the established fact and only then presents the contested interpretation—with nuance, not a headline.
This is what the reader sees first; the decision needs to fit in 60 seconds.
Established fact · contested interpretation · decision level.
The most important findings, ranked by reliability; each with a score from 1 to 10 and “Supported by” / “Challenged by” chips from the lenses.
The order and the note tell you what to trust first.
Ranking · score 1–10 · supports/challenges.
The non-obvious link that appears only when you cross-check all five lenses—it comes directly from the contradiction map.
This is the insight that no single lens can see.
Cross-check · non-obvious link · synthesis.
What no lens addressed, framed as the lens that could change the conclusions.
Shows the panel’s limits and informs the boundary question.
Blind spot · sixth lens · limit.
From 3 to 6 specific actions for the reader’s role defined in Phase 0—concrete, not abstract.
This is where the research turns into action for you.
Actionable · target reader · specific.
The guide to what to claim, qualify, and avoid; the boundary question that would change everything; and the references, each with a status tag.
Wraps up the report by saying what it’s safe to claim and where the limit is.
Claim safety · boundary question · reference with status.
🔎 The verification system
How STORM proves the numbers are real — banner, status tags for each citation, and the reliability hierarchy.
The badge at the top of the report: “X fabricated, Y corrected, Z downgraded” and the score “N/N checked.” It has to be true.
This is the one-line proof that Phase 4 actually ran.
Banner · N/N score · honesty.
Each citation gets a tag: CONFIRMED, PARTIALLY CONFIRMED, CORRECTED, DOWNGRADED, UNVERIFIED, or FALSE.
You see each source’s verdict without redoing the check.
Verdict · status · traceability.
How the 1-to-10 score comes from source quality: peer-reviewed causal > official data > single study > analogy > preprint.
Recaps 1.3 with examples and explains why one finding gets a 9 and another gets a 4.
Hierarchy · note · source quality.
Preprints and commissioned studies are downgraded; the statistic from a single study is honestly attributed to its source.
Avoids treating weak evidence as if it were strong.
Preprint · commissioned study · attribution.
The sidebar where disputed claims and preprints go, always with the strongest opposing source alongside.
Separate what’s solid from what’s still disputed.
Contested signal · opposing source · transparency.
Even when verified, the panel was written by the same author; check the numbers that actually matter to your decision.
Verification reduces error, not your responsibility.
Authorial convergence · critical reading · responsibility.
🧩 Customize and extend
The skill is yours: add lenses, tailor it to the reader, switch the model for a subagent, and carry the STORM principles forward.
You add a new lens (e.g., "THE AI BEGINNER" or "THE CONTENT CREATOR") following the template of the five—with the 3 required outputs.
Adapts the panel to your domain without breaking the method.
Lens patch · 3 required outputs · template.
Preload context about your business and your role into the skill so every report is tailored to you.
The actionable insight is much more useful when it’s aimed at the reader.
Tailored · reader’s role · context.
Where you can safely customize the HTML template while keeping the CSS and identity—without changing the entire visual style.
Customize the output without rewriting the template.
Template · CSS preserved · identity.
Run each lens on a different model based on cost and quality—the expensive one where it matters, the cheap one where it’s enough.
Balances research cost and depth.
Model by agent · cost/quality · trade-off.
The same skill fits in .codex/ or .agents/; only the folder changes; the method stays the same.
You’re not limited to a single agent.
Portable · .codex · .agents.
Apply the idea — multiple perspectives + verification — in other workflows, such as a council or an idea roast.
The value is in the method, not just this skill.
Multiple perspectives · verification · council/roast.
🧭 Best practices and pitfalls
The method’s guardrails—what never to skip, what never to inflate, and the checklist to use before trusting and sharing.
A report delivered without adversarial review and citation checks simply isn’t a STORM report.
This is the rule that protects confidence in the result.
Phase 4 · mandatory verification · accurate banner.
The limit is five lenses and one verifier per citation cluster; more agents don’t mean more truth.
Avoids cost and noise without improving quality.
Agent cap · cluster · no bloat.
Every lens and citation points to a real source that was consulted; anything that can’t be verified is downgraded or cut—never disguised.
This is what separates research from plausible invention.
Real source · downgrade/cut · no disguising it.
Always state in the report that the five lenses have the same author; agreement is a strong hypothesis, not a consensus in the field.
Intellectual honesty is what sustains the method.
Original panel · convergence · consensus.
A run creates about 9 to 11 agents (5 perspectives + 4 to 6 verifiers); predictable, but it needs to be managed.
You decide when and how many times to run it.
~9–11 agents · predictable · rate-limited.
The final check before using the briefing: Phase 4 is complete, the banner is accurate, the panel is disclosed as original work, and you've checked the numbers that matter.
Wraps up the course by turning the method into a habit.
Checklist · trust · share.