🔍 Truth and verification
The 10 commands in family 01 — from /truth a /freshness. Together, they form the complete antidote to hallucination and false certainty.
💎 /truth — separate fact from inference
The protocol for /truth: verifies assumptions, avoids stating unsupported claims as facts, and reports confidence.
It’s the language’s most-used command because it solves the most common problem—fluent answers mixing truths with assumptions.
New here? One inference is a conclusion drawn indirectly from data ("if sales went up and the price didn’t change, the customer base grew"). A fact is something that has been directly verified. Inferences are valid — as long as labeled as inference.
📦 The default /truth output
- •verified_facts — what the evidence supports.
- •inferences — conclusions drawn but not verified.
- •unknowns — what no one knows yet (and what matters).
- •confidence — stated level of certainty.
🎯 Copy and run
/truth sobre
How to verify: four separate blocks (facts, inferences, unknowns, confidence) and no inference dressed up as fact.
✅ /verify — confirm the specific claim
While /truth audits a set, /verify target a claim:
looks for enough evidence to confirm, correct, or reject. The result is always one of these three verbs—never "it seems so."
✓ Correct usage
- ✓Claim specific and testable (“competitor X launched a plan in March”).
- ✓Accept all three outcomes: confirmed, corrected, rejected.
✗ Incorrect usage
- ✗Ask for verification of something vague ("check whether our marketing is working").
- ✗Rejecting the answer for being inconclusive—inconclusive is data, not failure.
💡 Practical tip
Chain them when the stakes are high: /verify -> /confidence. A low-confidence confirmation calls for a second source before the decision.
🧱 /evidence — require evidentiary support
/evidence shows what evidence supports each important conclusion
and where data is still missing. This is the command that turns a well-written opinion into an auditable argument.
✓ Evidence-backed argument
- ✓Each conclusion lists its evidence (data, documents, measurements).
- ✓Declared gaps: "no retention data beyond 60 days."
✗ Argument without evidence
- ✗Safe conclusions with no evidence cited.
- ✗"Studies show…" without saying which ones, from when, or involving whom.
🎯 Copy and run
/evidence para cada conclusão em:
How to verify: conclusion → evidence → gap map. Conclusions without evidence must be explicitly identified as weak points.
📚 /sources — prioritize sources by authority
The protocol organizes sources according to the official criteria: authority, recency, proximity to the facts, and independence.
📊 The four prioritization criteria
Who has recognized expertise on the subject?
Is the information valid at the time of the decision?
Is it a primary source or seventh-hand?
Repeated citations count as a single source.
🎯 Copy and run
/sources para o tema; ranqueie por autoridade, atualidade e independência
How to verify: ordered list with source-based justification and a warning about interdependent sources.
🔎 /factcheck — audit claims one by one
Breaks the content into verifiable claims and classifies each into five categories. New here? One claim is a claim that can be true or false — "the product costs R$ 99" is a claim; "the product is expensive" is an opinion.
The /factcheck pipeline: raw text goes in, verdicts for each claim come out—with opinions kept separate from facts.
🎯 Copy and run
/factcheck in the article:— classify each claim and highlight the false ones
How to verify: table: claim → verdict → evidence used. Zero unclassified claims.
🎚️ /confidence — calibrate your confidence
Assigns a confidence level to conclusions and explains which factors would most increase or reduce that confidence. Calibrated confidence is different from a confident tone: it's a number accompanied by the conditions that would change it.
⚠️ Watch out for the classic mistake
Long, structured, and safe response isn’t right answer. Without explicit calibration, fluency becomes false credibility — exactly what this family exists to prevent.
🎯 Copy and run
/confidence nas conclusões de; diga o que elevaria e o que reduziria cada grau
How to verify: each conclusion with a confidence level + review triggers ("with quarterly churn data, I'd raise it to high").
🌫️ /uncertainty + /assumptions — surface what you don't know
/uncertainty surfaces missing data, unknown variables, and points with more than one plausible interpretation.
/assumptions lists the reasoning assumptions marked as verified, reasonable, or weak.
✓ Sound assumptions hygiene
- ✓Each assumption has a declared status before the analysis runs.
- ✓Weak assumption → low-cost test defined to validate it.
✗ Hidden assumption
- ✗Entire analysis supported by "the market keeps growing" — without labeling or testing it.
- ✗Uncertainty treated as something shameful to hide, not as information.
💡 Practical tip
Run /assumptions BEFORE expensive analyses (/deepresearch, /model). Finding a fragile assumption early saves the entire analysis.
🧬 /provenance + /freshness — origin and validity
The last two commands in the family complete the cycle: where it came from each critical piece of information (primary, secondary, inferred, or user-provided) and whether it still holds for today.
/factcheck — classify the claims
Breaks the text into claims and gives each a verdict. The foundation of everything.
/confidence — weigh the result
Confidence for each conclusion, with factors that would change it. Avoids performative certainty.
/provenance + /freshness — seal the origin and timeframe
Source of each critical data point + warning about what may have changed. Only then is the information ready for a decision.
🎯 Complete audit in one line
/chain(/factcheck -> /confidence -> /provenance -> /freshness) no relatório anexado
How to verify: final report with a verdict for each claim, confidence, source, and temporal validity — ready for a decision or publication.
💡 Practical tip
Using a 3-year-old statistic as current is the most common pitch deck slip-up. /freshness before the meeting, it takes 1 minute.
Quick self-check (optional): what sequence completes a truly comprehensive audit?
🎓 Module Summary
Next Module:
2.1 - Research and discovery — investigate in a structured way with /research, /scan, /discover, and related commands.