METHODOLOGY VERSION 1.0 · EFFECTIVE JUL 2026 · CHANGES WILL BE LOGGED
How we work
Everything LoopIntel publishes is derived from sources you can check yourself. This page is the contract: how we source, how we express certainty, how we judge claims, and what happens when we get something wrong.
EVERY CLAIM CITED Each factual claim carries a numbered citation resolving to a source ledger entry. No claim rests on evidence we cannot show you.
ARCHIVE-FIRST Every ledger entry links the live source and an archived snapshot. If a source changes or disappears, the record we cited remains checkable.
NO ANONYMOUS SOURCING We do not publish claims from unnamed sources. If it cannot be attributed, it is not evidence here — it is at most a lead.
CORRECTIONS IN THE OPEN Errors are corrected in place, logged publicly with severity, and never silently edited. The corrections count is on every page footer.
01 — SOURCE TYPES & HANDLING
Every ledger entry is typed with a controlled vocabulary. The type governs how much weight a source can carry:
PRIMARY DOC Original documents, imagery, telemetry, filings. The strongest evidence class; published at full resolution wherever rights allow
OFFICIAL Government and military statements. Attributed positions — treated as claims by an interested party, not ground truth
REPORTING Journalism from outlets with named authors and correction policies. Weight follows the outlet's record on the specific beat
FACT-CHECK Verification organizations. Cited for their findings and, where warranted, examined as claims themselves — as in Analysis No. 001
ANALYSIS Think-tank and academic work. Cited for data and assessments with the institution named
INTERVIEW First-person accounts, on the record. Named speaker, named venue, checkable transcript or recording
STATE MEDIA State-controlled outlets. Never sole-source for a factual claim; cited to document what a state said, or when independently corroborated
02 — CONFIDENCE LANGUAGE
Certainty is expressed in a fixed vocabulary. If a term is not on this list, we do not use it:
DOCUMENTED Stated in a primary document or directly observable in primary evidence. Not an inference
HIGH CONF. Multiple independent sources corroborate; alternative explanations examined and found weaker
MODERATE Credible but singly sourced, or dependent on an inference we flag explicitly
UNVERIFIED Reported and relevant, not yet corroborated. Published only with this label, in the signals feed
UNRESOLVED The available evidence does not answer the question. We say so rather than choose a side
Signals carry the short forms [VER] verified, [DEV] developing, [UNV] unverified. A developing signal is re-graded or retracted within 72 hours.
03 — VERDICT VOCABULARY
When we examine public claims against primary evidence — from any side — verdicts use these terms. Verdicts describe the claim's relationship to the evidence, never the claimant's motive:
FALSE Contradicted by the evidence on a checkable point of fact
CONTRADICTED A characterization directly at odds with the source's text or content
EXAGGERATED Contains a true core inflated beyond what the evidence supports
MISTRANSLATED Rests on a rendering of the source language that the text does not support
MISLEADING Technically defensible but framed so a reasonable reader draws a wrong conclusion
UNPROVEN Consistent with the evidence but not established by it
UNCORRECTED Shown wrong by later evidence or the claimant's own source, with no correction issued
04 — CORRECTIONS & DISPUTES
Anyone can dispute any claim we publish — the DISPUTE link is on every analysis. A dispute is logged publicly on receipt, adjudicated against sources within 14 days, and closed with a published outcome: corrected, clarified, or upheld, with reasoning. Corrections are graded minor (detail, no conclusion affected), material (a conclusion required re-support), or retraction. The item's original text remains visible, struck through, beside its correction.
THE STANDARD WE HOLD OTHERS TO Analysis No. 001 documents a fact-check that was never updated after its own source issued a correction. This page exists so that standard can be applied to us, symmetrically and in public.
05 — WHAT WE DO NOT DO
— No paywall, no ads, no sponsored content. Reader-funded; the budget is published annually— No anonymous sourcing, ever— No claims from evidence we cannot publish or link— No motive attribution beyond what the record establishes— No unlabeled machine-generated text or imagery— No silent edits — every post-publication change is logged