How we grade, judge
and correct.
Intelligence is only as useful as the reader's ability to weigh it. Every HELIOS product uses the same published rules, so you can see how sure we are — and why.
Open sources, opened and checked.
Primary over aggregator
News feeds are leads, not sources. We open the original article or record and cite that — never a redirect or a summary page.
Two independent sources
A top story needs two sources on different domains. One source is not enough — such items are labelled as intelligence gaps.
Links that still work
Every cited link is fetched again at publication. Dead links are removed; a story that loses its second source is dropped.
Numbers from instruments
Prices and indicator readings come straight from their data sources with capture times — never retyped from an article.
Two letters on every item: who said it, and how solid it is.
The Admiralty (NATO) system grades the source and the information separately. B2 means a usually reliable source reporting something probably true.
| Grade | Source reliability | Meaning |
|---|---|---|
| A | Completely reliable | No doubt of authenticity, trustworthiness or competence; a history of complete reliability. |
| B | Usually reliable | Minor doubt; has provided valid information most of the time. |
| C | Fairly reliable | Some doubt; has provided valid information in the past. |
| D | Not usually reliable | Significant doubt, though valid information has come from it before. |
| E | Unreliable | Lacks authenticity, trustworthiness and competence; a history of invalid information. |
| F | Cannot be judged | No basis to evaluate the source. |
| Grade | Information credibility | Meaning |
|---|---|---|
| 1 | Confirmed | Confirmed by independent sources; logical and consistent with other information. |
| 2 | Probably true | Not confirmed, but logical and consistent with other information. |
| 3 | Possibly true | Not confirmed; reasonably logical; agrees with some other information. |
| 4 | Doubtful | Not confirmed; possible but not logical; nothing else on the subject. |
| 5 | Improbable | Not confirmed; illogical and contradicted by other information. |
| 6 | Cannot be judged | No basis to evaluate the information. |
Seven words, each with a number.
Judgments use the calibrated likelihood terms of U.S. Intelligence Community Directive 203, always with the range in brackets — so “likely” never quietly means 51% to one reader and 90% to another.
| Term we use | Synonym | Range |
|---|---|---|
| Almost no chance | Remote | 01–05% |
| Very unlikely | Highly improbable | 05–20% |
| Unlikely | Improbable | 20–45% |
| Roughly even chance | Roughly even odds | 45–55% |
| Likely | Probable | 55–80% |
| Very likely | Highly probable | 80–95% |
| Almost certain | Nearly certain | 95–99% |
Likelihood says what. Confidence says how sure.
High
Judgment rests on high-quality information from multiple independent sources, and the facts are not in dispute.
Moderate
Information is credibly sourced and plausible, but not corroborated enough for a higher level.
Low
Information is scant, fragmentary, single-source or questionable — the judgment could easily change.
If any gate fails, nothing is published.
Global Watch editions pass these automated gates, a live-link check and an independent fact-check before they reach the site. A failed edition is held back and the desk shows the next edition as pending — it never shows stale news as live.
| Gate | Name | What it enforces |
|---|---|---|
| W1 | Structure & timestamps | Every edition, indicator and quote carries a valid timestamp; the next edition time is declared. |
| W2 | Analytic standard | Each top story has fact, judgment, likelihood term, confidence and an Admiralty grade. |
| W3 | Sourcing | Two or more graded, linked sources; top stories need two independent domains. |
| W4 | Indicators | Pizza Index readings and market quotes carry capture times; numbers come from code, not prose. |
| W5 | Integrity of voice | Blocks out-of-scope brand terms and flags sensitive language for review. |
| W6 | Language | North America desk in English, Asia desk in Korean — no mixing. |
| W7 | Fresh & publishable | Nothing older than 18 hours is published as new. |
| W8 | Scorable forecasts (advisory) | A forecast names one event, a deadline and a criterion. Problems are logged as warnings; a malformed forecast is dropped, the story stays, and publication is never blocked. |
Live record of the last run
Fun to watch. Kept out of our judgments.
Two trackers, same moment
Pentagon Pizza Index (uspizzaindex.com) and PizzINT (pizzint.watch) are read at the same moment from their public pages.
Compare, then show
DOUGHCON levels and shared stores are compared and shown as agree, partial or disagree. A closed store shown as surging counts as a conflict. The index carries no Admiralty grade, because we do not treat it as a source.
Never in a judgment
The threat level, headlines, flash items and story judgments rest on security news and market data only. The Pizza Index appears only in its labelled side panel. Both trackers draw on Google Popular Times, so they are not fully independent of each other either.
We grade our own likelihood words.
Stated in advance
Each scorable judgment carries its event, deadline and criterion on the edition page from publication. The first probability is the one graded.
Graded blind, checked twice
A grader sees the event and criterion but not our probability; an independent second review must agree before a result is final.
Shown as is
Misses stay on the page, voids are listed with reasons, and rates appear only once enough calls are graded.
Accountable by default.
Corrections and withdrawals
- Material errors are corrected in the next edition and noted.
- Any edition can be withdrawn from the site; withdrawn editions are archived, not deleted.
- Every edition keeps a permanent URL, so what we said stays on the record.
How AI is used
- An AI analyst drafts Global Watch under these standards.
- A separate AI fact-checker re-opens the sources; unsupported or legally risky items are removed.
- All numbers come from code. HELIOS staff receive every run report and can withdraw an edition.
- Client deliverables (ATLAS, AEGIS) are reviewed by HELIOS analysts before release.
Is Global Watch written by people or by AI?
Why show the Pentagon Pizza Index at all?
What happens when you get something wrong?
Do you use classified or leaked material?
Hold us to these standards.
Ask for a sample ATLAS or AEGIS deliverable graded exactly this way.