Most newsletters tell you what happened. We also grade whether we were right.

Every Daily Intelligence Report ends with a Watch List — specific situations we say to monitor next, each now with a confidence we commit at publish. Every call is scored against the news that actually followed, re-graded three days later, and shown here — hits and misses alike. This is the thing a "just the facts, no opinion" newsletter can't show you: a falsifiable forecast record.

80.1%
of our HIGH-confidence Watch List calls earned credit vs 64.6% for the calls we flagged MODERATE — confidence we committed at publish, before the outcome, that actually separates. (166 resolved)

Calibration scorecard

67.5%
Credit rate · 90d
68.4%
Last 7 days
596
Decided calls
720
Forecasts on record
Calibration by publication-stated confidence — committed at publish, before the outcome (distinct from the grader's after-the-fact confidence)
HIGH confidence (166 resolved)80.1%
MODERATE confidence (315 resolved)64.6%
LOW confidence (115 resolved)57.4%
Daily credit rate, last 21 scored days
Window tally: 263 hit · 279 partial · 30 miss · 24 counter. Credit = HIT + ½·PARTIAL of decided calls (HIT + PARTIAL + COUNTER + MISS). PENDING and NULL-COVERAGE are excluded. Window: 2026-07-25 → 2026-10-23.

How the grading works — and why you can trust the number.

Each Watch List call gets one of six verdicts: HIT, PARTIAL, MISS, COUNTER (evidence went the other way), PENDING (not yet resolved), or NULL-COVERAGE (the news corpus didn't cover the topic — un-scoreable, so it's excluded, never counted as a win or a loss). The headline number is a credit rate = (HIT + ½·PARTIAL) ÷ decided calls. We grade against the live news corpus, re-grade at +72 hours, and — the part most "we're accurate" claims skip — we commit each call's confidence at publish, before the outcome, so the per-confidence breakdown is a real calibration, not hindsight.

Recent calls — including the ones we got wrong

Graded from the 2026-10-23 Intelligence Report · re-scored through 2026-10-26.

HIT

Brazil Presidential Runoff (October 25)

(stated LOW at publish)

Lula won the runoff with ~51.8% (4.4M-vote margin) in an upset against polls; Flávio Bolsonaro alleged irregularities over the extended voting hours and said he'd seek annulment, with chaos at some polling places. Market turmoil followed, matching the watch triggers of fraud claims and contested aftermath.

AP News — LULA VICTORY IN BRAZIL! / Brazil election: Bolsonaro seeks to annul contested voting
HIT

PHP Critical Vulnerability Global Patch Race

(stated HIGH at publish)

Successful in-the-wild exploitation of the PHP2Shell flaws (CVE-2026-109618/109619) is confirmed, with working shells launched against honeypots and a suspected North Korean scanning campaign. Industrial vendors (Siemens, Schneider, ABB) confirmed affected products. No CISA directive or named victim breach is visible, but the exploitation trigger fired.

BleepingComputer — PHP2Shell security flaws now sucessfully exploited in the wild
PARTIAL

U.S.-Iran Military Escalation & Oil Infrastructure Strikes

(stated MODERATE at publish)

Heavy escalation signals (CNN-reported US readiness to strike within 96 hours, Houthi attacks on Saudi targets, Mecca-pact diplomacy in Tehran) but no confirmed Iranian strike on US assets, Gulf oil infrastructure or US-linked data centers, and no Brent $120 breach appears in the corpus. A tanker spill off the UAE hints at Hormuz incidents, but attribution is unclear.

Iran International — US forces ready for major Iran strikes within 96 hours of Trump order - CNN
PARTIAL

U.S. Midterm Election Integrity Breakdown

(stated HIGH at publish)

The DOJ filed additional election-related charges (48 noncitizens, 28 in California) and sued over Alaska's Senate certification, so enforcement and litigation did expand. However, no California mail-in ballot ruling, documented voting-machine failure, or new Russian operation was reported.

Washington Examiner — DOJ charges 48 noncitizens with federal voting crimes, including 28 in California

See every graded call in the Verdict Ledger →

Get the brief that grades itself →