Methodology

What it is

A daily, open-source early-warning brief on foreign developments with downstream consequences for border security, customs and trade, migration, and supply chains. It is situational-awareness analysis — not tasking, targeting, or official guidance.

Sources scanned

Each run sweeps a spread of open feeds, then corroborates on primary and wire sources:

• Broad news — Google News RSS and the GDELT project (no paywalled or proprietary feeds).
• Authority sources — official and multilateral feeds pulled directly: Europol (organized crime & seizures), the WHO, ECDC, and CDC travel notices (disease), WOAH (animal & agricultural disease), and ReliefWeb disasters — so the most consequential items come from primary sources, not just news aggregators.
• Humanitarian / displacement — ReliefWeb (UN OCHA).
• Conflict events — ACLED (armed-conflict and political-violence data), when credentialed.
• Targeted confirmation — direct web search and page fetches against primary and official sources (government, UN/IGO, Europol, INTERPOL) and the major wires (Reuters, AP, AFP) to verify facts and dates and to find a second independent outlet.

Wikipedia and aggregator redirect links are never cited; every source hyperlink points to the original publisher.

The set of automated feeds is not fixed: sources are added as they become available and dropped as they cease to be, and this page is updated to reflect those changes as they occur. Which feeds actually ran on any given day — and their rolling 30-day uptime — is tracked in the open on the Source Status dashboard.

What gets in

The window is the last 24 hours, anchored to US Eastern time; older items are out of scope. An item is kept only if it has a plausible border-security, trade, migration, or supply-chain nexus — foreign developments and their second- and third-order effects. Purely domestic policy is out of scope. The aim is roughly two to four items per mission area; noise is dropped, and a region with nothing over threshold is marked as such rather than padded.

Coverage begins with the 1 July 2026 edition — the brief's first. The archive, the Catalogue, the dossiers, and every ledger start there; events before that date are not catalogued.

Sourcing tiers

Every item carries one tier, shown on its source line and filterable on the Catalogue:

• Corroborated — two or more independent outlets, or a primary/official source.
• Single-source — one outlet only (flagged as such in the text).
• Developing — facts still actively changing; verify and revisit.

No more than a couple of items rest on any single outlet; where several carry a story, the most authoritative (a wire or official source) is cited alongside a second independent one.

The flash wire (Statim)

Between the daily editions, genuinely urgent breaking developments are flagged on Statim, the flash wire, within hours — a short What / So-what with a source link. Flashes trade some corroboration for speed: most are tagged Developing or Single-source and carry the same honest tier as everything else, so the standard is visible, not hidden. The daily brief remains the corroborated record — a flash is provisional, and the next morning's edition carries the confirmed account, folding the flash in and leading with whatever has since changed. The bar is high (urgent, clear frontier nexus, time-sensitive); most days carry no flash at all.

Significant seizures & the ledger

Alongside the analytic items, each edition flags significant foreign seizures and arrests — major interdictions announced by foreign authorities that wouldn't otherwise reach the reader. Each is recorded in a structured ledger, published continuously as The Vault (the seizures) and summarized monthly by the Confiscata. The ledger counts interdictions; it does not sum value, and never totals incommensurable units. The one deliberate exception is a rough drug tonnage: drug weights are normalized to kilograms from reported weights or pill-count conversions only — never inferred from a monetary value, with value-only and precursor-bundled seizures excluded — so any total is an honest order-of-magnitude read, not a precise figure.

Forecasting & accountability

Each edition closes with Watch Ahead — directional, resolvable calls. Every call fixes, at publication, an expectation, a resolution window, the counter-signal that would overturn it, and a confidence level in the brief's fixed confidence language (defined below: almost certain, very likely, likely, roughly even, unlikely…). Those words are never revised. Calls go to an append-only ledger; each week The Reckoning grades the ones that have come due — hit, miss, or non liquet (set aside when the evidence can't fairly decide) — and the standing record lives on the Tabularium, where nothing is removed or regraded. Confidence is shown so the record demonstrates skill rather than a wall of safe bets. Beyond the win-loss record, the Tabularium publishes calibration — for each confidence word, how often those calls actually came true versus what the word claims, flagged where they run over- or under-confident — with a Brier score once enough calls resolve. And the accountability closes a loop: each Reckoning distills the misses into lessons and recomputes that calibration, and the next day's Watch Ahead calls are written against them — so the track record is used to sharpen the forecasting, not merely displayed.

Confidence language

Every forecast uses a fixed estimative vocabulary, defined here and set at publication so calls can be graded consistently. The words map to probability bands; they are conventions of this brief, chosen for clarity and calibration — not a citation of any external standard.

Terms are used consistently across editions so that recorded confidence can be scored against outcomes over time. Where a judgment cannot be resolved cleanly for or against, it is marked non liquet (“not proven”) and excluded from scoring rather than counted as a hit or a miss.

Against the market

Where a call genuinely lines up with a liquid prediction market (e.g. Polymarket), the market's implied probability at the moment of the call is recorded next to it, the live odds are tracked as the call ages (shown on the Tabularium as “at call → now”), and the record is also scored against the market's — a check on whether the desk beats the crowd, not just its own past words. It is a secondary lens: most calls have no aligned market, one is never forced onto a call that doesn't fit, and a market only ever informs a probability, never a fact.

Standing desks

Alongside the daily editions, every subject the brief covers gets a standing desk that accumulates over time: one per country, region, and mission area, and — once they cross a coverage threshold — per named organization. Entities are resolved to a canonical name so one actor maps to one page. The heart of each region and mission-area desk is its ‘state of play’ — a short, human-readable synthesis of where that desk stands, drawn from its last seven days and refreshed daily by the brief run — sitting atop the full archive of stories that have touched it. Named individuals appear only as public figures acting in a public role (officials, leaders, major candidates, IGO figures — and actors a reputable authority such as law enforcement or a court has named in that role); never victims, private individuals, or unaffiliated suspects.

How it's produced

Editions are generated by an automated pipeline — an AI analyst working from the sources above — and are sent automatically, without a second analyst reviewing each edition before it goes out. That is a deliberate trade-off for a daily, one-person product. The compensating controls are a pre-send verification pass — a mechanical gate that blocks broken, placeholder, or aggregator/redirect links, plus a re-read of each item against its cited sources — and transparency: open sources only, every claim linked, sourcing tiers on every item, forecasts graded and calibrated in public, and a standing corrections record that the model refreshes with a weekly self-check — re-verifying the week's biggest claims against their sources — with recurring mistakes fed back into the model. That verification is the pipeline checking its own work, not a second analyst. All content is drawn exclusively from publicly available information and contains nothing classified, law-enforcement-sensitive, or otherwise protected.

Known limitations

• Language skew. The source universe is predominantly English-language; developments reported only in other languages are under-represented. Treat coverage of non-English media environments as partial.
• Selective, not exhaustive. A handful of items a day is heavy filtering of a global feed. Much is considered and dropped; absence from the brief is not evidence a development didn't occur or didn't matter.
• No independent review. Each edition runs an automated pre-send check — a mechanical link gate and a re-read of cited sources — but that is the pipeline checking its own work, not a second, human analyst. Whatever slips past is corrected after the fact and logged.
• Single maintainer. A personal project run by one person; there is no institutional continuity guarantee.
• Unofficial. The So what? and Watch Ahead lines are AI-generated analytic judgments, curated by the project's maintainer — a starting point to verify, not official positions and not an authoritative source.

Corrections

Errors are fixed when found and recorded — with the edition and what changed — in the corrections log. Spot something wrong? Send it here.