Every convention Throughline uses, in one place — what the grades mean, how sources are weighed, what the Ledger is for, and what this brief promises and refuses to promise.
Throughline is a daily continuity brief on conflict, statecraft and political economy. Its organising claim is that most news products restart every morning, and that the more useful thing is to carry stories forward — to say where a situation now stands, what moved since the last edition, and how confident the brief is in its own account.
It is deliberately slower than an alerting feed on breaking events. If you need to know within minutes that something has happened, this is the wrong product. What it offers instead is a continuous, graded, and openly audited account of a small number of running stories.
A first-time reader can start at any edition. The Board at the top of each one summarises every thread; you do not need to have read the back numbers.
The Board opens every edition: five figure tiles and a table of every thread with its direction, when it last moved, its confidence, and a one-line statement of where it stands. A reader who reads only the Board should come away with an accurate, if coarse, picture.
The lede states the single most important thing the edition has to report — which is sometimes that the previous edition was wrong.
Threads are the body. A substantial thread is laid out in three parts:
Short threads skip this scaffolding and simply run as prose.
The margin block beside each thread carries its grades and its sourcing basis. On a narrow screen it appears below the thread instead.
Then: the Ledger, the Calendar of dated commitments, Off This Board, Known weaknesses in this edition, and Notes.
Direction glyphs describe consequence, not quantity. This is the convention most often misread, so it is stated explicitly: a red ▲ on a price tile means the disagreement between sources is widening, not that the price went up.
| ▲ | The situation is worsening, escalating, or becoming less certain. |
| ▼ | The situation is easing or resolving. |
| ▬ | Holding — no material movement since the last edition. |
How much the brief trusts its own account of the thread — a judgement about evidence, not a forecast. It is encoded in the number of filled dots rather than in colour, so it survives greyscale printing and colour-blind rendering.
| ●●● | High. Rests on primary documents or on multiple independent sources that agree. |
| ●●○ | Medium. Sources conflict on magnitude, or key claims rest on parties to the events. |
| ●○○ | Low. Load-bearing figures are single-sourced, disputed, or unreachable. |
Every thread graded below High carries a Why not higher line stating the specific reason. A confidence grade without a stated reason is not worth much, so the brief does not print one.
How much would follow if the thread's central claim is right — deliberately separate from confidence. A low-confidence, high-consequence thread is exactly the kind a reader should keep watching.
| ◆◆◆ | High — structural stakes: war, chokepoints, the rate path. |
| ◆◆ | Medium — significant but bounded. |
| ◆ | Low — worth tracking, limited spillover. |
Written as "Story day n · tracked since No. x", because those are two different facts and conflating them was a defect in earlier editions. Story day counts the age of the situation itself; tracked since says when this brief began covering it. A war on story day 1,619 that this brief has followed for three editions is honestly described only by giving both.
The source classes a thread's assessment rests on — named so a reader can judge the foundation without reading all the notes.
Every load-bearing fact has two attributes: who asserted it (the claimant) and which outlet reported it (the carrier). The brief names the claimant in the sentence, always. The carrier stays in the notes unless it is the sole outlet, is partisan, or is effectively the claimant itself — as with insider process reporting, where an outlet's anonymous sourcing is the claim.
Collapsing these is a real failure mode, not a stylistic one. An edition once attached an analytics firm's name to a baseline figure that firm had never published but that a newspaper had computed. That is a Ledger entry, and it is typed as Framing.
State media, party organs, and parties to the events they describe are identified inline, because the source's interest is part of the fact. A defence ministry's account of what it struck, a security service's account of its own operation, a company's disclosure of its own incident — all are reported, none is laundered into neutral voice.
This includes sanctioned state outlets. They are cited as claims of a party, never as corroboration, and a link is not an endorsement of the source.
If a claim rests on one outlet, that outlet is named in the sentence. The words "reportedly" and "one outlet" are not used as substitutes for naming.
When credible sources disagree, the brief publishes the disagreement and downgrades its confidence, rather than averaging them or silently picking a favourite. When two conflict trackers differ threefold on territorial magnitude but agree on direction, both figures appear and the thread is graded Medium. This rule costs tidiness and buys the reader an accurate picture of how well anything is actually known.
When a source could not be reached — a paywall, a refused request, a dead link — the brief says so in place rather than quietly substituting a weaker source. "Could not check" is reported as different from "checked and found nothing."
Sources have terms, and this brief reads them. The governing distinction is that facts are not copyrightable and expression is: a casualty count, a barrel price or a territorial figure can be taken from anyone's reporting and restated here in this brief's own words, while another writer's phrasing cannot.
Two consequences a reader may notice. First, the brief reports prices in prose and never as a table or an auto-updating element — a table of one benchmark priced by several vendors stops being reporting and starts being a data product, which is what market-data terms actually prohibit. Where feeds disagree, the brief says so and explains why they differ rather than reproducing their quotes side by side. Second, some analytical sources require exact attribution wording, which is why the notes carry strings like "Source: Institute for the Study of War" rather than a looser paraphrase.
One rule is worth stating openly because it shapes the writing. The Institute for the Study of War permits quotation but bars use of its material for purposes tied to markets or speculative markets. This brief therefore keeps conflict assessment and market commentary rhetorically separate: it reports what a conflict tracker assessed, and separately reports what a market did, and does not present the former as a driver of or basis for the latter. Readers should not read the adjacency of a war thread and an oil thread as an implied causal claim. It is not one.
The Ledger is the brief's record of where its own published account diverged from the documentary record. It is the reason this product exists in the form it does. Corrections are entered in a later edition and carried cumulatively; entries are appended, never rewritten.
Errors are typed, because different failures deserve different names:
Symmetry. Throughline is issued alongside SITREP, a separate OSINT event feed by the same author, and the Ledger grades both on identical terms. Because Throughline makes the longer-range interpretive claims, it should normally accrue the more corrections; if that ratio ever inverts, the ledger is being gamed. Entries are marked SITREP-eligible when a SITREP edition was available to grade, and excluded when one was reviewed and found to contain nothing gradeable. A rot signal fires if excluded entries accumulate while none are ever counted — the pattern that would indicate the symmetry claim has become decorative.
A caution about the count. The number of Ledger entries in an edition measures how hard that edition audited its predecessor at least as much as it measures how wrong the predecessor was. A cycle with sixteen entries is not necessarily a worse edition than one with two — it may be a better-audited one. Earlier editions should be assumed to carry undiscovered errors at a similar rate.
Off This Board names what was happening outside the tracked threads, so a reader can size the blind spot rather than being handed a disclaimer. Known weaknesses in this edition states what the brief itself thinks is shakiest about the edition you are reading — before a reader has to find it.
Throughline is written and edited by Jackson Kemper, who is accountable for what it publishes.
The method, stated plainly. Each edition is assembled with automated retrieval and AI assistance. Source material is gathered by automated web search and retrieval; per-thread research and drafting are AI-assisted; and the edition's structure, standards and editorial judgements are set and reviewed by the editor. Where a claim could not be verified against a primary source, the edition says so in place. The brief is disclosed this way on principle: a publication whose product is a correction ledger cannot be quiet about how it is made.
This method has consequences a reader should weigh. Automated retrieval is fast and broad but is refused by many primary sources — government release feeds, some agency pages, and paywalled outlets frequently block it — which is why "this source refused retrieval" appears in editions as often as it does. AI-assisted drafting is fluent, which is a hazard as well as a convenience: fluency can make a thinly-sourced claim read as well as a well-sourced one. The Ledger, the confidence grades and the Why not higher lines exist as counterweights to precisely that.
If something here is wrong, please say so: corrections@tryrerun.com.
Substantiated corrections are entered in the Ledger of a subsequent edition, typed, with the original claim quoted. Corrections are not made silently in place, and a correction is recorded whether it was found by a reader or by the brief itself.
Not investment advice. Prices quoted are indicative, frequently disputed between feeds, and reproduced to illustrate disagreement between sources rather than to establish a level. Nothing here is a recommendation to trade, and no figure should be relied on for that purpose.
Not a forecast. Confidence and consequence grades are this brief's judgements about its own evidence. They are not probabilities, and they say nothing about what will happen next.
Not comprehensive. Seven threads is a deliberate limit. Off This Board exists because the alternative — implying completeness — would be the more misleading choice.
Not a primary source. Everything here is assembled from public reporting and public records, each linked in the notes. Where the brief and its source disagree, the source is the record.
Two conventions are not yet stable, and it seems better to say so than to let a reader assume otherwise.
"Independent priors." Editions 3 and 4 carried a masthead count of independent priors — the number of genuinely independent source families the notes traced back to, as distinct from the raw citation count. Fifty-seven citations resting on nine independent origins is a weaker evidentiary base than the number 57 suggests, and the metric existed to say so. It was never defined for readers, and No. 5 grouped its notes topically instead, which makes the figure non-comparable; the count is therefore not currently published. The concept is sound and is intended to return with a stated definition.
The archive has a hole. Vol. I No. 2 is not preserved. Because the cumulative Ledger count rests substantially on entries recorded there, most of that count is currently reported rather than auditable. The Ledger page states exactly which entries it can and cannot produce.