How we build every Echo

§How we work

Pattern recognition over opinion. Two sources over one. Verified before published, not after.

Every Echo is built the same way, and the way is written down here so that a reader can check our work rather than take it on trust. This page sets out the pipeline, the rules we use to choose historical cases, the library those cases sit in, and the publications we read.

It sits alongside the Editorial Code of Practice, which governs conduct. This page governs method.

§Section 1The TEE Pattern

Every body paragraph in every Echo follows the same structure: Topic, Evidence, Explanatory. A topic sentence states the claim. Evidence supports it with a citation. An explanatory line ties it back to the central pattern.

This is the discipline used in academic essay writing, and it gates editorial quality before publication. A paragraph that cannot carry all three parts is a paragraph we do not have the evidence to write.

§Section 2The Six-Step Pipeline

2.1 Source ingestion

30+ UK and international publications, listed in Section 5. Hansard from 1803, the UK National Archives, the ONS and the Bank of England, plus press archives back to 1900.

2.2 Pattern detection

Current events are classified into recurring patterns identified from the archive. Bills with similar dynamics. Polling collapses that preceded reshuffles. Policy announcements that fit a wind-down trajectory. The library is described in Section 4.

2.3 Draft generation

First draft written in TEE format across three blocks: the news today, the historical pattern with its cases, and the market hypothesis with base rates. Every paragraph carries an inline citation.

2.4 Dual verification

Every factual claim is cross-checked across a minimum of two independent sources. Different publications, not two copies of the same wire report. Statistics are confirmed from the authoritative source. Discrepancies are flagged for editorial review.

2.5 Anti-fake check

An automated classifier flags retracted claims, known hoaxes and machine-generated noise, cross-referenced against fact-checking and publication retraction records. Anything flagged is routed to the editor.

2.6 Editorial pass

A human editor reviews the draft for tone, factual nuance and Editorial Code compliance. No Echo publishes without explicit editor sign-off.

§Section 3How We Choose the Cases

The historical block of an Echo is the part a sceptical reader should interrogate first, because a pattern can be made to say almost anything if the cases are picked to suit it. These are the rules we apply, and they are the same for every Echo.

• The object is an action, not a topic. We name the action in the news in one line, for example a broadcaster merging commissioning departments, or a national operator suffering a third system failure. Cases are other instances of that action. A calendar event standing next to the news is background, never the object.
• One consequence, tested the same way in every case. Each case is marked held or not held against a single stated consequence, measured in a fixed window after the action. That consequence is what the first market asks.
• Cases are chosen by similarity, not by date. Each candidate is scored against the current event on the same set of features, everything below the threshold is dropped, and the highest scoring cases are taken. Date breaks ties, nothing more. Where recency matters, because a rule or a regulator changed, it lowers the score of older cases by itself.
• The make-up of the sample is printed next to the base: how many cases were taken, how many passed the threshold, and the range of similarity. Three events in twenty years and one a decade ago is a weak comparison, and the reader should be able to see that it is.
• Where the country is thin, we widen to comparable regimes, not to comparable headlines. If Britain has too few instances of an action, we look for the same action under similar institutional conditions, and we say in the Echo that we have done so.
• Fewer than four genuine cases and the Echo does not publish. We say the record is too thin and stop, rather than stretch the definition until the number is reached.
• Below six cases we publish the fraction only. Three of four is shown as three of four, five of five as five of five. A percentage implies a precision that five cases cannot carry. From six cases upward we show the percentage alongside the fraction.

§Section 4The Pattern Library

Every Echo we publish deposits its case bank into a shared library, organised by the action at its centre rather than by subject. A pattern in the library is a named action, the consequence we test it against, the cases found, and the record of how often that consequence followed.

The library is the reason the second Echo on a recurring action is faster and better evidenced than the first, and it is the part of Hunch that compounds. It is built from published work only. A pattern enters the library when an Echo carrying it has passed editorial sign-off, and it carries the same citations the Echo did.

Patterns are grouped across the six sections we publish in: politics, finance, tech, sport, culture and the UK regions. Each entry links back to the Echoes that built it, so the base rate can be traced to the cases and the cases to their sources.

§Section 5Sources

A fact exists for us only if it appears in a publication on this list. Nothing is written from memory, and nothing is written from a source that is not named here.

UK national press

• The Times. Westminster, government policy, Sunday Times investigations
• Financial Times. Fiscal, markets, business-government interaction
• The Guardian. Politics, public policy, investigative reporting
• The Telegraph. Conservative politics, foreign policy, defence
• The Observer. Sunday political analysis, long-form essays
• The Economist. Policy analysis, economic context, leaders

Political weeklies and magazines

• The Spectator. Conservative-leaning political commentary
• New Statesman. Left-of-centre political analysis
• Prospect. Long-form political and policy analysis
• Tortoise Media. Long-form journalism, slow news

Broadcast and wire

• BBC News. Political reporting, Newsnight, Today programme
• ITV News. Political reporting, regional coverage
• Sky News. Westminster coverage, breaking political news
• Reuters. Political wire reporting, fact verification
• Associated Press. International political coverage
• Bloomberg. Fiscal and monetary policy, market reaction to politics

Specialist coverage

• Politico EU. Westminster and Brussels intersection
• Press Gazette. Media industry coverage, journalist accountability
• Sifted. European tech and startup coverage
• UK in a Changing Europe. Post-Brexit UK-EU relations analysis
• Variety, Deadline, The Hollywood Reporter. Film and television, on demand by topic
• ESPN. Sport, on demand by topic

Government and primary sources

• Hansard. Verbatim record of parliamentary debates
• UK National Archives. Cabinet papers, official records
• Gov.uk. Department announcements, official statements
• Office for National Statistics. Official UK economic and social statistics
• Bank of England. Interest rates, inflation, monetary policy
• Office for Budget Responsibility. Independent fiscal forecasts and analysis

Think tanks and policy institutes

• Institute for Fiscal Studies. Tax, public spending, distributional analysis
• Resolution Foundation. Living standards, low-to-middle income policy
• Institute for Government. Whitehall, civil service, government effectiveness
• IPPR. Progressive policy research
• Centre for Policy Studies. Free-market policy research

Fact verification

• Full Fact. UK independent fact-checking
• BBC Reality Check. BBC fact-checking team
• FactCheck.org. Cross-border political claim verification

How sources are used

Every Echo references a minimum of two of these publications, with at least one primary source where one exists, such as Hansard, the ONS or a government press release. The full citation list appears at the foot of every Echo in Harvard reference style.

Where one source disagrees with another, we say so explicitly in the Echo body. Where a claim cannot be verified across two independent sources, the Echo is held until it can be.

If a source has retracted, corrected or updated a claim we cited, the Echo is amended within 24 hours and an update note appears on its revision log.

§Section 6History and Hunch

Every market shows two numbers. History is the record: we scan the archive for the same kind of action in the same kind of setting, take the cases that qualify, and count how often the consequence followed. It is a fact, shown with its sample, and it does not change with our view of today.

Hunch is our reading of this case: how the same scenario feels against the record, given where events are heading and the circumstances around them. It starts from History and moves only where today differs from the cases behind it. Hunch is scored separately, and its accuracy is published on the Accuracy page beside the pattern's.

§Section 7The Numbers

• 125 years of news archive coverage
• 30+ publications referenced across each Echo
• 2 minimum independent sources per factual claim
• 8 cases shown per pattern, with the full record published alongside
• 4 cases minimum, below which an Echo does not publish
• 24-hour target latency from event to Echo publication
• Several Echoes a day, more when the news demands it

§Section 8What We Will Not Do

• Publish opinion without evidence backing it
• Take partisan editorial positions on policy questions
• Use direct quotes from primary sources without analysing them in our own words
• Recommend market positions. We publish the pattern, you decide what to do with it
• Trade against our own published markets