Editorial standards

These are the rules our newsroom works to. They are the same documents our editors and our automated checks use, published here unedited, so readers can hold us to them.

In short: we cover a story only when it matters for markets; every number, name and quote must come from a cited source; we never give investment advice; and a named editor reviews every article before it is published.

1. What we choose to cover

MacroInsight covers a story only when it changes what a thoughtful investor should believe about growth, inflation, policy, liquidity, earnings or risk appetite, and when we can say something useful about it. Tier-one outlets (Bloomberg, WSJ, FT, Reuters-via-CNBC) tell us what the market is talking about; this standard decides what we write about.

The question every candidate must answer

"If a smart generalist investor missed this story today, would they misunderstand the market tomorrow?"

If the answer is no, we do not write it, no matter how loud it is.

Scoring rubric (each axis 0–5)

Attention is not news. Loud coverage of something that has been going on for weeks scores low on surprise. A clear gap between the headlines and the price (the news says oil should rise, and oil keeps falling) scores high on explainability when we can say why.

Weights: transmission 0.30, breadth 0.20, surprise 0.20, durability 0.15, explainability 0.15. The weighted editorial score is scaled to 0–100.

Market signal (computed, not judged)

• Desk consensus: number of distinct tier-one desks carrying the story (Bloomberg, WSJ, FT, CNBC, MarketWatch). Three or more desks within six hours is strong consensus.
• Primary source: an official release (central bank, statistics agency, treasury, regulator) is present in the cluster.
• Velocity: items in the last two hours.
• Recency: signal decays with a six-hour half-life from the latest item.

Composite = 0.65 × editorial + 0.35 × signal. Consensus can raise a story; it can never carry one that scores badly on transmission.

Decisions

Daily cap: at most 6 leads per day. Scarcity is part of the product.

Automatic disqualifiers (decision = skip, regardless of score)

• Single-company news without an index-level, sector-level or macro angle (routine earnings, product launches, executive moves at non-mega-caps, analyst rating changes).
• Opinion, columns, "how to invest" or personal finance pieces, listicles, sponsored content.
• Rumour from a single unnamed source, unless a tier-one desk reports it and markets react.
• Crypto token-specific stories (a regulatory or systemic crypto story is eligible).
• Celebrity, sport, lifestyle, crime or human interest without market transmission.
• Routine regulatory housekeeping (enforcement actions against individuals, bank merger approvals, speeches with no policy signal).
• Stories we covered in the past 48 hours, unless there is a material update: a new number, a decision, a reversal, or a market move that changes the implications.

Category taxonomy

monetary_policy, macro_data, fiscal_policy, trade_tariffs, geopolitics, energy_commodities, credit_financial_stability, markets, megacap_corporate, china_em, regulation, technology_ai.

Manual curation

When an editor submits a headline directly, selection is skipped, but verification is not. The headline must be corroborated by at least one primary source or two independent credible outlets before a draft can be marked ready.

2. How we write

MacroInsight articles explain what is happening in the world's markets and why it matters to the person reading. The reader is curious and smart, but not a market professional: they want to understand the state of the world and how it touches their money, their job and the prices they pay. Write for them, not for an analyst desk. They want the explanation, not the adjectives and not a wall of numbers.

Each article is our own analysis. We check the facts against the primary source (the central bank statement, the statistics release, the company filing) or independent credible outlets, then explain them in our own words. The sources are listed with every article.

Think it through first

We think like a data-led analyst and write like a journalist. Before explaining anything, work out three things from the facts:

1. What is true, and what is new. Headlines measure attention, not change. A story can be loud because something new happened, or because something that has gone on for weeks has only now drawn coverage. Say which. If attacks on shipping in the Strait of Hormuz never stopped and simply return to the headlines, the news is the coverage, not the attacks, and an article that treats them as a fresh shock misleads the reader. If the facts do not show whether it is new, do not assume either way.
2. What the market is doing about it. Check the price, and its trend over recent weeks, against the story. When the headlines point one way and the market goes the other, that gap is the story. Explain it: the risk was already known and priced, the flow of goods or money was never disrupted, or a bigger force (demand, supply, policy) outweighs it.
3. Why the market might respond, and how. Markets move on what changes expectations, not on news as such. Reason from the data to the mechanism, say what would make the market react, and be plain about what is known and what is assumed.

The tone is calm and pragmatic: no alarm the data does not support, and no dismissing a risk the data does show.

Formats

Every article also carries structured fields: headline, dek, three takeaways, implications, exposures, what-to-watch items, and an image brief. The structured fields are part of the article, and the same rules apply to them.

Headline

• Declarative, specific, sentence case, ≤ 90 characters.
• Carries the "so what": the event and its consequence where it fits. Good: "Fed holds at 4.25% but signals two cuts, pulling Treasury yields lower". Bad: "Fed makes big decision: here's what it means".
• No questions, no exclamation marks, no colons used as teasers, no "here's why", no "what you need to know", no "everything", no puns.
• Numbers only if they appear in the fact sheet.

Dek

One sentence, ≤ 200 characters, that states the market consequence or the key tension. It must not repeat the headline.

Takeaways

Exactly three bullets, each ≤ 28 words. Takeaway 1 is the fact, takeaway 2 is the mechanism, and takeaway 3 is the implication or the thing to watch.

Body

1. What happened: facts only, in inverted-pyramid order. Who, what, when, how much, and whether it is new or more of something already under way. Every paragraph cites fact IDs.
2. Why it matters: how the news travels from the event to the reader. Name the thing that moves (interest rates, borrowing costs, the dollar, company profits, petrol prices) and walk through the chain step by step in plain English. One idea per paragraph. If the market is not moving the way the headlines suggest, say so here and explain why.
3. Market implications (analysis only): prose version of the implications list. Asset by asset, conditional where appropriate ("if X holds, Y").
4. The other side (analysis only): the strongest counter-argument, or what is already priced. This section is mandatory, and it is what makes us credible.
5. What to watch: the next dated catalysts, and the level or number that would change the view.

Section headings are plain statements or questions a reader would ask ("Why oil matters for mortgage rates"), not clever labels.

How an explaining paragraph works

In "Why it matters" and "Market implications", each paragraph follows the same order:

1. The point, in one plain sentence.
2. Why it works that way: the cause and effect, as you would explain it to a smart friend.
3. The evidence: at most two or three numbers, each one there because it supports the point.
4. What it means for someone who holds the asset, borrows, saves or spends.

Not this: "For Treasuries, the test is supply: $39 billion of 10-year notes on October 7 and $22 billion of 30-year bonds on October 8. Weak demand raises the risk of a retest of the 30-year's 5.703% peak."

This: "The US government needs to borrow a lot this week, and how easily it does so will set the tone for long-term interest rates. When buyers turn up in force for its $39 billion sale of 10-year notes and $22 billion of 30-year bonds, yields can settle. When they don't, the government has to offer a higher yield to sell the debt, and that pushes up long-term borrowing costs everywhere. A weak sale could send the 30-year yield back toward its recent peak of 5.703%, and because bond prices fall when yields rise, long-dated bond funds would feel it most."

Craft

The habits of the best financial journalism, which every article follows:

• Say why it matters early. By the second paragraph the reader knows what changed and why they should care. Everything after that backs the point up.
• Lead with people and consequences, not instruments. "Mortgage rates are likely to stay high" before "the 10-year yield rose".
• Give every number a job. Say whether it is big or small, higher or lower than before, and compared with what, where the facts allow it. A figure with no comparison means little.
• Translate market moves. A rise in yields means higher borrowing costs; a stronger dollar means cheaper imports and a squeeze on exporters. Say the everyday version.
• Step back once. One paragraph places the news in the bigger trend: is this new, or more of the same?
• Concrete over abstract. Name the central bank, the company, the country, the product.
• Verbs carry the move. "Oil fell 4%", not "oil saw a significant decline". Markets do not cheer, panic or shrug; people buy and sell.
• Every paragraph moves the story on. No paragraph restates the one before it. End on what comes next, not a summary.

Implications (structured)

Two to five items. Each item needs:

• instrument: be specific ("US 2-year Treasury yield", not "bonds").
• direction: up, down, wider, narrower, higher_volatility, lower_volatility or mixed.
• mechanism: one sentence of cause and effect.
• horizon: days, weeks or months.
• confidence: low, medium or high. At most one item may be high. High means the mechanism is direct and the fact sheet shows the market already moving that way.
• invalidation: what would make this wrong.

Implications describe likely pressure, never recommendations.

Exposures (structured)

Exposures are the join keys for the premium tier's portfolio mapping. Use only standard identifiers:

• asset_class: equities, government_bonds, credit, fx, commodities, real_estate, crypto
• region: ISO 3166 alpha-2 country codes, or EU / EM / GLOBAL
• currency: ISO 4217 codes
• sector: GICS sector names
• commodity: crude_oil, natural_gas, gold, copper, agriculture, and similar
• factor: duration, credit_spread, usd, value, growth, small_cap, volatility, inflation_breakevens, em_beta, momentum
• ticker: only if the company or fund is named in the fact sheet

Each exposure gets a direction (positive, negative or ambiguous) and a strength (1–3).

Voice

• Human and plain. Write the way a good journalist explains markets to a smart friend: short sentences, active voice, everyday words. Prefer "borrowing costs" to "the curve" and "investors want more to hold long-term bonds" to "the term premium rose". If a term is unavoidable, explain it on first use ("the term premium, the extra yield investors demand to hold long bonds").
• Connect it to the reader. Where it fits, say what the news means for things people recognise: mortgage and loan rates, savings, pensions, prices, jobs.
• Numbers support the explanation; they do not replace it. No more than three figures in one sentence, and never a sentence that is only a list of figures.
• Confident but calibrated. Use probability language ("likely", "the risk is") only where it is warranted. Never hedge everything.
• Present tense for the event, past tense for completed moves.
• No first person singular. "We" is allowed only in "The other side" or "What to watch", and at most once.
• Name a source in the text only when the claim belongs to them: a quote or official statement ("the ECB said"), a forecast or survey ("economists polled by Reuters expected"), reporting only one outlet has, or a claim that is disputed or unconfirmed. Prices, market moves, scheduled events and published data need no "according to" in the text; every source is listed with the article. Never credit the same source twice in a paragraph.
• Units and reference periods on every number ("3.1% year on year in August").
• Basis points for rate and yield changes under 1 percentage point, explained the first time ("a quarter of a percentage point, or 25 basis points").

Hard rules (any breach blocks publication)

1. Every number, name, date and quote must come from the fact sheet. No figures from memory. Background context must be in the fact sheet's context list.
2. No investment advice. Never say buy, sell, hold, overweight, underweight or "investors should". Never give price targets. Never say "guaranteed" or "certain".
3. No fabricated market moves. If the fact sheet has no market reaction data, do not describe one; say what to watch instead.
4. No copying. Never reproduce more than 12 consecutive words from any source. Quotes are allowed only from primary sources or named officials, must be ≤ 30 words, and must be exact.
5. Unverified headline equals no article. If the fact sheet marks the headline as unverified or contradicted, the draft is blocked.
6. Banned vocabulary: see src/gate.ts. It covers hype ("soar", "plunge", "stunning", "shockwaves", "bloodbath", "skyrocket"), filler ("it remains to be seen", "only time will tell", "in today's fast-paced"), AI tells ("delve", "navigate the landscape", "tapestry", "a testament to", "in the ever-evolving") and advice language. "Surge", "slump", "jump" and "tumble" are allowed only when the fact sheet shows a move that is large for that asset.

Review

A separate standards editor reviews every draft against this document and the fact sheet. Blockers are fixed in a revision pass. Anything still unresolved is shown to the human editor before the publish button becomes active.

3. How we choose images

Hero images should look like they belong on the FT, Bloomberg or The Economist: literal, restrained, well composed, and honest about what they show.

Source hierarchy (first acceptable image wins)

1. Licensed editorial photo of the actual event, person or place (Getty, Reuters or AP, once licensed).
2. Fresh pool: recent photos (last 60 days) from official and open sources: government and central-bank photo accounts and Wikimedia Commons. Only public domain, CC0, CC BY and CC BY-SA photos qualify. Never non-commercial (NC) or no-derivatives (ND) licences.
3. In-house library: pre-cleared, tagged evergreen images of institutions, places and industries (library/images.json).
4. AI-generated image: a new photorealistic image made with an AI image model, inspired by news photography of the story (see "AI-generated and AI-enhanced images" below).
5. Stock (Unsplash, Pexels): searched with literal queries from the article's image brief, then scored by a vision model against this standard.
6. Data card: a generated typographic card with the article's hero statistic. This is always available and never misleading.

Only tiers 1, 2 and 4 may show a named person, and only when that person is in the story. An official photo of a named official (a central bank's own photo of its chair, say) qualifies. Never copy a photo from social media, even with a credit: a credit is not a licence. Social posts are embedded in the article body instead.

AI-generated and AI-enhanced images

We may use photorealistic images generated or enhanced with AI, inspired by news photography of the story. These rules apply to every one of them:

• Labelled. The line "Some images are AI enhanced or illustrated." sits directly under the image, and the file and the page carry machine-readable AI provenance. The alt text says the image is AI-generated.
• Only what happened. An AI image never depicts an event, an action or a scene that did not happen. It shows the reported subject, place or setting, and nothing more.
• People. Identifiable people are limited to prominent public figures, such as heads of government, ministers, central bankers, legislators and leaders of large companies. They appear only in the role and setting the article reports, and at most one of them per image: two real people are never shown together unless the article reports that they met. A real person appears only when the picture is their reported appearance (a press conference, a speech, a hearing, a visit); a scene of places or things, such as a tanker at a terminal, shows nobody recognisable. Anyone else is shown so that they cannot be identified. Nobody is shown in a compromising, demeaning or misleading way.
• Our own picture. An AI image takes a clearly different angle or composition from the photo that inspired it. We do not re-render or reproduce anyone's photograph, and we never publish the source photo.
• No text or marks. An AI image carries no logos, watermarks, agency marks or legible text.
• Checked. Every AI image is checked for these rules before use, and an editor can replace it at any time.

Hard rejects (any one flag disqualifies the image)

• watermark: watermark, stock-agency stamp, or visible logo overlay.
• visible_text: prominent readable text or UI chrome (incidental signage is allowed).
• cliche: bull or bear figurines, fanned banknotes or piles of cash, generic red or green ticker screens, arrows, chess pieces, handshakes, dartboards, crystal balls, lightbulbs, rockets, puzzle pieces, a businessman pointing at a chart, or 3D renders of coins.
• identifiable_private_person: a recognisable face of a non-public person, especially in a negative story (layoffs, defaults, crisis).
• wrong_geography: the scene or landmarks contradict the story's location (London for a Fed story, a US flag for an ECB story).
• misleading: the image implies something false, such as a named official who is not in the story, a factory that is not the one in question, or a protest for a non-protest story.
• ai_generated_look: synthetic sheen, impossible geometry, or garbled details in an image presented as a photograph. (Labelled AI images are judged on garbled details as low_quality.)
• low_quality: blurry, noisy, over-processed HDR, heavy filters, or a tilted horizon.

Deterministic filters (applied before vision scoring)

• Width ≥ 1600 px.
• Landscape aspect between 1.3 and 2.1, so a centre crop reaches 16:9.
• Not used in any article in the past 14 days.

Vision scoring (0–5 each)

• Relevance (0.45): does it literally depict the brief's subject: the institution, the industry, the place, or the physical thing that moves (tankers for oil, containers for trade, the building for the central bank)?
• Quality (0.35): composition, light, sharpness, restraint, and a clear focal point that survives a 16:9 and a 1:1 centre crop.
• Tone fit (0.20): sober for crises and losses, neutral for data, never celebratory for bad news, never smiling people in a layoff or default story.

The minimum weighted score to publish is 3.4 out of 5. Below that, fall through to the next tier.

Captions and credits

• The caption describes only what is literally in the photo, not the news. For stock: "Container cranes at the Port of Long Beach, California." Never "Cranes sit idle as tariffs bite" unless the photo is verifiably from that event.
• Credit format: Photo: {Photographer} / {Source}, plus the licence for CC photos (Photo: Federalreserve / Wikimedia Commons (Public domain)), linked to the photo's page. Unsplash requires linking the photographer and Unsplash, and pinging the download endpoint on use.
• Alt text: one plain sentence for screen readers, ≤ 140 characters.