One headline. Opposite trades.
“Iran threatens to close the Strait of Hormuz” lifts crude, LNG and tankers, and sinks airlines. Newsleaf tells investors which way each energy story pushes their holdings, every weekday at 06:00.
Example event · Day 1, 05:40 · 4 outlets
Energy news is loud, repetitive and out of date by Wednesday.
Investors holding oil, gas, shipping or airline positions read the same story again and again and still can’t tell which way it moves their positions.
Dozens of articles. One event.
One development produces dozens of near-identical stories. A reader can’t quickly tell a new development from the same story published again.
Bad news is not a direction.
Sentiment tools score the tone of a headline. “Iran threatens to close the Strait of Hormuz” scores as bad news. It raises crude prices and lowers airline shares.
Monday’s threat, Wednesday’s retraction.
News expires. A threat made on Monday can be withdrawn by Wednesday, and readers are left tracking which earlier stories still hold.
Brent +45%since the start of the war in Iran, as of May 2026
Oil flowing through the Strait of Hormuz, million barrels a day · EIA via Transport Topics · EIA
The morning issue lists only what changed.
New, weakened, unchanged. Then which way each change pushes crude, LNG, tankers and airlines, with every item linked to the articles behind it.
One line per development. Overnight, Newsleaf pulls every new energy story, merges the duplicates and compares the result with the previous graph.
Direction comes with a reason. Each call names the event behind it and why it moves the price: jet fuel follows crude, and fuel is one of an airline’s largest costs.
Old stories fade. They don’t vanish. When talks undercut a threat, the issue says so, and the threat stays on record at a lower weight.
Search the graph
Answers questions about any topic, company, ticker or place, as of today or any earlier day.
How a headline becomes a direction.
Newsleaf keeps the news as a graph: events, the countries, companies and places involved, and the investments each event moves. Eight steps build it, shown here on the Hormuz story.
Collect
Six sources, watched around the clock. Nothing is thrown away.
01Watch
A watcher agent sits on each source and checks it on its own schedule. It keeps a bookmark at the last item it handed over, drops anything that isn’t about energy, and passes the rest to a run. When a story breaks, several watchers catch it within minutes and their items land in the same run.
The first run has no bookmarks, so everything each source shows goes in and becomes the first graph. Every run after that adds only what’s new. Bloomberg and the Financial Times are out of version 1: their terms forbid automated collection.
02Ingest
Every new article is stored exactly as fetched, with the fetch time. Nothing downstream edits the archive, so old articles can be reprocessed whenever extraction improves.
fetched 05:40
Understand
From prose to a typed event, linked to the investments it moves.
03Extract and classify
A language model reads each article and marks who acted, what happened and where. A classifier tags its topics. This one is oil and shipping.
type: supply threat
04Merge
Four outlets, one event. Articles about the same development merge and the outlet count is kept. “Tehran” joins “Iran”; “the strait” joins “Strait of Hormuz”.
05Link to investments
A table kept by the team maps each type of event to the investments it moves, the direction and the reason. A person reviews every link. A threat to Gulf supply moves four.
Remember
Every arrow knows when it was true. The graph can be replayed to any morning.
06Store with time
Each link becomes an arrow carrying its direction, the day the event happened, the moment Newsleaf learned of it, and a weight that fades with age. For this type of event it halves every 14 days.
Arrows are never edited or deleted. A later story that weakens an earlier one is a new arrow between the two.
- Happened
- Day 1
- Learned
- Day 1, 05:40
- Weight
- Day 1 · fullDay 8Day 15 · half
07Delta
After each run Newsleaf lists what it added and what it weakened. That list is the delta, and it is what the morning issue is made of.
- pushes up
- pushes down
- slow effect
- weakened
- weakens
08Publish
At 06:00 on weekdays an editor checks the delta and it goes out as the issue. Search reads the whole graph, at any hour.
Others sell scores or opinions. Newsleaf sells a record.
Every direction Newsleaf publishes is logged and scored against the price move that followed. The scorecard is public.
News analytics feeds
- For example
- RavenPack, Bloomberg’s news sentiment data
- Sold to
- Funds with quantitative analysts
- The buyer gets
- Scores and data feeds for building trading models
- Checked by
- The buyer’s own backtests
Paid energy newsletters
- For example
- Doomberg, Energy Outlook Advisors
- Sold to
- Individual investors
- The buyer gets
- One writer’s analysis
- Checked by
- Usually nobody, systematically
Newsleaf
- Sold to
- Individual investors and financial advisers
- The buyer gets
- A daily list of changes, with a direction for each affected investment
- Checked by
- Anyone, on the public scorecard
Priced under the incumbents until the record earns more.
Newsleaf starts below the energy newsletters because it has no track record yet. The price rises once the scorecard covers a full year.
$300 a yearor $30 a month, sold through Substack
After Substack’s 10% fee and card fees, each $300 subscription pays Newsleaf about $259.
Forty annual subscribers cover the running costs.
Build plan
- Months 1–3The weekday issue
Watchers, archive, extraction, merging, the graph and the delta. The team writes each issue’s investment links by hand.
TestsDoes it catch the stories that move prices, and will readers pay?
- Months 4–6A public scorecard
The event-type table, and a log scoring every published direction against the price move that followed.
TestsHow often the directions are right.
- Months 7–12Search for subscribers
Search over the live graph.
TestsDo subscribers come back between issues?
- After month 12Data for funds
A feed of the graph’s full history, sold per contract, and a second sector.
TestsDoes the stored history sell on its own?
$25,000
for the first twelve months.
It covers data, models, hosting, legal work and the first subscribers while the weekday issue, the scorecard and search are built.
- News data$5,388
NewsAPI.org Business plan at $449 a month
- Model usage$3,000
$250 a month. Extracting 1,000 articles a day costs about $3.50 on Claude Haiku 4.5; the rest writes issues and runs search
- Hosting and database$1,800
About $150 a month
- Legal$5,000
A founders’ agreement and a securities lawyer’s review of the issue and search
- First subscribers$6,000
Free trials and advertising aimed at energy investors
- Contingency$3,812
About 15%
Sairam Krishnan
Product and technical lead
Designed the architecture and the build plan. Builds the product.
Katera Mujadidi
CFO and venture capital specialist
Originated the Newsleaf idea.
The graph starts recording on the first run. Every later product sells that history.