Independent equity research,
built into repeatable workflows.

Seeking Edge produces research on U.S. equities and the AI supply chain: a scored map of the AI value chain, published here in full; single-name equity research, delivered to clients, with one note published as a sample; and the records of a quantitative programme that has since been retired. Every piece states its evidence, what would prove it wrong, and what has been corrected.

Read the sample research note Explore the AI Atlas Request research

Research for informational and educational purposes only. Not investment advice. Nothing here is sold, and no live model is offered.

What is current, as of 2026-09-21

AI Supply Chain Atlasedition 28 August 2026

Scoring data 8 August 2026. Thesis status board dated 21 September 2026: thesis 01 contested; scores unchanged.

Single-name equity researchsample v2 · 19 September 2026

Delivered to clients on request. One note is published as a sample, marked to 18 September 2026; it supersedes v1 of 18 September 2026.

Quant research programmeretired · last run 2026-04-29

Final snapshot published as a record. Window closed 2026-05-29; no evaluation published.

Working with Seeking Edge

What we help with, what you receive, how to start, and why the work can be trusted.

What we help with

Single-name expectation gaps: what today’s price requires from a company’s economics, whether the evidence supports it, and which observable decides it.
Supply-chain screening: where in the AI value chain supply is rigid while demand compounds, who can collect the rent, and which named holders are leads worth underwriting.
Thesis updates: an existing view re-scored against a new filing, print or data series, with the triggers that moved and the ones that did not.

What you receive

A written research note that answers six questions on its first screen: the question and its date, our judgment against the market’s expectation, how it was produced, the evidence that carries it, what would prove it wrong, and how it becomes a playbook.
The scenario model behind it: required-growth or value grids that re-solve from the stated inputs, rather than a point target.
An evidence list by source tier, with each load-bearing figure traced to a filing, a transcript or a data series.
Monitoring is not offered as a service. Each note lists the observables, thresholds and sources that would update it, so the view can be re-scored on the next print.

How to start

Send the name or the sector, the question you want answered, the time horizon, and the delivery format: a full note, a note with its model, or an update of an existing view.
We reply with the scope we can commit to, the sources we would rely on, and a timeline. Nothing about client work is published.

Why trust the work

A sample note is published in full with its version history: the correction from version 1 to version 2 is shown, not hidden.
Key sources are cited by tier on every note; a fitted parameter is never an anchor.
Corrections stay visible: the Atlas carries a corrections register and a primary-source triage, and the retired quant programme’s record stays published.
A named research lead is responsible for every published conclusion.

Request research Read the sample note

Three places to start

Each one states the question, the approach, the evidence that carries it, and its current status.

Industrial research

AI Supply-Chain Atlas

Where is supply rigid while demand compounds, who can actually collect the rent, and what evidence would change the call?

Approach: A 12-tier map of 131 chokepoints scored on six dimensions, with rent capture scored separately from how binding a constraint is. Five attempts to turn a scored node into a priced position all came back no; the rule that came out of them is that a named holder is a lead, not a position. A corrections register and a primary-source triage record what reversed.
Evidence: By the Atlas’s own count roughly 21% of its 1,050 source URLs are primary and 58% relay; 57 claims are machine-verified verbatim against a primary document.
Status: Edition 28 August 2026; scoring data as of 8 August 2026. Thesis status as of 21 September 2026: thesis 01 is contested by the triage; scores unchanged pending adjudication.

Start with the 90-second case →

Equity research

Single-name equity research

What does today’s price require, which observable decides it, and what would retire the view?

Approach: A full-length note on one company: what the market expects and how that was established, the mechanism and the two or three pieces of evidence that carry it by source tier, a scenario range rather than a point target, the observables that would retire the view, and a version history that shows each correction against the earlier version.
Evidence: Filings, analyst calls, exchange and market-curve data, and where relevant on-chain or industry data. A fitted parameter is never an anchor.
Status: Delivered to clients on request. One note is published as a sample: Circle Internet Group (CRCL), v2 of 19 September 2026, marked to the 18 September 2026 close.

Read the sample note →

Quant research programme (retired)

A cross-sectional one-month model

What a 30-name cross-sectional model of one-month forward returns produced over a 19-month walk-forward panel, and how it was validated.

Approach: A factor screen, a three-model ensemble, a Monte Carlo options overlay and a fixed validation battery, run on a bi-weekly cadence until April 2026. The programme is retired. Its final snapshot, its validation record and its unedited output file remain published as a record of the method.
Evidence: Walk-forward out-of-sample panel of 19 months (2024-09-30 → 2026-03-31); mean IC 0.027; the long-short leg carries no execution costs.
Status: Retired. Last run 2026-04-29; evaluation window closed 2026-05-29 without a computed outcome.

See the final snapshot →

The question and the evidence standard come first

Each piece states the question it answers, the date its data is marked to, and the source tier behind every load-bearing figure. AI-assisted tooling is used for sourcing and extraction; the question, the model and the triggers are set and reviewed by a human researcher, not by the tooling.

Every call carries its own falsifier

A view is published with the observable that would retire it and the threshold at which it would. When a leg of an argument fails a primary-source check, the status is written next to the claim, not buried at the end.

Corrections stay visible

A withdrawn thesis leg, a demoted source, a retired programme: each is kept, dated, and linked to what it changed. Corrections are shown in the open rather than overwritten.

Read the work, then argue with it.

Send the name, the question, the horizon and the delivery format. We reply with scope and timing.