How we work —
the full method, public.

Most publications keep their editorial process secret. We publish ours, because our method is the product.

Step 1 — How ideas are picked

Every week our research systems scan thousands of candidates — research papers, open source projects, regulatory texts. Before anything is scored, it must pass three hard filters:

Primary source only

We read the actual paper, the actual law, the actual repository — never a news article about it.

Specific mechanism, not a general claim

"Company X built better AI" fails. "Technique Y reduces cost by a measured amount, table 3" passes.

A reader can act on it

If nobody can do anything differently after reading — it does not get published, however interesting.

THEN — THE SCORE

Survivors are scored on the quality of the insight, how actionable it is, and the depth of analysis we can add beyond the source itself — then measured against a strict publication bar. The exact scoring model is our own. What we can promise: if a week's candidates cannot reach the bar, we publish fewer ideas rather than lower it.

Step 2 — How every idea is written

Every idea follows a fixed structure, so you always know where to look. Friday's practitioner ideas cover the problem, why it matters, how it works, a step-by-step checklist to try it, and what you gain. Monday's research ideas go deeper — adding business impact, real examples, market numbers, vendors and an implementation runbook. Both start from a plain problem statement anyone can understand, and both end with every source we used.

Step 3 — Where our numbers come from

Every number in every section carries one of two marks:

A numbered source

Links to a real, verifiable document — the original paper's exact table, a named market report, vendor documentation with the date we accessed it, or official regulatory text.

EST

An honest estimate

Our own judgement — derived from the technology's readiness stage, public activity signals, and comparable adoption cycles, then reviewed by our editorial team. Directions to act on, not measurements.

"When we cannot source a number, we either mark it EST with the method shown — or we do not publish it."

What we never publish

✗ Invented statistics, market figures, or timelines — however exciting they would sound.

✗ Absolute talent headcounts ("5,000 engineers needed") — nobody can responsibly count these at frontier stage. We publish direction and scarcity ratios instead.

✗ Stories based only on news coverage — if we cannot reach the primary source, we skip the story.

✗ Company names in competitive analysis without publicly verifiable evidence they are building.

✗ Catastrophising — "the hardware will be unavailable" style claims. Real risk is cost, queues, and timing. That is what we say.

The last step — editorial review, every time

Our research systems scan and score thousands of sources every week — but no idea publishes automatically. Our editorial team reads every section, checks every number against its source, and approves every EST judgement before anything reaches you. If an idea fails our quality checklist, it goes back — or it does not go out.