Facts first. Then the question worth asking.

Tech and AI news, sorted each morning: what actually happened, what is still a pitch, and the one question worth asking. Short, free, and labelled.

Side A: the daily

Every story, split down the middle. Here’s how one reads.

Sample issue, built from our research notes. In reporting; not published yet.

ElevenLabs v4 wants one voice model to carry a whole conversation

Sep 28, 2026 · ElevenLabs announcement

Confirmed: ElevenLabs announced v4 and v4 Turbo: expressive voice with several speakers in one scene.
Not settled: The speed figures are the company’s own numbers for generation. They are not the time from your request to sound in your ears.
The question: When two speakers switch languages mid-scene, does each keep their own voice and pronunciation?

Also on this side

A2Canva ProSuiteWhat stays editable when a design moves into motion?

Confirmed
Editable creative workflows that reach into Affinity and Cavalry.

Not settled
Some announced integrations are still “coming soon.”

Source
Sep 16, 2026 · Canva announcement
A3Claude Code modsHow do you check, or switch off, a change you asked for last week?

Confirmed
You can change how the coding tool behaves by asking it to.

Not settled
Mods are not sandboxed. They run with your machine’s access.

Source
Oct 1, 2026 · Claude announcement
A4Google CC for householdsWhose information goes into the plan, and who gets to change it?

Confirmed
An agent that plans for a whole household, not one person.

Not settled
A U.S. early experiment with a waitlist, not a broad launch.

Source
Sep 17, 2026 · Google announcement
A5Cloudflare Clef and Clef-flashWhat happens when a typed, tidy answer is simply wrong?

Confirmed
Small models built for repeated classification and routing decisions.

Not settled
The benchmarks are Cloudflare’s own. Self-service fine-tuning is promised for later.

Source
Oct 1, 2026 · Cloudflare announcement · model

Side B: long reads

For when a story needs more than four lines.

OpenArm 2.0 wants robot experiments to travel

A robot succeeds in one lab. Another team runs the same model and gets different results. Was it the software, or did the camera angle, lighting and hardware change the test?

Confirmed: OpenArm Cell puts the arm, lighting, cameras and calibration into one standard enclosure.
Not settled: The documentation doesn’t establish how much the new gripper improves task success.
The question: Can a shared setup make a result from one lab repeatable in another?

Review draft · Robotics · publishes after review · OpenArm documentation

How we work

Confirmed and not settled are labelled, every time. If a sentence is a promise, it’s dressed like one.
Vendor numbers are called vendor numbers. Every company’s benchmark says the company did great. We tell you whose benchmark it is.
No required verdicts. Some stories end with an opinion. Most end with a question, because that’s usually where the story is.
A person reviews every story. Research tools help us draft. Nothing goes out until a person has read it.

Questions, answered plainly

How often does it arrive?

Daily. One email, short enough to finish before your coffee does.

When does it start?

Soon. The first issues are in review now, and signup opens with them.

What does it cost?

Nothing. Donations will be optional, and the email reads the same either way.

Who’s behind it?

An independent publication. Research tools help with drafts; a person reviews every story before it goes out.

Can I leave?

Any time, with the link at the bottom of every email.