Hichem Benali Case note Cabinet ↗

Case note ·

Convergence
before causality.

On 13 June 2026, the first Sendable? V0 appears in my Git history.

Its premise was simple. AI can produce the message. That does not decide whether the message should be sent. The V0 put a judgment layer immediately before that action.

01 / Provenance

A dated commit. Not a public claim.

monkidy/sendable-landing · f4ae11edprivate repository · screen capture, unmodified
A terminal showing the Git remote github.com/monkidy/sendable-landing and the commit f4ae11eddf617fa089e6b3d305322a3905a3b75d, authored 13 June 2026 at 17:02:21 +0200 and committed at 17:24:08 +0200, with the message Build Sendable V0 static landing.
Git history for the repository that became trysendable.com. The repository is private today. This note does not claim it was publicly visible in June, and nobody outside can audit that commit from here.

A second timestamp, written by a different system, agrees. The hosting project for that repository was created on 13 June 2026 at 17:34:40 Europe/Paris, roughly ten minutes after the commit. It is infrastructure metadata from my own account, not an independent attestation.

02 / What existed

A verdict. Not a rewrite.

trysendable.com ↗
commit f4ae11ed · app/page.tsxhistorical source, rendered at that commit
The Sendable? V0 page. A label reads PRE-SEND VERDICT LAYER above the line AI wrote it. You still own the outcome. Below, three receipt cards return the verdicts SEND, REVISE and DON'T SEND, each with a context, reasons, allowed actions, blocked actions and a receipt identifier.
Rendered from the historical source at that exact commit, in a detached checkout. Nothing was added, removed or restyled for this page.

Draft, then context, then risk, then verdict, then receipt. The three verdicts were closed: send, revise, don't send.

The output was not simply better prose. It was a bounded judgment about whether the draft should pass, with the allowed and blocked actions written down next to it.

03 / What I found

Another Sendable, live today.

usesendable.com ↗
usesendable.compublic website · captured 22 August 2026
The usesendable.com home page. The brand reads Sendable. The headline reads Paste the AI slop. Get the one-pager you can send. A panel invites the visitor to paste a draft and choose an audience, a goal and an output format.
Their public home page as it stood on the day of this note, captured whole and unannotated. No arrow, no circle, no highlight has been added to it.

It also operates in the space between AI-generated material and a human send. Its centre of gravity is different: a reader, a goal, a review, and a professional one-page memo you can forward.

That is a different object from a closed verdict with blocked actions. Related territory is not the same product, and I am not going to claim it is.

04 / The boundary

What this proves. What it does not.

Proven

  • Sendable? existed in my Git history on 13 June 2026
  • That V0 implemented a pre-send verdict layer
  • Its vocabulary included send, revise, don't send, and receipts
  • A separate current product named Sendable occupies a related pre-send problem space

Not proven

  • That anyone behind the other Sendable knew about Sendable?
  • That anyone copied anyone
  • Any causal relationship between the two products
  • That the private repository was publicly visible in June 2026

The left column is what the evidence on this page carries. The right column is what it does not, and I am not going to borrow from one to fill the other.

05 / Current reading

Convergence observed. Causality unknown.

A plausible story is not proof. Two products can arrive at one boundary without either having seen the other, and that possibility does not get quietly dropped because the other story is more interesting.

06 / The falsifier

The test that would change this page.

My earliest verified Sendable? artifact is dated 13 June 2026. If the team behind the other Sendable has an earlier dated artifact, I would genuinely like to see it.

If it predates mine, this page will be updated publicly. This is not a trap. It is the test that decides the question.

If no answer comes, that changes nothing. An absent reply is not an admission. Silence stays unknown.

The part I care about goes beyond the name. When AI makes production cheap, judgment becomes the scarce thing.

What should be allowed to pass? On what evidence? Who remains responsible once it is sent? Sendable? was one small answer to that question.

Receipts over claims.