Progress

Ariadni came to life over the July 4th weekend of 2026 and has been live in private beta testing ever since. This page is the running build log: what shipped to production, in order, on the way to the pilot and since it began.

July 6, 2026. Live, in private beta testing.

The app goes up at app.ariadnilegal.ai, invitation-only while testing runs: connect an inbox, name your counsel, and every document that arrives gets the four reads (responsiveness, who did the work, work quality, and value), with a dashboard that scores by lawyer, firm, and matter, pillar weighting so the overall reflects your priorities, and invoice review before you pay a bill.

July 8, 2026. AskAriadni, engagement letters, and the de-identification gate.

Email a document to Ariadni’s intake address and get an automated read back; the message is not retained after the reply is sent. Engagement-letter review ships, reading the letter as an arm’s-length contract before you sign. The de-identification demo goes public at /try, and every deploy now fails automatically if de-identification regresses.

July 9, 2026. The marketing site.

ariadnilegal.ai goes live with the waitlist.

July 19, 2026. Promises, kept.

When counsel commits to a date in their own words, Ariadni remembers, and tracks whether the commitment was kept.

July 21, 2026. Zapier.

Every event Ariadni produces (a scored document, a missed deadline) can fire into any Zap.

July 26, 2026. The methodology, written down.

Scoring moves to a versioned, evidence-gated rubric: every score carries its rubric version, past scores are never silently restated, and an automated gate fails any change that weakens issue detection.

August 3, 2026. The labyrinth.

Brand refresh: the drawn labyrinth and the thread.

August 8, 2026. The return benchmark.

Ariadni reports how long work of a given type and length has taken your counsel to come back, from their own observed track record. Descriptive pace, never a commitment. A sample dashboard also goes live: the real product on invented data, nothing to sign up for.

August 9, 2026. Benchmarks that test themselves, and signed scores.

Every rendered return benchmark is scored against the delivery that follows it, and a bounded correction learned from those results sharpens future benchmarks. Signed score records ship: a share link proves a record is authentic without revealing scores, names, or content. Sign-in grows to Google or Microsoft, with either mailbox.

August 10, 2026. Pilot-ready.

The pilot gate ships: a click-through pilot NDA at first sign-in, a one-click add-your-pilot-attorney flow, a participant guide, and the Terms of Use and Privacy Policy hardened for launch.

August 12, 2026. Scoring, rebalanced.

Work sent by the other side is no longer scored. It gets a plain-English summary and a risk read, and Ariadni says plainly that it is not your lawyer’s work. On later drafts of the same document, only your lawyer’s own edits are scored, and every matter carries two numbers: a score for the turn and a running score across the engagement. Risk is reported in bands rather than false precision, and engagement letters are read and explained but never graded. De-identification changes with it: names, firms, contact details and dates still go before anything is analyzed, but figures now stay, because a number identifies no one and removing it made the fee and invoice reads meaningless.

August 13, 2026. The pilot begins.

Invitations go out to the first group, and the pilot runs on synthetic matters end to end. Shipping alongside it: Ariadni now works out where a document came from. A draft your lawyer forwards from the other side is recognized as the other side’s, so your lawyer is neither credited nor blamed for someone else’s work, and you get told which edits favor you and which favor them. Sign-in moves to the front door at ariadnilegal.ai, with Google or Microsoft.

August 16, 2026. The first live sequence.

The pilot goes fully live: participants receive a synthetic engagement letter, a first draft, the counterparty’s redline, and the counter-markup, each read by Ariadni inside their own inboxes minutes after arrival. Every read lands as a plain-English email with a provision-by-provision translation attached as a PDF and, where Ariadni has questions for the lawyer, a polite draft reply staged in the participant’s own Drafts folder, ready to review. Scoring follows the August 12 rules: your lawyer’s own work is scored, and the other side’s paper is read but never graded.

August 17, 2026. A guarded lawyer list, and a sharper return benchmark.

Two upgrades land together. The lawyer list, which decides whose work gets scored and who Ariadni’s staged drafts address, is now guarded: when a new name appears in a document’s metadata, Ariadni asks whether the person is your lawyer rather than assuming, because analyses stay sharp when only your own lawyers are on the list; on the other side’s paper an off-list author is expected and taken in stride. And the return benchmark now reads deeper than the clock: beyond the email timestamps that time each turn, it draws on the document metadata of the deliverable itself, so the expectation reflects the kind of work actually in front of your lawyer, benchmarked against their own history with work of that kind. Self-scoped to one client’s own inbox, as always.