Real operations. Running live.

On the recordFictional identities; all outbound email routed to one stand-in inbox; the signature ran against a real e-signature provider; the chat exchange was a real person at the keyboard. One gap named on the page: after the Teams close, the same case briefly still showed the underlying request as open on another screen, found and fixed the same day. Not yet shown: the inbound requirements letter arriving in a real mailbox (deferred to the next run).

Some screens shown here are from the current development build; the hosted demo environment is one release behind.

On the recordChase cadence time-compressed; the secure link is standard sensitivity (the platform refused to downgrade a PHI-grade request rather than send one); the settlement letter is issued but not yet attached to the email; outbound routed to a stand-in inbox.

On the recordThe vendor’s reply is sent programmatically from the workspace’s own mailbox because the mail provider can only send as its credentialed account (the accepted design, named in the report). The signature is applied through the e-signature provider’s own API rather than drawn on screen (headless cannot draw). Outbound routed to a stand-in inbox narrated as the vendor’s AP mailbox. The chase-loop case keeps its August captures.

Some screens shown here are from the current development build; the hosted demo environment is one release behind.

On the recordRecorded against the public demo box at release 1.3.0; the box is kept stopped when not in use. The wrong-vertical gap the August edition named on-page is now closed, and captured as a scene.

On the recordWhat was arranged for the demo is stated on the screen it affects: the safety-limit test uses a disposable case with a deliberately low limit and pre-filled details, the calibration labels are entered by the test script from a fixed rule, and the emergency brake is pulled and released within seconds. Four results are shown as honest “not yet” states rather than cleaned up: the drift check refuses to call a trend, the judge is not calibrated so rubric scores read “—”, the audit panel’s event preview sits below the fold in its capture, and the safety limit is not exact under parallel work. The July capture of this demo had the breaker leak captioned as a quirk; the re-drive proved it was a real defect, and it was fixed.

On the recordRe-driven 5 October 2026 to add the applicant chase. The one-day reminder was sent by running the real reminder process as if 25 hours had passed; nothing was edited by hand. Lab, physician and signature documents are generated sample files, and the records, underwriting and catalog services are simulated. Fictional applicants; the stand-in contact address only ever appears masked.

Some screens shown here are from the current development build; the hosted demo environment is one release behind.

On the recordFictional company and people. The two board items that opened the second and third cases were written by the cases themselves onto a real monday.com account and opened through its real webhooks; the Microsoft account and the two register rows were checked directly in the directory and the lists. The candidate’s email reply was sent over Microsoft Graph from the company’s own mailbox. The inbox, the matching and the fulfilment are real; only the sender identity is stood in for. Both signatures were applied through the signing service’s own interface. The background screen is an outside vendor; the plant historian is outside the platform by design; company groups and a mailbox were not created. The candidate’s stand-in contact address is masked in seven frames.

Some screens shown here are from the current development build; the hosted demo environment is one release behind.

On the recordFour things were arranged by the test and each is flagged on the screen where it happens: the broken configuration was planted by a script, the monitoring sweeps that normally run hourly were triggered on demand, the fix was applied by a script (the operator’s own action is the re-run), and the deep link was opened by the test rather than clicked from an inbox. Stated on the page: the alert email is recorded as accepted by the mail provider, but no inbox was checked, so arrival is not claimed.

On the recordEvery frame is from one continuous build-and-run session against a live environment. The workflow, agent, form, and case were all created during the session shown.

On the recordRetailer decisions and payments are simulated, so the recovered dollars show a case handled through to cash, not a win rate. The detection engine is a stand-in with scripted scores. One 30-day hold was released by running the real timer as if 427 hours had passed. Fictional firm, supplier and data; all email went to a test inbox.

Some screens shown here are from the current development build; the hosted demo environment is one release behind.