These cards model the calls owners worry about, from corrected contact details to a caller asking for a person. The published results come from a browser simulation with fictional businesses, not live telephone calls or customer outcomes. Each card also suggests a separate test on Chleb's own business line; no live voice results are recorded in this log.
A receptionist that greets well and books a routine visit tells you little. The calls that cost a business money are the awkward ones: a caller who gives the wrong number and fixes it, a caller who wants a price the owner never set, a caller who needs a person right now, and a caller whose problem is an emergency. Each card names one of those moves, says what to try and says what a good answer does, so you can judge the result yourself instead of taking our word for it.
The sandbox runs a simulation of the call flow in your browser. It follows the rules Chleb configures, such as reading details back before booking, offering only open times and sending urgent calls to a person, but it is a small set of fixed rules, not the production receptionist. The businesses in it are fictional. You can type, tap the card's lines in order or play the whole attempt, and you see the records the business would get: the contact, a note, any booking and any handoff.
The live line is Chleb's own business line, (813) 694-8671, and it books on Chleb's real calendar. The live voice agent there answers as Chleb's receptionist, not as the fictional businesses in the sandbox, so each card says how to try the same move in Chleb's context: correct your email when it spells it back, or ask for a person halfway through. Interruptions and silence can only really be tested on a phone call, so those cards point there.
Before publishing, we ran every sandbox attempt through the simulation and wrote down each reply that fell short of the card's good answer. Each entry gives the date it was found, the card, the exact words said and the exact reply, whether it happened in the simulation or on the live line, and its status: open, fixed, or not fixing with the reason. An automated test replays every quoted failure, so an entry cannot stay open once the simulation stops making that mistake, and it cannot be misquoted. An attempt shown as passing says what the simulation did, and the same test checks each part of that sentence.
Only the simulation has been tested so far. No results from the live voice agent are in the log. Any that are added carry the date and what was said, and are marked as live so they are never mixed up with the simulation.
The report form opens your own email app with the report filled in, addressed to chleb@chleb.ai; nothing is sent from the page itself. A report is added to the log only if it reproduces: a sandbox report when the same words give the same reply in the simulation, a live report when Chleb's line does the same thing again. An entry gives the date, the card, the words and where it happened. Adding one means changing this page, so a report can take a while to appear, and a report that does not reproduce is not logged. Your name and contact details are never published.
When a failure is fixed, the entry stays in the log, marked fixed with the date and a line on what changed, and the replay test then checks that the old reply no longer comes back. If a failure will not be fixed, the entry says so and says why.
A failure in the sandbox shows where a small rule-based model of the call flow falls short. It is not a measurement of the live receptionist, and a pass in the sandbox is not a guarantee about any real call. What a buyer should take from this page is a list of moves worth trying on any receptionist, including ours, before moving a business number to it. A Chleb setup is configured around a business's own services and tested with calls before the number goes live.
No. It is a simulation of the call flow that runs in your browser with fictional businesses and your browser's voice. The live voice agent answers Chleb's business line, (813) 694-8671, where it answers as Chleb's receptionist.
Because a demo that only shows the easy calls tells a buyer nothing about the calls that go wrong. Quoting the exact words lets you judge how bad each failure is, and the log shows whether it gets fixed.
The sandbox uses fictional businesses. To walk through your own services, hours and hardest calls with the people who set the receptionist up, book a demo.
Only if it reproduces. Then the card, the words said and the reply go in the log with the date. Your name, business and contact details are not published.