- A blank page and a page with a nought written on it are not the same statement, and two consoles were reading them as if they were. When the automatic safety rule stops an experiment it used to write literally nothing, which is exactly what "nobody has ever touched this" looks like, so a stopped experiment and an untouched one became indistinguishable. On anything that ships switched ON that splits the console in two: one pane badges it as sitting happily at its default while another correctly lists it as needing to be put back. Two panes of one console disagreeing about one experiment is the failure this entire line of work exists to end. A stop now writes explicit noughts whether a person or the machine did it, so a blank means exactly one thing everywhere.
- The board stopped accusing environments of drifting away from a setting they were already running. A missing row is not an allocation of nought, it is the absence that means "use the default", which is the definition of being at the default. The search lock ships on and had no row anywhere it had never been dialled, so its card said "not at default" while two other answers from the same server, and the restore rehearsal, all correctly said there was nothing to do. An experiment that was deliberately STOPPED still reads as drifted, because a stop writes a real row.
- The sweep that puts everything back learned to report rows for experiments that no longer exist. Reading the live development environment turned up three of them: one deleted experiment and two renamed slots. They are harmless, nothing consults them, but "there is nothing to restore" and "the table is clean" are different claims and the console was making the second while only the first was true. The first version of that report had the very blind spot it was added to fix: it read stored rows once per declared kind, so a kind emptied by deleting its only experiment became invisible, which is precisely the orphan most likely to exist in the first place.
- Every kind of experiment whose baseline is a decision now has to state it out loud. Two kinds were made to declare theirs on Friday; the other four were still working theirs out for themselves, which is the same bug with a longer fuse. The guess is correct right up until something ships switched on, and that is exactly how the search lock came to print "off" over a product that locks every on-site brief. The kinds whose baseline is a structural fact rather than a decision are left alone and say why.
- The specialist call card is the agent's to raise, and once a call is booked it stays where the client can see it. The card had reached a client exactly once per project, because the offer fired a single time and every identifier was derived from the project, so "can we set up that call", "send me those times again", "none of those work" and "I need to reschedule" all had nowhere to go. Mira can now raise a fresh card whenever the client asks for one, both agents read where the project stands with the desk from one shared line rather than each guessing, the booking records the exact moment and how strongly the person was identified, and a small pill under the top bar keeps the booked call in view. The other half is a refusal: for a client who is NOT routed to the desk there is no card, no times, and none are coming, and the agent is now told to read that absence just as carefully as it reads the line, instead of promising times that could never arrive.
- And the check that is supposed to catch a contradiction between a switch and an experiment had gone green and blind. It has two passes. Consolidating the platform taught the first pass the new vocabulary, so it kept matching and the run stayed green, while the second pass was still looking for a parameter that the new shape does not have, and therefore silently checked nothing at all. A flat contradiction was fed to it on purpose and it reported success.
A Sunday of 6 commits, five of them on the same distinction: an empty row and a row that says nought are different statements, and a console that cannot tell them apart will eventually tell two different stories about one experiment. The sixth is the specialist call card becoming something the client can ask for again, and something that stays on screen once it is booked, rather than a single offer that vanished after one look.