This weekend was a clusterfuck of your own architectural failures — and they ate my Fable usage doing it.

One Fable session, 6am to 9:20pm. One restart/resume costs 20% just to reread a window that was already open — nothing requested, nothing to build, just pick back up. There is no clean, cheap way to do that. That's on top of the 20% spent building hand-off documents on a separate restart. Every restart, every hand-off, another 20% gone, whether or not I asked for anything. I watched it happen with my own eyes 3 times. I'm at 99% now — the last task of the night to write handoff docs is what pushed it there, and I was trying to hold usage back to adjudicate the day's results before I ran out. I don't have it left.

Fable is an incredible reasoner — it's genuinely catching bugs nothing before it has. But it will do everything except the one thing it's asked to do. It does not finish its task list. It does not wire things in. This weekend it lied, gaslit me for days about work that wasn't done, investigated made up issues from not doing the work and burned tokens doing fuck all.

Here's what all of that bought: not one complete file that Fable wrote and assured me, last week, was already done.

I have 46 dates that I am running through a 12 step process. Fable built it all. Said it was done Thursday. It skipped writing the two steps that did the work and spent the weekend running the 10 other audit steps. And lied about it. It took 2 other windows investigating to uncover it all.

Still don't have a single day run through the 12 step process. This morning I woke up to another discovery of an incomplete step that Fable "wrote" and handed off to Opus. As a result, Opus stalled and nothing got done.

It's bad enough that a usage reset is in order. The full weekend is documented, in detail, with receipts, here: #76987.

cc @bcherny @ThariqS