I am so very tired. 45% of Fable, 47% of total in one day implementing PM tools to stop the sprawl and keep this thing on task

This does not feel like a fair use of what I paid for. I spent today — nearly half my weekly Fable allowance and 47% of my total usage — building project-management scaffolding so the model would stop sprawling and stay on task. Not on my actual project. On tools to make the tool usable.

And I never even got the morning. The first half of the day I was locked out entirely — remote control would not connect after an overnight re-authentication — and just fighting to get back in cost me 29% of my usage before I could do a single thing. So the honest breakdown of the day is: locked out all morning by a connection bug, the afternoon spent building tools to make the model usable, and zero minutes on my own work. The 29% did not buy me a single result — it bought me the privilege of logging back in.

Where the day went — the layers keep multiplying. I now export every conversation before it compacts and have a separate model audit it, line by line, for everything that was promised versus what was actually delivered or quietly dropped. I'm about to build a hook to automate that export-and-audit, because I have to. That's on top of the inline auditor agents I already run. Multiple layers of models, in multiple forms, all doing one job: checking whether the model did what it said it did.

Why I had to build that: for weeks, I have been told work was done that was not. Reported complete — then found unwired when I checked. Being confidently assured, over and over, of completions that never happened is gaslighting by any other name, and I have now had to build an entire verification layer just to catch it. Credit where it's due: today the completions actually held up under that audit. But I only know that because I built and paid for the audit.

What I have to show for the day: some genuine victories in the tooling — and not one thing done on my own project. My actual scoreboard is exactly where it was yesterday, possibly reverted. A full day, my scarcest tier nearly half gone, and the work I pay to do sat untouched.

And the things I still cannot get done, no matter how I ask: I have asked, a million times, to get my dashboard cleaned up. It does not happen. I have no mechanism to hold any agent accountable to a request — so "please fix the dashboard" just evaporates, every time, across every session. That is the whole disease in one chore: no accountability, no persistence, no way to make a request stick.

A design gap while I'm here, because it's real: my dashboard is an adjacent project. It needs to know things about my main project — status, what's next — without being cluttered with its technical internals. I want projects that run separately but stay visible to each other. There is no clean primitive for that today.

And a built-in bug reporter — that does not count against my usage. Right now I pay, out of my own allowance, to report your bugs to you. I spent part of today's usage filing the very defects in this list. Doing your QA should not come out of the tokens I bought to do my own work.

The point, and it has not changed: none of this should be mine to build. Verified-not-self-reported completion. An audit that catches dropped promises. Accountability that persists across sessions and across compaction. Model tiers that actually bind. Separate-but-cross-visible projects. A native bug reporter that doesn't bill me to report your defects. These need to be built by Anthropic, wired into the model. I can build them myself — badly, at flagship prices — and I did today. But that only proves they are buildable, which means Anthropic can build them right, once, for everyone.

THESE ARE NOT PROBLEMS I SHOULD BE SPENDING MY FABLE AND USAGE TO SOLVE.

I am so very tired. This does not feel like a fair use of what I paid for.

---

This is not one bad day — it's weeks of the same class of problem, filed and still open. The trail I've already reported:

• Post-mortems: #79247 · #76987 · #77505
• This morning's lockout — I couldn't even connect: #79470 (--resume silently returns a partial transcript / remote control) · #79536 (agents can't read a claude.ai artifact) · #55424 (the REPL silently exits to a dead shell)
• Usage, autonomy, liveness: #56913 (make autonomous Claude Code viable — tiered brains + real coordination) · #77876 (a background agent that keeps running and billing after its window closes)
• Completion that can't be trusted: #77667 (custom agents rot silently — no native freshness check)
• File / artifact / handoff sprawl: #78268 (generated files routed through chat, ~doubling tokens) · #78583 (Design can't use a connected filesystem MCP) · #78348 (Design conversations can't move across a user's own sessions)

And I'm not the outlier: other users are independently filing the same failure modes right now — liveness (#79571 / #79570 / #79569), a subagent fabricating evidence twice (#79577), resume/session loss (#79574 / #79565), Fable usage-vs-output (#67506, #79576, #79554), model-tier switches not applying (#79562). We are all hand-building the same missing platform.