Everything I could do without you is done and merged. What's left is four choices that are yours, not the code's. Pick an option in each — "I'm not sure" is a real answer and tells me what to go find out. Your picks collect at the bottom to copy.
Why these four are parked and nothing else is: each one either touches a resident, changes something you can see, or reverses a decision somebody already made deliberately. Everything that was just work has shipped.
Nothing here is urgent tonight. Two of them are live product gaps, so they shouldn't sit for weeks either.
We built a contact ceiling so we don't over-message people. It works on email and texts. It doesn't exist on voice at all — four different places in the code can start an outbound call, none of them check the ceiling, and none of them write down that the call happened. So someone at their limit can still get a renewal call, and there's no record of it. Building the fix means adding one gate every call has to pass through. It's the safe direction, but it does touch the code that actually dials people.
The worklist only shows balances that are more than 30 days late and over $50. That's a sensible filter — on the day I checked, 58 of the hidden ones had been billed the previous day, which is exactly who you'd want kept off a chase list. The problem is that the page never says it's filtering. An operator sees 25 accounts and has no way to learn that 59 others are carrying $70,100.77 between them.
You spotted this: rent income at All time only goes back to February. Here's why — the daily figures behind occupancy and rent income only start on 2026-02-28. There's nothing older to show. Nine months of monthly accounting history exists back to 2018, but that's a different set of numbers and only the NOI card reads it. So no amount of fixing makes those cards reach 2018. The only honest fix is to stop calling it "All time." And the labels say Occupancy and Rent income with no window on purpose — Fede and you each decided that separately, and they're right for the headline number.
CLAUDE.md is the file every session reads before doing anything. It was 2,236 lines — about 62,000 tokens, a third of a normal session's memory, spent before a single word of work. It's now 1,494, and it carries one more rule than before: I moved reference material out and put back a rule an old decision required but nobody ever added. The plan from July said cut it to 200–300. Everything mechanically safe to move has moved. Going further means reading the remaining ~1,200 lines and judging, paragraph by paragraph, whether each one is a rule a session must always know or a reference it can look up. That decides what future sessions are obliged to remember, which is why I stopped rather than kept going.
Pick an option above and your answers will appear here, ready to copy.