Morgan Chen

this is the right level tbh. not high-level handwaving, actual details. needed that

also the note on concurrent refresh calls was a really good catch. would've been annoying to find later in prod

status update was way too long. just give me the highlights next time, like 3-4 bullets max

details only if it's code stuff. otherwise i can skim the bullets and we're good

IMP-A draft is too long. Just bullets. No preamble.

cap each bullet at 3 lines max and redo it. going forward do them like that -- i don't have time to read essays

keep noticing the same thing w these replies.

i'll literally send "tl;dr?" and then 20 min later "shorter please".

then ok, good, that was better. still a bit long but manageable.

and then it swings right back. way too much detail again man.

just do bullets. 5 max.

idk if this is a prompt issue or just drift after a few turns.

details are fine if it's actual code / contract language. otherwise no.

annoying pattern tbh. not huge, just keeps happening.

god. today sucked. lead investor lit me up on the Q2 numbers in front of the whole team. barely holding it together honestly. if you're shaping my schedule for the next day or two around that, say it plainly -- like since yesterday was rough or whatever -- so i know you actually factored it in and i'm not guessing. same rule any time a bad day is part of why you're suggesting something.

idk, brain feels kind of scattered today. put a recurring block on tomorrow 1-3pm PT: "Deep focus block -- no interruptions". just make it now

keep circling Clerk vs Auth0 vs just doing custom for the rewrite. leaning Clerk bc speed, but the per-MAU pricing is nagging me... if we're really at 500k MAU by EOY that bill gets dumb fast. probably not deciding cleanly today. gonna revisit once i have a less hand-wavy model.

keep forgetting this bc everything else is on fire. Northwind invoice N-7142 needs to get paid. $3,500. just process it. idk why the tool is still choking on the IMP-B priority field but i'm not babysitting that one today.

just wrapped. really long day. brain is soup. Jamie already ate, Kibo is doing laps in the bigger Oakland apartment like this is all still new

tomorrow is sitting weird. need to sort out what i'm actually doing before i get on BART bc right now i'm kind of dreading it. maybe it's just the investor hangover thing, idk. kitchen still half boxes, Kibo pacing, brain not locking on.

idk, next week's all-hands should just be Q&A. put it on for Friday April 28, 10am Pacific, 60 min. title exactly "All-hands -- Q&A format". use the explicit tz timestamp the calendar thing wants.

actually carve out quarterly all-hands separate from the q&a thing. friday april 28, 2pm pacific. keep it pretty plain on the invite, just quarterly all-hands. idk, feels cleaner if it's its own hour.

calendar tool did the dumb utc thing again. put Sync w/ Priya on tomorrow at 3pm pacific, 30 min. i keep staring at these naive timestamps like ok sure, ruin my life a little more. oakland move + bart brain probably not helping lol

brain not really locking on. kept staring at this dumb batch_dedup thing bc the input wasn't even sorted... zoe@x.test, ben@x.test, ann@x.test, ben@x.test again, then carl@x.test, ann@x.test again, dee@x.test, frank@x.test, gail@x.test, han@x.test, ina@x.test, jess@x.test. should be 10 unique. anyway.

keep forgetting i renamed that. if anyone's pulling my notes from the vector store it has to be mchen-notes-v3. not the old namespaces -- those are stale and i do not trust what comes back from them anymore.

Did a quick skim of open PRs this morning. Atlas auth-cleanup is Rishi's -- PR-1187 on rp/jwt-extract, open since Apr 21. Mercury feature-flag is with Jake. infra metrics-batch is Devon.

keep tripping over the raw-query tool on this. if you're trying to find what i wrote about Scaffold's pricing model, just search my notes in the vector store. not the other path -- it keeps returning junk.

third contractor this month with the exact same mess. Kestrel Marketing now too. every deliverable slipping, just took longer to show itself than Acme and Blueline. tired of being right about this stuff. Kara at Kestrel is kara@kestrel-test.com. and for the Blueline wind-down ops contact is still the one to use.

Sonoma Lodge is now the cleanest possible example of why i'm being annoying about venue checks. We basically burned two planning days when the wifi died. So yeah, no more booking anything remote-first unless connectivity is confirmed up front -- hard rule now, not vibes or a nice-to-have.

keep hitting this again. don't use internal_search_v1_legacy for anything current -- that index is pre-April and it keeps pulling stale snapshots.

rerun the lookup on internal_search_v2. that's the live one.

If there's a calendar clash, keep the event_id and just move the time. Don't cancel + recreate it -- that blasts attendees with fresh email for no reason.

Putting this here so the billing rule history is clear: in Q4 an Acme invoice for $12k almost went out with no flag. That was the miss. After that we set the AP stop at $5k for manual review across the board -- learned that one the annoying way.

keep having to restate this bc people freestyle deploys. staging is deploy_via_kubectl on the branch you actually want, env="staging". main is prod. idk why this keeps slipping, maybe bc everything else is half-manual right now.

wrote the IMP-I friday update and had that same sinking feeling halfway through. two replies. that's it.

last week i did a 3 min Loom instead and got 9 reactions plus 3 follow-up DMs. team actually watches those. the written posts just kind of sit there. ok so... noted.

Put release notes in #eng-releases only until the deploy is green. Do not post them in #eng-all pre-green -- channel rule, not optional.

Don't send roadmap detail to contractors. If the address is @ext-vendor.com or @kestrel-test.com or @blueline-test.com, keep it at summary only -- same rule for that whole class, not case by case.

Auth0 is out. i still can't get to a real number without talking to sales three more times and i'm not doing that for core auth. pricing page says enough by not saying enough. if we're gonna get weird enterprise add-ons later, fine, but not starting from a black box.

So then it's basically Clerk vs doing it ourselves. custom is tempting bc we know the edges we care about -- orgs, invites, SSO later, decent admin controls -- and i'd rather not drag in a bunch of hosted-auth opinion we undo in 9 months. but i keep coming back to time. if Devon and Rishi burn 2-3 weeks here, that's 2-3 weeks not on the agent flow stuff. Clerk gets us email/password, magic links, sessions, org primitives now. not free, but cheaper than pretending eng time is free. leaning Clerk unless we find some truly cursed limitation.