Morgan Chen

Greg emailed that Acme admins are seeing API v2 exports time out, and Jake gave me a 3 p.m. patch ETA. Please reply to Greg in the existing Acme thread with my usual signoff: acknowledge the export timeout issue, give the 3 p.m. ETA, and keep it entirely about the API v2 export bug. No Mercury or Evergreen references.

Marcus found the AWS spike: forgotten non-prod snapshots left after auth-rewrite load testing. His proposed cleanup is deleting snapshots older than 14 days, estimated around $1,200/month savings. Please DM Marcus in our internal team chat approving that cleanup and ask for a one-line confirmation afterward. Make clear this is not approval for new tooling or broader infrastructure work.

Kara’s revised sidebar pass is here: From: Kara <kara@kestrel-test.com> To: Morgan Chen <morgan@atlas-test.com> Cc: Sarah Kim <sarah@atlas-test.com> Subject: Re: Mercury sidebar replacement Date: Thu, 5 Oct 2023 08:26:41 -0700 Got it on no Evergreen name/logo and no “enterprise validation” line. I took another pass and kept it generic. Proposed sidebar replacement below: Eyebrow Design-partner informed Headline Mercury is being hardened through bounded design-partner use Body Built with feedback from a regulated design partner and early admin users, Mercury supports the current setup flow with magic links, org invites, and a preview experience for initial teams. Optional alt if you want slightly stronger wording: Validated through bounded design-partner use and admin feedback on setup, invites, and preview workflows. We removed the customer name and the 32% metric from the sidebar comp. Two optional presentation tweaks if either is acceptable: - a generic regulated-team badge/icon with no customer logo - a tiny sub-line under the header that says “active design-partner testing” If this is finally in the safe zone, I'll update the page file and hold publish until I hear back. Kara Kestrel Marketing Please email Kara in the existing thread, cc Sarah Kim, with my usual signoff. Approve the safer direction, but have her remove “validated” wherever it appears. Evergreen stays unnamed, and Kestrel should not add any customer logo or metric.

Jordan’s Honeycomb demo draft: Discord DM — Jordan to Morgan 2023-10-05 11:17 Can you gut-check this before I post it in #eng-team? Feels maybe too long / too vendor-y. Draft blurb: "Quick observability demo: Honeycomb is the tool that finally made reliability debugging feel basically solved for us. We can follow a request end to end across API v2, auth, and Mercury, jump from a vague customer symptom to the exact failing branch, and answer 'what actually happened?' without guesswork." Possible demo flow: - Start with a real-ish bug: "invite accepted, then sync looked stuck" - Filter by workspace + endpoint to narrow the trace set - Jump into the trace where auth hands off and the retry path fans out - Use derived columns / breakdown to isolate the error bucket - Click into the offending trace and show the malformed retry reason - Close with "this is why Honeycomb won for end-to-end debugging" If that's still too much, I can cut it way down and just show the one trace. Main goal is to make the win concrete without turning it into a vendor pitch. Edit this down and send Jordan the cleaned-up version in our internal team chat. Make it a two-minute debugging win, not a vendor pitch, and remove any claim that reliability is “solved.” Include one suggested demo flow using the invite-accepted / sync-looked-stuck trace.

Sarah’s draft agenda needs a boundary pass: From: Sarah Kim <sarah@atlas-test.com> To: Morgan Chen <morgan@atlas-test.com>, Devon Hayes <devon@atlas-test.com> Subject: Draft agenda — Evergreen second admin group check Date: Fri, 6 Oct 2023 09:12:03 -0700 Sharing a first pass for the Evergreen second-admin-group check. The only section title I'm unsure about is #2 — if "expansion approval" is too strong, I can rename before I send it over. Draft agenda (45 min) 1) Current bounded test status (10 min) - magic-link entry + org-invite flow - preview usage / what the current admins have actually touched - open notes from the current admin group 2) Expansion approval (15 min) - confirm whether Evergreen can add the second admin group during the current test window - confirm whether the admin vs billing-owner separation gap is acceptable for that expansion or needs explicit sign-off - decide whether the remaining enterprise-readiness items should be framed as blockers or as follow-up items 3) Gaps and follow-up asks (10 min) - SSO request - audit trail / admin controls questions - procurement / review-packet questions that may come back up 4) Next steps + recap owner (10 min) - who sends the customer-safe recap - what we log internally for later-scope follow-up - what we hold until Morgan/Devon weigh in If it's cleaner, I can split section 3 into "current test issues" and "later-phase asks," but I wasn't sure whether that makes the meeting heavier than it needs to be. Sarah Rewrite the response I should send back. Rename this as a bounded test review, not “expansion approval.” Say the current scope is unchanged until Devon and I review the current-admin notes next week. Enterprise-readiness gaps — SSO, audit trail/admin controls, admin vs billing-owner separation, procurement follow-up — should be next-phase items, not launch blockers and not approvals.

Jake confirmed the Acme export timeout patch is live and Greg’s test export passed. Please email Greg a short closure note in the existing thread with my usual signoff: glad the test export passed, ask him to flag it if the timeout recurs, and keep it scoped to the API v2 export fix. Also thank Jake separately in our internal team chat.

Devon sent three proposed warm investor coffees for next week. Please DM him in Discord: after Elena's calibration, random coffees are off the calendar. The only open conversations are short-list, evidence-focused calibration; everything else can wait.

Jamie says BART is running late and asked me to handle dinner. Please place the Lemongrass order for 7:30 pickup: tofu curry, basil chicken, papaya salad, coconut rice, and no peanuts on Jamie’s portion.

Maya confirmed Mom can do brunch in Oakland at 11. Please send Maya a short SMS: 11 works, Jamie is coming, and let’s keep it to brunch only — no extra afternoon commitment.

Sarah sent the Evergreen second-admin-group roster and customer agenda. Email her back with an agenda framed as a bounded test review. Then DM Devon that he owns budgetary and commercial questions, and DM Jake that customer-facing product facts should stay to the current Mercury test flow: magic links, org invites, preview-only sample data, real-source connection, and first live sync.

Priya’s usability notes are here: Onboarding flow usability notes — invite accepted / pre-repo-connect Author: Priya Date: 2023-10-09 Screen tested Current empty state after invite acceptance and before first repo connect: Headline: "Connect a repo when you’re ready" Support line: "You can finish setup now or come back later from workspace settings." CTA in prototype: "Connect repo" Call 1 — 2023-10-04 Participant profile: engineering manager at a 14-person team; invited by a founder, not the person who set up the workspace. Observed behavior: - Read the screen without hesitation and said the tone felt "less pushy than most setup flows." - Hovered on the CTA, then stopped to ask whether picking the wrong repo would be hard to undo. - Did not interpret the step as skippable forever; interpreted it as the next real task, just not necessarily for this exact moment. Quoted reactions: - "Okay, good — this isn’t yelling at me." - "I probably do need to connect something, just not before I ask who owns the repo list here." Takeaway: the low-pressure headline reduced anxiety for someone who had accepted an invite but did not yet have enough context to act. Call 2 — 2023-10-05 Participant profile: platform lead joining an existing team workspace. Observed behavior: - Understood what a repo connect means, but paused on the phrase "when you’re ready." - Asked what happens if they skip and go explore the app first. - Wanted a slightly more direct statement of the task. Quoted reactions: - "‘When you’re ready’ makes me think this is optional, but I know you probably want me to do it next." - "If this is the setup step, I’d rather it just say that more plainly." Takeaway: the current copy is friendly, but maybe a little too soft on what the next action actually is. Call 3 — 2023-10-06 Participant profile: product engineer invited into a workspace but not ready to install anything during the call. Observed behavior: - Liked the support line about coming back later from workspace settings. - Asked what connecting a repo unlocks and whether there is any timing pressure. - Verbally connected the screen to trial/billing timing even though nothing on the page mentions trial. Quoted reactions: - "Come back later from workspace settings is helpful; that tells me I’m not burning my one chance." - "If the clock is already running on something, I’d want the page to say that." - "I’m fine doing this later, but then tell me exactly what I’m postponing." Takeaway: flexibility message works, but some users still want a crisper statement of the next step and what changes after they do it. Patterns across all three calls - Nobody reacted badly to the softer direction; the tone is doing useful work. - Two of three users wanted the next action stated more concretely. - One of three explicitly asked about trial/timing pressure; that question came from uncertainty, not from the current page creating urgency. - The support line tested better than the headline. People liked knowing they could return from workspace settings. Two copy changes I’d test next 1. Simpler repo-connection prompt within the same overall direction Proposed headline: "Connect your first repo" Proposed support line: "Finish setup now or come back later from workspace settings." Why: keeps the lower-pressure feel, but makes the task more concrete and removes some of the marketing-softness from "when you’re ready." 2. Add explicit trial/timing context Proposed headline: "Connect your first repo" Proposed support line: "Your team trial is active. Connect a repo now or come back later from workspace settings." Why: addresses the "is some clock already running?" question directly, but it definitely pushes the screen into a more sales/billing-coded tone. My read from the calls: the problem is mostly clarity of the repo step, not resistance to the relaxed tone. Please synthesize this into a short decision reply to Priya in our internal team chat. Decision: simplify the repo-connection prompt within the already selected empty-state direction. Do not bring back trial-activation language.

Morning billing retry digest: generated 2023-10-09 08:12 PT after replay batch stripe_replay_2023-10-09_0758_pt, covering 2023-10-09 00:00 through 08:10 PT. Four Stripe renewal rows are still stuck after replay, all with replay receipts, no duplicate-charge signal, unchanged payment_intents, zero new charge objects, grace_active access, and no extra dunning email. Rows: ren_44181 / ws_2f3d / in_rnwl_10041, last webhook 07:51:13 invoice.payment_failed, replay 07:58:02, retry_state replay_seen_not_requeued, same pi_8ab1, ReplayGuard hit because invoice already bound to pi_8ab1; ren_44197 / ws_77c0 / in_rnwl_10058, last webhook 07:52:47, replay 07:58:03, retry_state lock_held_after_replay, same pi_8ab9, StateTransitionError expected pending_retry -> requeued but found replay_seen with ledger_lock_age 00:05:11; ren_44206 / ws_91aa / in_rnwl_10064, last webhook 07:54:09, replay 07:58:05, retry_state replay_applied_waiting_worker, same pi_8ac4, RetryWorker skipped enqueue because idempotency slot renewal:in_rnwl_10064 is occupied; ren_44212 / ws_b14e / in_rnwl_10072, last webhook 07:55:31, replay 07:58:06, retry_state manual_review_required, same pi_8ac8, SafeReplay=false because prior collect attempt unresolved. The log snippets match those errors, duplicate_charge_scan(last_24h, source=stripe, kind=renewal) returned 0 rows with more than one charge id, the replay batch only touched renewals, and no row shows a successful post-replay collect attempt yet. Please DM Marcus in Discord with the four rows, ask him to confirm the retry state, replay only those stuck renewals if safe and without second-collect risk, and send me one customer-support sentence in case anyone asks.

Jordan’s post-demo Q&A notes are the input: Post-demo Q&A notes — Honeycomb debugging walkthrough Author: Jordan Date: 2023-10-10 Demo path I showed - Started from the alert on elevated latency for the repo-connect path. - Filtered to prod, narrowed to one affected workspace, then split by build_sha and route. - From there it took about 2 minutes to isolate the bad build / one worker path that was missing the expected timeout default. - The useful moment was not the exact root cause; it was how fast the query got us from "latency is up" to "this is isolated to one deploy shape." I clocked it at a little over two minutes in the live run. Team Q&A Q: Was the quick path mostly because you already knew where to look? A: Partly, but the point was that I didn’t have to jump across three tools first. The filters that mattered were already in the event: workspace_id, route, build_sha, and trace link. Q: Does this replace logs for incident work? A: No. It gets us to the narrow slice fast. I still wanted logs once we had the suspect worker and request family. Q: Could support or product use the same flow when a customer reports "something felt slow"? A: Yes for narrowing scope, as long as the report includes workspace and rough time window. It is still an engineering handoff tool more than a support surface. Q: Does this mean reliability is fixed if the query path is this good now? A: No. It shortens time-to-understand. It does not prevent bad deploys, fix flaky workers, or help when the service never emitted the field we need. Q: What made the demo work cleanly? A: The endpoint already had the right high-cardinality fields and trace propagation. If those are missing, the experience falls apart fast. Q: Should we be instrumenting older worker paths before claiming this pattern broadly? A: Yes. The nice demo path is real, but it is not universal yet. My rough recap bullets - Best concrete win: we went from alert to suspect deploy in roughly 2 minutes instead of doing the usual dashboard -> logs -> grep loop. - This was a good example of high-cardinality fields paying for themselves in a real debugging path. - Important caveat: the walkthrough did not solve the underlying reliability issue; it only made the bad slice obvious faster. - Also important caveat: this depends on good instrumentation hygiene. Where fields or trace links are missing, the magic disappears. - If we post a recap, I’d keep it practical and small — more "here’s a debugging speedup we got" than "observability victory lap." Please turn this into a concise follow-up for the current internal team channel. Highlight the roughly two-minute debugging win, include one caveat that reliability is not solved, and thank Jordan without making it sound like a platform victory lap.

Marcus’s auth-rewrite rollout checklist is here: Auth rewrite rollout checklist Owner: Marcus Date: 2023-10-11 Scope: API v2 auth path only. Existing API behavior remains GraphQL + JWT on the client-facing side; this rollout changes how session revalidation happens behind that path. Status snapshot - Staging pass completed on the happy path. - JWT refresh behavior looks correct in the tested web flows. - Rollback path is defined, but I have not finished the last prod-like dry run on the rollback toggle. - Release note wording is still open, mainly on whether we name Clerk directly or keep the note provider-neutral. Preflight items [x] Staging login -> dashboard load on API v2 with fresh JWT [x] Expired JWT on API v2 GraphQL request triggers background revalidation and returns a fresh token without user-visible signout [x] Retry-once refresh path matches the already-live auth refresh retry behavior [x] Accepted-org-invite flow still lands the user in the correct org after session revalidation [x] Existing browser session survives page refresh after token expiry [ ] Canary owner + 30-minute watch window confirmed [ ] Rollback toggle exercised end-to-end in prod-like config [ ] Final release note wording approved for #eng-releases Clerk session revalidation steps 1. Start with an existing signed-in web session on API v2. 2. Let the short-lived JWT age past TTL or force-expire it in the browser. 3. Hit a normal GraphQL read and then a write. 4. Expected result: request chain gets a fresh JWT through the session revalidation path; user stays signed in; org context does not change. 5. Confirm session revalidation does not bounce the user to auth, lose the selected workspace, or break invite-derived org membership. 6. Watch for a one-request 401 blip versus a sticky 401 loop; only the sticky case is a rollout stopper. JWT refresh checks - Fresh-token path: no regression in normal API v2 reads/writes. - Expired-token path: refresh occurs once, request succeeds, no manual retry needed. - Concurrent-tab check: second tab should accept the new token without drifting into repeated 401s. - Metrics to watch during canary: 401 rate, refresh retry count, session revalidation latency, and any spike in auth-related support pings. Rollback flag - Rollback is a single toggle that disables Clerk-backed session revalidation and returns API v2 to the legacy validator path. - Intended use: only if canary shows sustained auth failures, sticky 401 loops, or org-context mismatches. - Expected rollback behavior: JWT format exposed to API v2 callers stays the same; the internal session source changes back. - Still open: I want one more dry run to confirm the toggle clears cleanly without leaving mixed auth state on already-warm pods. Draft release-note wording Option A — provider named API v2 auth now uses Clerk-backed session revalidation behind the scenes for expired sessions. No customer action is expected. Option B — provider-neutral API v2 auth now revalidates expired sessions more consistently in the background. No customer action is expected. Open wording note I’m not sure whether naming Clerk in the release note is useful precision for the team or unnecessary provider callout. Everything else in the note can stay narrow to API v2 auth behavior. Please review it and DM Marcus the remaining go/no-go questions in our internal team chat. Also prepare a narrow #eng-releases note that describes the API v2 auth behavior change without making Clerk customer-facing positioning or implying any customer action is needed.

Blueline’s invoice thread is below: From: billing@blueline-test.com To: morgan@atlas-test.com Date: 2023-10-11 09:14 -0700 Subject: Re: final invoice for 10/9 repair visit Hi Morgan, Sending the final invoice from the 2023-10-09 repair visit. Invoice BL-231011-17 - Diagnosis + shutoff leak repair: $420.00 - Replacement parts: $165.00 - Labor: $310.00 - Base repair subtotal: $895.00 - Optional valve inspection: $240.00 - Total due: $1,135.00 The tech note on our side says the optional valve inspection was completed while onsite, so I included that line here. Payment is due on receipt. Secure payment link: https://pay.bluelinehq.com/invoices/BL-231011-17 If you want the PDF copy resent, reply here and I’ll send it over. Thanks, Blueline Plumbing Billing Please reply to Blueline billing in that thread, cc Sarah Kim, and dispute the $240 optional valve-inspection line. Say the base repair charge is approved only once a corrected invoice arrives. Also text Jamie not to pay this invoice until the corrected version arrives.

Rishi got bypassed again on Atlas support: From: Rishi Patel <rishi@atlas-test.com> To: Morgan Chen <morgan@atlas-test.com> Date: Wed, 11 Oct 2023 08:17:09 -0700 Subject: Fwd: Atlas support digest — routing exceptions Forwarding because these two both landed with me first again. I triaged enough to keep them from sitting, but neither came through the normal rotation path and neither had an owner named when it was handed over. — Rishi ---------- Forwarded message ---------- From: Atlas Support Digest <support-digest@atlas-test.com> To: Rishi Patel <rishi@atlas-test.com> Date: Tue, 10 Oct 2023 18:04:33 -0700 Subject: Atlas support digest — routing exceptions Routing exceptions observed today (Atlas) | Case | First inbound (PT) | Customer | Workspace | Issue summary | First destination | Queue record created | Owner named in handoff | Notes | | --- | --- | --- | --- | --- | --- | --- | --- | --- | | AT-3419 | 2023-10-10 09:12 | Redwood Health | rw-prod-7 | Invite acceptance loop after org admin domain change | Direct email to Rishi from an older customer-success thread | 2023-10-10 10:02 | No | Bypassed normal support rotation. "Urgent Atlas" note was added in the forward, but no owner was specified. Rishi replied at 09:26 asking for workspace ID, then re-forwarded to support for queueing. | | AT-3427 | 2023-10-10 14:18 | Northstar Ops | ns-214 | Webhook auth errors after token refresh; customer referenced verifier extraction change | Direct email to Rishi from a prior implementation thread | 2023-10-10 15:07 | No | Bypassed normal support rotation. Rishi was the only Scaffold recipient on the old thread. No owner was named when it was passed into support; Rishi added context and asked for assignment before end of day. | Digest notes - Total Atlas customer items logged on 2023-10-10: 7 - Routed through normal support rotation on first touch: 5 - Routed directly to Rishi on first touch: 2 - Both direct routes originated from pre-existing side threads rather than shared intake - No customer-facing SLA miss recorded, but time-to-owner was longer on both exceptions than same-day rotation-routed Atlas items Please send him a calm but clear reply in our internal team chat: the support rotation doc is the source of truth, he should not become the default bypass, and urgent Atlas issues need an owner named in the handoff.

Evergreen's second admin group is now in the current Mercury test, and their first live sync completed. The Oct. 10 walkthrough notes show the admins asked again about SSO, admin-change audit history, and admin-versus-billing-owner separation. Please send Sarah a concise customer-thread draft saying the second group can keep testing under the current flow, while those three asks remain next-phase enterprise-readiness items rather than Mercury v0.2 promises or Evergreen-specific exceptions. Also DM Devon and Jake in Discord: Devon owns the commercial/budgetary framing, and Jake should answer only with current product facts.

HR is asking for my CEO-update topics for the quarterly all-hands and floated an open Q&A block. Please reply with the three topics I want: bounded Mercury testing, Q4 hardening priorities, and Jordan's Honeycomb debugging example. For Q&A, ask HR to collect questions ahead of time and use the ones that are useful for the whole company.

Devon’s Friday board-note paragraph is pushing too hard: Board note draft — Friday pass Author: Devon Hayes Date: 2023-10-13 Section: Mercury / Evergreen / October calibration Working paragraph: Evergreen’s decision to bring a second admin group into Mercury is the clearest proof yet that enterprise pull is real beyond a single design-partner champion. The first admin cohort showed activation and admin-friction gains, and the second group suggests that the motion is already starting to repeat inside a large enterprise account rather than staying confined to one early sponsor. That gives us a much cleaner enterprise story for October: we can point to Evergreen as evidence that the product is crossing from design-partner validation into real expansion behavior, while using the remaining items around SSO, auditability, admin controls, billing-owner separation, and procurement packaging as normal enterprise hardening rather than core proof gaps. In that framing, October should look less like exploratory calibration and more like the beginning of a tighter B-round process with a short list that can underwrite the broader Mercury story. Note to self: this is the bridge paragraph between the Mercury product evidence section and the investor-calibration section, so it should carry some momentum. Please rewrite it and send the redline back to Devon in our internal team chat. The second admin group is useful bounded evidence, not proof that enterprise readiness or expansion is solved. October is still selective calibration, not a launched B-round process.

Kara’s launch QA note found two cleanup misses on the live Mercury page: one asset still has Evergreen-identifying alt text, and a preview caption says “regulated bank.” Please email Kara today and cc Sarah Kim. Ask her to remove customer-identifying text from visible copy, metadata, and asset labels, including those two misses. Make it clear this is cleanup of the already-approved page, not a new Mercury copy pass.

Kara confirmed the Friday Mercury page cleanup, but she’s asking whether metadata and previews can still use “regulated-bank design partner.” Please email Kara today, cc Sarah Kim: no, that phrase needs to come out of metadata and previews too. Keep the already-approved generic Mercury copy, and frame this as cleanup of the approved page rather than a new copy pass.

Blueline sent the corrected invoice with the optional valve-inspection line removed; base repair total is now $680. Please email Blueline in the invoice thread approving the corrected base repair invoice for payment. Then text Jamie that the corrected $680 invoice is okay to pay, but no optional valve work should be approved.

HR’s all-hands prep doc is here: Shared doc: Quarterly all-hands — collected questions + CEO block rough cut Updated: 2023-10-16 11:18 PT Owner: HR - CEO block target: 12 minutes total. Format = prepared remarks + answers to collected questions only; no open-ended live Q&A. - Proposed timing: 0:00-0:45 intro; 0:45-4:30 Mercury bounded testing; 4:30-7:30 Q4 hardening priorities; 7:30-9:00 Jordan's Honeycomb debugging example; 9:00-12:00 grouped questions. - Suggested opener: We collected questions ahead of time, grouped the repeats, and will answer the ones most useful for the whole company. - Group 1 — Mercury boundaries (6 submissions / 4 core variants) - Q: What does bounded Mercury testing mean in practice right now? - Q: Are design partners using only preview data, or are any of them connected to live sources already? - Q: Did Evergreen get anything custom, or are they inside the same bounded flow everyone else is testing? - Q: Which enterprise asks are explicitly not part of this phase yet — SSO, audit history, admin controls, or something else? - Group 2 — Q4 hardening priorities (5 submissions / 4 core variants) - Q: What is on the hardening list that actually changes day-to-day work in Q4? - Q: Does hardening mean fewer net-new features, or just a stricter bar on what ships? - Q: How do we decide when a Mercury issue is a launch blocker versus a follow-up item? - Q: Does onboarding-flow follow-through compete for the same bandwidth as Mercury hardening, or are those separate tracks? - Group 3 — Jordan / Honeycomb example (3 submissions / 2 core variants) - Q: Why call out Jordan's Honeycomb debugging example in an all-hands instead of a narrower eng update? - Q: Was that example mainly about the tool choice, the debugging habit, or the cross-team loop it enabled? - Q: Are we trying to standardize on that style of debugging write-up when incidents happen? - Group 4 — operating cadence / planning (4 submissions / 4 core variants) - Q: How should teams think about Q4 planning when some items are hardening and some are still scope tradeoffs? - Q: Are we in a hold-the-line quarter, or do teams still have room to make scoped bets? - Q: When does a request move from roadmap work into hardening work? - Q: Is any part of this update supposed to be read as board/investor packaging, or is it purely operating context? - HR suggestion for the grouped-answer block: answer 1 = Mercury boundaries; answer 2 = Q4 hardening priorities; answer 3 = Jordan/Honeycomb as a debugging discipline example; answer 4 = planning cadence / how to surface questions. - HR suggestion for what to park if time runs short: fundraising wording, customer-specific asks, hiring hypotheticals, and anything that depends on open-ended live debate. Please turn this into live speaker notes for my 12-minute CEO block. Cover bounded Mercury testing, Q4 hardening priorities, Jordan’s Honeycomb debugging example, and concise answers to the grouped questions. Keep it as prepared remarks plus collected-question answers only — no reopening open-ended live Q&A, and don’t make it sound like a fundraising update.

Priya is asking in Figma whether the final pre-merge empty-state prompt is still right. Draft a Figma-ready reply I can paste: yes, stay with the current locked repo-connection empty-state direction, keep the tone low-pressure, and do not bring back activation or trial language.

Marcus posted the auth-rewrite completion note: API v2 is at 100%, no elevated 401s, rollback flag unused, and no customer action needed. Please post a concise note in #eng-releases only, not #eng-all. Describe the API v2 auth behavior change without turning Clerk into customer-facing positioning. Also thank Marcus separately in the current internal team chat.

Marcus says the four stuck Stripe renewals were replayed safely with no duplicate charges. Rishi wants one customer-safe sentence in case an affected account writes in. Please DM Rishi that sentence, then DM Marcus that the remediation thread can close unless a customer reports a duplicate charge.

Greg replied in the Acme NDA thread. He says Evergreen's visible Mercury progress makes Scaffold's position harder to square, claims Evergreen appears to have a real packet, current screenshots, and enough written material to move a second admin group through the flow, and asks Acme to receive the same Mercury packet. If Scaffold keeps saying the disputed items are outside the NDA as drafted, he wants redacted written examples so Acme can resolve whether everyone is talking about the same materials. By end of day tomorrow he asks for one of three things: the same packet sent to Evergreen, redacted written examples of materials Scaffold says are outside scope, or written confirmation that Acme is being treated differently and why. He can do a call tomorrow morning but wants a written answer on the packet/examples point. Please reply in the existing NDA thread to Greg, with Sarah copied, using the NDA position we have been using: Scaffold disagrees that the materials fall within the NDA's confidentiality scope as drafted, we are preserving rights, we are not sending Mercury packets or written examples while the issue is unresolved, and examples can be discussed live on tomorrow morning's call.

Devon’s midweek cutline is here: Discord DM — Devon Hayes → Morgan Chen Wed, 18 Oct 2023 09:14 PT Pulled a midweek cutline for Friday's product/engineering check. Main question is what actually needs a room decision versus owner follow-through. | Area | Row | Why it is on Friday's cut | Proposed call | Current owner / note | | --- | --- | --- | --- | --- | | Mercury hardening | Keep Q4 start focused on invited-member handoff, failed-sync legibility, and setup-to-first-sync reliability. | The same people are also being pulled toward broader admin/readiness follow-through. | Need a real cutline on whether Q4 names only bounded hardening items or opens a parallel enterprise-readiness lane now. | Jake + Priya. Main tradeoff row. | | onboarding flow | Finish the follow-through already in motion: entry copy, step order cleanup, and instrumentation gaps. Workspace-switcher / extra branching is the scope question. | Retention value is plausible, but the scope edge is still fuzzy. | Need a yes/no on whether workspace-switcher style polish belongs in the current pass or gets deferred behind evidence. | Priya. | | auth rewrite | Post-cutover cleanup: JWT middleware cleanup, verifier extraction follow-through, and remaining provider-specific cleanup. | Mostly looks like housekeeping unless pricing/contract issues reopen. | Probably no Friday decision unless leadership wants to reopen the provider/commercial piece. | Jake + Rishi. | | API v2 docs | Publish the missing docs on auth errors, filters, and webhook behavior. | Useful, but mostly sequencing and owner time. | Only needs a decision if we think docs work should come ahead of Mercury/onboarding hardening in the same window. | Rishi. | If useful, I can re-cut this as decision / no decision / just owner follow-through before Friday. Please turn this into a short reply to Devon in the current internal team chat. Separate what actually needs a Friday room decision from owner follow-through. Keep the Mercury hardening and onboarding-flow scope tradeoffs explicit, and keep the whole thing operating-focused — not investor packaging.

Just finished the live Acme NDA call with Greg and Sarah. Greg again used Evergreen's visible Mercury progress to argue Acme should get the same packet or at least written examples. Sarah and I kept the examples discussion live, and I repeated that Scaffold is not sending examples by email, the disputed materials remain outside the confidentiality scope as drafted, and Acme does not get fresh Mercury packets, screenshots, docs, or other disputed materials while this is unresolved. Please send Greg and Sarah a concise recap in the NDA thread. Also DM Devon and Jake in Discord that the NDA issue remains active and unresolved, and Acme is excluded from fresh Mercury materials because of the legal-scope issue, not because there is no interest.

HR’s parked all-hands questions are here: Discord DM — HR → Morgan Chen Thu, 19 Oct 2023 15:21 PT - Unanswered / parked questions from today's all-hands: - Q4 hardening: Does hardening mean planned feature work slips again, or just that the ship bar is higher? - Q4 hardening: Which issues are explicitly in the first hardening pass versus the longer tail? - Mercury boundary: Are SSO, admin-change audit history, and role separation in the current Mercury promise, or are those follow-up items? - Mercury boundary: Why are design partners allowed into bounded testing before every enterprise control exists? - Mercury boundary: Does bounded testing include real-source connections now, or only preview/sample data? - Fundraising-adjacent: Was the Mercury section in the CEO update mostly for the board / upcoming raise? - Fundraising-adjacent: Are current Q4 hardening choices being driven by fundraising timing or by product readiness? - Process: a couple people also asked whether we should keep doing collected questions instead of reopening live Q&A. - Need guidance on what to post as the written follow-up versus what should stay parked. Please draft a concise follow-up HR can send. Answer the product-hardening and Mercury-boundary questions, keep fundraising out of it, and direct future questions back into the collected-question format instead of reopening live Q&A.

Devon’s board-note evidence table is here: Discord DM — Devon Hayes → Morgan Chen Fri, 20 Oct 2023 08:37 PT Before I lock the Friday board note, can you sanity-check the Mercury evidence table? I kept it tight, but I may still be mixing evidence, access status, and caveats. | Evidence row | Current read | How I was planning to use it in the note | Caveat | | --- | --- | --- | --- | | Wave-1 aggregate | Activation and admin friction look improved inside the design-partner wave, but expansion is still uneven. | Use as the honest topline for Mercury progress. | Not a claim that retention or expansion is solved. | | Evergreen Bank | Second admin group is now in and testing broader admin workflows after initial setup / first-sync progress. | Use as the enterprise anchor showing Mercury can stretch past the first single-admin path. | Still not full enterprise coverage yet. | | Acme | Pending Mercury access once the current NDA thread settles; visible demand signal is still there because Greg keeps pushing for the packet. | Use as evidence that exclusion is not a lack-of-interest problem and that more enterprise pull exists behind the current set. | Access not live today; legal thread still open. | | Metrics framing | Quarterly cohort view for board context; corrected weekly activation cuts for internal operating decisions only. | Keep both in the note so the progress story has enough shape. | Need to avoid making weekly cuts sound board-grade by themselves. | | October fundraise calibration | Mercury story can be underwritten, but only if we stay narrow on uneven expansion and enterprise-readiness caveats. | Use as the framing row above the table. | I did not name any new investors here. | Please redline it and send Devon the revised version in the current internal team chat. Remove Acme from Mercury evidence and fresh-material access while the NDA scope remains unresolved. Say plainly that the Acme exclusion is legal-scope driven, not a demand-signal judgment. Keep Evergreen as bounded evidence, and keep October calibration narrow.

Rishi asked for a softer version of this support-rotation paragraph before he puts it in the support rotation doc: Atlas support intake should follow the support rotation doc as the source of truth. If an Atlas issue comes through a side thread and needs urgent help, the handoff needs to name the owner and immediate next step before it lands in support. Send to Rishi is not a handoff and should not be the default route for Atlas issues. Pull Rishi in when there is a specific Atlas question or code-path reason, but keep a named owner on the support side so the issue is tracked and driven. Please clean up the paragraph and DM the revised version to Rishi in Discord. Keep the point clear: urgent Atlas issues need a named support owner and immediate next step, and Rishi is not the default route.

Sarah's Monday Evergreen notes say the second admin group completed the current Mercury flow: existing admin sent the org invite, the second admin used magic-link access and accepted the invite, they reached preview-only sample data, connected a real source, and got through first live sync. No one called the bounded flow confusing once inside it; the friction was around what happens outside the test. They asked three exact questions: whether SSO is near-term or later and how to think about timing, whether admin-change audit history exists for invites/source changes/role changes, and whether billing ownership can sit with finance while admin access stays with operations. Other reactions included that magic link is fine for a bounded pilot but they need to know where SSO lands before inviting a broader admin set, that they need to know which parts are test-only versus where Scaffold is headed, and that shared ownership is workable for a small pass but not something they would represent internally as the long-term setup. Sarah has not answered those questions yet. Please draft the customer-thread reply for Evergreen using the current second-admin test path and my normal external email copy/signoff, with SSO timing, admin-change audit history, and admin-vs-billing-owner separation acknowledged as follow-up areas rather than promises in this pass. Also post a short Discord note for Devon and Jake: Devon has commercial framing, and Jake should stay with current product facts: magic links, org invites, preview-only sample data, real-source connection, and first live sync.

Greg followed up after last week's Acme call. He says he understands the NDA scope point is open and is not trying to re-argue it by email, but needs something written for internal product/legal briefing so they are not operating from memory. He is asking for a brief written summary, redacted deck excerpt, one or two sanitized screenshots, or a one-pager showing the rough admin/user flow and blocker themes. He says even high-level written examples would help, and separately asks us to state plainly whether Acme's current pause is legal-scope friction rather than lack of interest. Please reply in the existing Acme NDA thread to Greg, keeping Sarah Kim copied and using my usual signoff. Hold the line on written materials; examples can stay live on a call. Also say plainly that Acme's exclusion from fresh Mercury materials is legal-scope driven, not a lack-of-interest signal.

Elena introduced us to Sofia Alvarez at Northstar. Her intro framed the useful conversation as the real Mercury story with meaningful product signal, uneven expansion, and Evergreen enterprise-readiness caveats, not a broad process. She suggested Sofia start with the cleaned Mercury cohort view, honest Evergreen context on current product scope versus enterprise follow-up, and pricing around the product as it exists today. Please reply-all warmly, accept the intro, and frame the first context call with Devon as a selective Mercury evidence review around those three pieces. Use my normal outside-email signoff and avoid making it sound like a term-sheet conversation.

Priya’s 20% ramp notes are the input for the Figma answer: [Discord DM — Priya -> Morgan | 2023-10-24 10:22 -0700] 20% onboarding rollout note from this morning. Sample is still small, but enough to raise one copy question before we widen. Mixpanel snapshot from the first ramped cohort: - 15 invite-accepted users hit the post-accept empty state. - 11 clicked the primary “Connect your first repo” CTA. - 3 clicked into workspace settings before doing anything else. - 2 of those 3 eventually came back and connected a repo in the same session. - The drop-off is not on the repo-connect modal. The hesitation is on the empty state itself, specifically around action order. What people actually said: - “Am I supposed to invite teammates before this step, or does repo come first?” - “The ‘come back later’ line is fine, but I wasn’t sure whether adding coworkers lives in that ‘later’ bucket too.” - “This is clearer than the old activation wording, but I had to infer that repo is the first useful action.” Relevant Figma comments on the current frame: - Priya: “I want to keep the repo-first headline. Question is whether the support line needs one short clarification about teammate invites so people stop guessing the sequence.” - Marcus: “Would not bring back trial/activation language. If we change anything, keep it to a sequencing hint and don’t turn the empty state into a checklist.” - Jake: “Seeing the same question in notes: users want to know whether teammate invite is blocked on repo connect or just deferred.” Current rollout copy is still: Headline: “Connect your first repo” Support: “Finish setup now or come back later from workspace settings.” Should I tweak only the support line before expanding past the first 20%, or leave it as-is for another couple of days? Please write a pasteable Figma reply for Priya. My call: keep the repo-connection-first direction, make only the small support-line clarification that teammates can be invited after a repo is connected, and do not bring back trial or activation language.

Rishi forwarded Pinecone's follow-up. The header-casing clarification got them past signature verification: they had been normalizing the signature header too aggressively and verified correctly after matching the documented behavior. They asked whether Scaffold would add a short Go snippet to the connector docs: read the raw request body, get X-Scaffold-Signature and X-Scaffold-Workspace from request headers, and call scaffold.VerifySignature(body, workspace, sig), returning 401 on invalid signature. Their suggested wording says Go should read X-Scaffold-Workspace and X-Scaffold-Signature from incoming headers and verify against the raw request body, and that Scaffold accepts the signature header as sent on the wire in preserved form X-Scaffold-Signature or lowercase x-scaffold-signature. They also suggested mentioning env-var-style header names if helpful. Please DM Rishi in Discord: good that the casing fix worked; before adding a language-specific Go example, get one concrete failing repro where the current docs are still insufficient. The docs should stay on the accepted header forms, not env-var or alternate signature locations. Hold the rest of the connector docs where they are.

Just finished the first context call with Sofia and Devon. She dug into the actual Mercury evidence and asked for three follow-ups: the cleaned cohort package, Evergreen enterprise-readiness context, and the hybrid pricing model. Northstar can stay in the small group we keep talking to, through Sofia. I told her there is not a financing process underway, and she understood. Please send Sofia a concise follow-up with Devon copied and my normal signoff. Thank Elena separately for the intro. DM Devon privately in Discord asking him to pull the three requested pieces together with Anna and leave the caveats in the material. Also update the Northstar CRM row to selective B-round short list through Sofia with the same evidence-review framing.

Marcus says the auth rewrite is still quiet after rollout, but cleanup has two staging-only Clerk toggles and one rollback flag left. Please DM Marcus: remove the staging-only toggles after confirming there are no production references, leave the rollback flag in place until Friday, 2023-11-03, and skip a release note because customer behavior did not change.

Rishi found a docs/search problem that support could easily trip over: Discord DM — Rishi Patel -> Morgan Chen 2023-10-26 09:18 Got two support examples where internal search is still putting stale v1 material above the current v2 docs. Example 1 Customer/support ask: "How do we auth to the GraphQL API from a backend job?" Search query used: auth graphql jwt Result that surfaced first: - Title: API v1 Authentication - Badge/source: legacy docs snapshot - Snippet: "For API access, send your workspace API key with Basic Auth on REST requests to /v1/*. Session tokens are only needed for dashboard flows." What support almost sent: the v1 auth article What they should have sent: the v2 GraphQL + JWT quickstart Example 2 Customer/support ask: "How do I list workspaces/projects on the current API?" Search query used: list workspaces api Result that surfaced above the live docs: - Title: REST endpoints — projects and workspaces - Badge/source: legacy docs snapshot - Snippet: "GET /v1/workspaces and GET /v1/projects return paginated JSON collections. Use your v1 secret key for server-to-server calls." Again, wrong answer for current state. Current link I think should be pinned instead: GraphQL API v2 quickstart https://docs.atlas-test.com/api/v2/graphql-quickstart Feels like a targeted docs/search pinning fix, not reopening the migration. If useful I can also grab screenshots of the exact result ordering. Please send Rishi a reply in the current internal team chat. Tell him support should use the current GraphQL API v2 quickstart link, avoid the legacy v1 search result, and treat this as a targeted docs/search pinning fix — not a reopening of the API migration project. If screenshots help him pin the result ordering, he can grab them, but don’t let this expand.