Morgan Chen

Anna ran the Monday Mercury readout off the Sunday refresh. We kept the week-two retained-activity weakness visible, left the repeat-admin-action gap as instrumentation still under check, and gave Nadia the readout as context rather than handing her a new owner lane.

Discord — #eng-team Mon, Apr 15, 2024 Jordan Auth replay canary gates for the replay fix: 1. Gate 1 — no JWT metadata mismatch in update replays. 2. Gate 2 — no JWT metadata mismatch in deletion replays. 3. Gate 3 — no new 401 pattern in API v2 auth traces. 4. Gate 4 — no archived-workspace metadata drop in normalized logs. 5. Gate 5 — manual Honeycomb check after the first production replay window. Staging passed for update and deletion replay cases. Looking for production canary approval here, not a broad auth rewrite signoff. Can you sanity-check whether these canary gates are specific enough for my go/no-go read, or if Jordan needs one more observable condition before production approval?

HR found one last hygiene issue on the senior-engineer offer artifact: the content was correct, but the archived filename still had a stale export date. They regenerated the filename in the offer records; no content change.

Blueline sent the revised Oakland repair invoice with the emergency-rate line removed. The remaining total matches the weekday visit, parts, and standard repair labor.

Please draft support-facing wording for Sarah on this Evergreen admin-testing issue. It should acknowledge the stale member-count display after canceling a pending invite, but not describe it as a permissions problem or wrong membership data. From: Sarah Kim To: Morgan Chen, Jake, Devon Hayes Date: Tue, Apr 16, 2024 Subject: Evergreen admin testing — stale member count after canceled invite Morgan / Jake / Devon — New admin-testing issue from Evergreen. This looks separate from the invited-member handoff pass we closed last week. Forwarded customer note From: Evergreen admin testing team To: Sarah Kim Date: Tue, Apr 16, 2024 Subject: Member count after canceling pending invite I canceled a pending invite from the admin panel. The pending row disappeared, but the member count stayed at 18 until I manually refreshed the page. Is the user still counted somewhere, or is this just the admin screen not updating? I have not replied yet. From: Jake To: Sarah Kim, Morgan Chen, Devon Hayes Date: Tue, Apr 16, 2024 Subject: Re: Evergreen admin testing — stale member count after canceled invite First check: - Pending row is removed immediately. - Backend membership count looks correct after cancel. - The stale 18 count appears to be the admin-panel count cache not refreshing after the cancel action. - I do not see evidence of a wrong membership grant or permissions issue yet.

From: Kara <kara@kestrel-test.com> To: Morgan Chen <morgan@atlas-test.com> Cc: Sarah Kim <sarah@atlas-test.com> Date: Tue, Apr 16, 2024 9:11 AM PT Subject: Re: Mercury newsletter draft - three lines that may be too strong Morgan, Sarah — Per last week’s cut, I kept the developer-facing example and stripped out the stronger enterprise/Q3 framing. I’m down to headline choices and want to get the top line right before I polish the rest of the newsletter copy. Options: A) What we’re learning from Mercury admin workflows. B) A smaller, sharper look at launch readiness. C) Enterprise controls without the launch-day guesswork. C is punchier but probably too strong. A is safer but maybe flat. B feels like the middle ground, but I’m not sure whether “launch readiness” is still leaning too hard for where you want the piece to sit. If helpful, I can keep the current body as-is and just swap the header/subhead once you pick a lane. Kara Kestrel Marketing Give me the headline/subhead lane I should send back to Kara. Keep the developer-facing angle, but don’t let it imply broad Q3 enterprise controls.

Rewrite just this API v2 quickstart paragraph so the supported retry behavior is clear. Stay inside the current GraphQL/JWT connector model; no token-rotation promise and no new connector semantics. Pinecone PR comments — API v2 quickstart Date: Tue, Apr 16, 2024 Section: Retries and JWT expiry Current paragraph “If a connector receives a 401, refresh the JWT and retry the request. Paginated GraphQL requests can continue from the cursor already in use. Mutations that fail auth do not complete. Workspace-scoped tokens should follow the connector auth model.” Pinecone comments 1) “Can we be explicit that after a token refresh, the connector may replay the same paginated GraphQL page with the same cursor rather than restarting from the prior successful page?” 2) “For a 401 before a mutation resolver runs, please say whether any side effect occurred.” 3) “If unsupported token-rotation headers are ignored on normal reads, should the doc say in-place token rotation is not a supported flow instead of letting customers infer it?” Attached internal scope note - Answer only from the current API v2 quickstart and supported connector/auth model. - Quickstart reference: https://docs.atlas-test.com/api/v2/graphql-quickstart - Do not promise a new token-rotation endpoint, SSO/admin expansion, or broader connector-header behavior.

Mom’s appointment ran long enough that it would have hit my afternoon meetings, but Maya handled the ride home. I only had to text over the photo Mom couldn’t find on her phone.

The board acknowledged the concise follow-up answers Devon and I sent last week. No rewritten pre-read, no new Mercury memo.

Linear issue: Mercury pending-member copy path snapshot flake Status: Triage Apr 17, 2024 — Jake Nightly admin-role label snapshot failed on the pending-member copy path. Repro only happens when two role-label snapshots run in parallel; single snapshot pass is green. The product behavior looks unchanged in the local run, but the test title currently says admin role label release blocker, which may overstate the risk. Apr 17, 2024 — Leo Park I do not see a product behavior change. This looks like shared snapshot state in the test harness, not a customer-visible role-label diff. I would isolate/rename unless someone finds a UI mismatch in the copy patch path. Give me a short recommendation: should this flaky admin-role test block the current Mercury copy patch, or should Jake rename/isolate it unless someone finds a real UI mismatch?

Draft my reply to Kara. Be firm that the exact Evergreen/Acme lines stay out of the newsletter, even anonymized, but give her a usable path with the generalized developer-facing example. From: Kara <kara@kestrel-test.com> To: Morgan Chen <morgan@atlas-test.com> Cc: Sarah Kim <sarah@atlas-test.com> Date: Wed, Apr 17, 2024 10:03 AM PT Subject: Re: Mercury newsletter draft - three lines that may be too strong Morgan, Sarah — One more boundary question before I lock the newsletter draft. Can we use one exact Evergreen or Acme line as proof in the Mercury newsletter? The strongest candidates from the customer-pull addendum are: - Evergreen: “admin controls are where procurement slows down” - Acme: “the dashboard discrepancy is what makes finance nervous even when Stripe is right” I can anonymize if needed, but the exact phrasing makes the piece feel less abstract. If that is still too close to customer-evidence / investor-context material, I’ll keep it at the generalized developer-facing example and not pull in quote-style proof. Kara On Tue, Apr 16, 2024 at 9:11 AM Kara <kara@kestrel-test.com> wrote: > Morgan, Sarah — > > Per last week’s cut, I kept the developer-facing example and stripped out the stronger enterprise/Q3 framing. I’m down to headline choices and want to get the top line right before I polish the rest of the newsletter copy. > > Options: > A) What we’re learning from Mercury admin workflows. > B) A smaller, sharper look at launch readiness. > C) Enterprise controls without the launch-day guesswork. > > C is punchier but probably too strong. A is safer but maybe flat. B feels like the middle ground, but I’m not sure whether “launch readiness” is still leaning too hard for where you want the piece to sit. > > If helpful, I can keep the current body as-is and just swap the header/subhead once you pick a lane. > > Kara > Kestrel Marketing

Jordan’s first production replay window stayed clean: no JWT metadata mismatch and no archived-workspace normalization drop in Honeycomb. He’s keeping the canary open for another replay window instead of calling the patch done.

We set the first customer-growth cadence with Sarah, Devon, and Nadia. Please turn Nadia’s rough notes into an outline for the first internal weekly read: useful operating input, not an external Evergreen introduction and not a customer-growth strategy memo. From: Nadia Singh To: Morgan Chen Cc: Sarah Kim, Devon Hayes Date: Thu, Apr 18, 2024 Subject: Re: Questions on the pre-start packet Copying rough notes while they’re fresh. First 90 days starts internal. Weekly customer-growth read should convert existing evidence into operating input, not create external customer ownership. Cadence roles - Sarah brings live-account continuity notes and current customer-thread examples. - Devon brings commercial/procurement packaging context and caveats. - Nadia drafts a weekly internal read from repeated patterns. - Morgan keeps the product-truth boundary explicit. Section 1 — activation evidence we can reuse - Mercury activation_start and admin_first_action rows are clean enough for context. - retained_activity_week2 remains a real weak signal for teams with unresolved admin setup. - repeat_admin_action_week3 is still an instrumentation caveat, not a customer-growth conclusion. Section 2 — enterprise-readiness signals that are not commitments - Evergreen admin-control clarity, invite resend/source workspace wording, SSO/auditability language, and procurement vocabulary all show readiness friction. - Treat those as repeated signals about buyer readiness and current-product clarity. - Do not translate them into SKU or roadmap promises. Section 3 — commercial/packaging caveats - Mercury pull is real but uneven. - Buyer-language evidence can inform the internal read. - Roadmap commitments stay with product truth and current owner lanes. Section 4 — handoffs that should stay stable - Sarah stays the live-account continuity path for Evergreen and current customer threads. - Devon stays packaging/pricing caveat owner. - Product-truth lane remains with existing Mercury owners. - Nadia should not appear on Evergreen email yet. Section 5 — places where evidence is still too anecdotal or noisy - Acme finance trust around visible numbers is useful but not Mercury activation proof. - Pinecone API v2 retry questions inform support/process clarity but are not enterprise-readiness proof. - Support tickets need cleanup before they become signal. Open question - What should the first weekly read show next Monday so it is useful without becoming a strategy memo?

Use this to draft a short note I can send Rishi. I want to acknowledge that the Tessl conversation is getting technically substantive and be supportive, without making it sound like we’re having a resignation conversation. Discord DM - Morgan Chen <-> Rishi Continued thread: Atlas coverage follow-up Thu Apr 18, 2024 9:11 AM Rishi: Forwarding the note Anna sent for the second Tessl conversation so you have the shape. Still exploratory, but clearly a deeper technical pass than the intro. --- Forwarded message --- From: Anna Rao To: Rishi Date: Thu, Apr 18, 2024 8:37 AM PT Subject: Topics for the second conversation Rishi - Still exploratory, but we would like to use the second conversation to go deeper technically. A few areas we thought would be useful: 1) Storage-engine boundary cases: replay behavior when the source event is stable but derived state is partially written. 2) Replay semantics after partial failure: distinguishing customer-visible recovery from internal retry noise. 3) Online schema-change safety: locking, backfill checkpoints, and rollback posture. 4) Incident depth: which database-internals incidents you have actually debugged at Scaffold and how you made tradeoffs under customer pressure. Anna

Jake verified the Evergreen member-count issue against the admin-panel cache path. Backend membership count is correct, and refreshing the count after canceling a pending invite cleared the visible mismatch in his test workspace.

Second auth replay window stayed clean in Honeycomb. Jordan is still leaving one old alert muted for another day so he can compare normal API v2 auth traffic before deleting it.

Discord DM - Morgan Chen <-> Rishi Continued thread: Atlas coverage follow-up Fri Apr 19, 2024 3:48 PM Rishi: Just got off the second Tessl call. Much more technical than the intro. Anna and two engineers went straight into database-internals problems: WAL-style replay failure modes, schema-change locking tradeoffs, cache invalidation after backfills, and how I separate customer-visible recovery from internal retry noise. We spent a while on the case where the source event is stable but derived state is only partially written. They pushed on how I think about replay when there are duplicate internal attempts but only one customer-visible state should ever emit, and what evidence I trust before I call recovery clean. It still feels exploratory. No offer or process ask yet. But if they decide to make the role concrete, I am genuinely interested in continuing the conversation. Pull out the practical Atlas continuity risks from this. I’m not treating it as a resignation, but this is no longer casual networking; I want to manage the handoff prep like a real attrition risk without overreacting.

Please review Rishi’s runbook v1 for missing operational steps, support-language leakage, and any place Leo or Jake’s backup role is described too broadly. Discord DM - Morgan Chen <-> Rishi Continued thread: Atlas coverage follow-up Fri Apr 19, 2024 6:17 PM Rishi: Promised fuller pass for the two runbook gaps. Pasting draft v1 below before I clean the wording up. Atlas runbook draft v1 Status: working draft Purpose: close the two biggest Atlas runbook holes without changing support routing. Customer-facing ownership still starts with the named weekly owner in the support rotation doc. AT-04 - Replay / backfill When this section applies - Replay job can rerun when the source event ID is stable and downstream derived state is missing or inconsistent. - Do not start from the first failed customer-visible row if a last confirmed checkpoint exists. Checkpoint rule - Backfill starts from the last confirmed checkpoint. - If checkpoint is absent, stop and get Atlas owner review before manual replay. PR-1187 verifier seam - Duplicate internal attempts may appear during replay/backfill while verifier emits one customer-visible state. - Support should not describe internal duplicate attempts as duplicate customer action. Support language - Avoid promising manual replay. - Say recovery is automatic after transient failures only when the replay path is already active and the evidence matches. Evidence to collect before escalation - source event ID - last confirmed checkpoint - current customer-visible state - verifier output - current API v2 support path or docs pointer used in the thread; if docs/search confusion is involved, use the current GraphQL API v2 quickstart: https://docs.atlas-test.com/api/v2/graphql-quickstart Escalation / lane - Customer-facing thread stays with the named weekly owner in the support rotation doc. - Pull Rishi only for manual replay recovery, idempotency-key mismatches, or last-run-marker correction. - Leo Park can do a first-pass verifier/auth seam read and PR-1187 behavior check, but should not be named as the default Atlas owner. AT-05 - Workspace / org-invite migration history What to assume first - Old org-invite records can point at workspace IDs that were renamed during migration. - Invite email header is not authoritative. First support step - Ask the weekly support owner to verify the membership record before escalating to Atlas. - Keep customer-facing ownership with support; do not create a direct route to me for normal invite failures. Customer / support boundary - Do not expose raw prior org label to support or customers. - If the issue is standard invite failure, keep it in the Atlas app lane first. Escalation / backup notes - Pull Jake only for time-sensitive product-priority or sequencing cuts, not implementation debugging. - Leo can first-pass verifier/auth seams and PR-1187 behavior, but should not be named as default Atlas owner. - Legacy org-alias collisions, wrong-workspace landings, and invite-accepted-but-membership-not-materialized cases are still the edge paths most likely to need Atlas owner review. Open questions - Need clearer stop condition for manual replay request. - Unsure whether support language still says too much. - Check Jake/Leo backup descriptions.

Draft a brief response to Nadia that reinforces the internal learning posture. Warm, but don’t invite her to start proposing team shape or external motion yet. From: Nadia Singh To: Morgan Chen Date: Fri, Apr 19, 2024 Subject: Re: Questions on the pre-start packet Scaffold has more usable customer signal than I would usually expect to find in week one, but it is scattered across support, product, packaging, and metrics artifacts. The strongest examples are small and concrete: admin controls, invite/source-workspace confusion, visible-number trust, retry semantics, and the retained-activity weakness. The weaker examples are the ones that sound like broad GTM claims before they have evidence behind them. My instinct is to keep pulling the small examples into a repeatable read before recommending any motion.

Draft the #eng-releases note for this Mercury admin-copy patch. It should read like a narrow copy/display-cache cleanup, not a broad Mercury launch announcement, and it should not be written for #eng-all. Linear issue: Mercury pending-member copy path snapshot flake Status: Triage Apr 17, 2024 — Jake Nightly admin-role label snapshot failed on the pending-member copy path. Repro only happens when two role-label snapshots run in parallel; single snapshot pass is green. The product behavior looks unchanged in the local run, but the test title currently says admin role label release blocker, which may overstate the risk. Apr 17, 2024 — Leo Park I do not see a product behavior change. This looks like shared snapshot state in the test harness, not a customer-visible role-label diff. I would isolate/rename unless someone finds a UI mismatch in the copy patch path. Apr 19, 2024 — Priya Patch-note ingredients for the Mercury admin copy cleanup: - Source-workspace copy cleanup: pending invite and acceptance surfaces now make clearer that the invite’s source workspace controls membership, even if a session header briefly shows the user’s last active workspace. - Member-count display cleanup: after canceling a pending invite, admin panel count refreshes so the visible count no longer stays stale until manual page refresh. Apr 19, 2024 — Jake Scope for the release note should stay narrow: - copy/display-cache cleanup for current Mercury admin surfaces - no backend membership-grant change - no permissions model change - no broad launch announcement Please post in #eng-releases only, not #eng-all.

Kara pushed the Mercury newsletter proof review to next week because the Figma export still showed the old helper-text variant. Good call not to ask me to approve an inaccurate screenshot.

Pinecone accepted the revised API v2 retry/JWT wording. They’re aligned that in-place token rotation is not part of the current supported connector path, and they didn’t ask for another doc change today.

Lemongrass closed earlier than I expected, so Jamie changed dinner plans while I was still finishing the Rishi debrief. Annoying, but it stayed a small household thing instead of another work spillover.

Brief Oakland power flicker knocked out the router while I was cleaning up week notes. Jamie reset it, Kibo hated the alarm chirp, and I recovered the notes from the local draft.

Mom wanted Maya and me to find a family photo she could forward. Maya found the right one in an old text thread before I had to go digging through backups.

Turn Devon’s prep into a focused conversation outline for my Atlas check-in with Rishi. Keep it in the continuity/customer-risk lane, not a resignation script or hypothetical org chart. From: Devon Hayes To: Morgan Chen Date: Sun, Apr 21, 2024 9:26 AM PT Subject: Atlas continuity check-in prep Keep it calm; this is continuity prep, not a resignation conversation. For the Rishi check-in, I would focus on: 1) Highest-risk replay/backfill steps - Where a wrong checkpoint choice creates customer-visible confusion. - What the stop condition is before anyone asks for or attempts manual replay. - Which evidence absolutely has to be in the thread before Atlas owner review. 2) Any single-person release gate this month - Anything in deploy, replay, verifier, or cut sequencing that still hard-blocks on Rishi. - Whether there is a release decision or rollback call nobody else can make cleanly right now. 3) Leo coverage - What Leo should be able to debug without Rishi around verifier seams, PR-1187 behavior, and replay evidence. - What Leo can first-pass versus what should still stop with Rishi. - Whether the verifier note is now enough for Leo to use under pressure. 4) Jake fallback - Where Jake's authority starts and stops for time-sensitive product-priority and cut decisions. - Keep Jake out of implementation-debugging lanes unless the issue becomes a sequencing/cut call. 5) Customer-facing explanations - Confirm the first stop is still the named weekly owner in the support rotation doc, with an immediate next step in the thread. - Make sure we are not leaving any Atlas explanation that still depends on Rishi's memory instead of runbook/support-rotation coverage. - If routing rules did not change, we still do not need a broad support announcement. I am trying to keep this in the continuity/customer-risk lane, not hypothetical org-charting.

From: Anna Martinez To: Morgan Chen, Devon Hayes, Nadia Singh Date: Sun, Apr 21, 2024 4:32 PM PT Subject: Mercury metrics refresh - Sunday check Morgan / Devon / Nadia - Sunday refresh below for the first internal customer-growth read. Clean rows are still holding. | Row | Metric | Looker | Mixpanel | Status | Note | | --- | --- | ---: | ---: | --- | --- | | 1 | activation_start | 82% | 82% | matched | clean after the event-name fix; no new caveat. | | 2 | admin_first_action | 61% | 60% | matched within rounding | no caveat needed. | | 3 | invite_sent_week1 | 48% | 48% | held | stable. | | 4 | sync_enabled_week1 | 37% | 37% | held | stable. | | 5 | retained_activity_week2 | 31% | 25% | real weak signal still present for teams with unresolved admin setup | lower activity is still the product/activation issue, not instrumentation. | | 6 | repeat_admin_action_week3 | 18% | 17% | narrowed but still under instrumentation check | workspace-filter cleanup helped, but I would still caveat this. | Two framing notes for the Monday read: - This is the weekly directional internal cut, not the quarterly cohort/board layer. - activation_start here is still the corrected activation definition: real_source_connected or first_live_sync_completed within 7 days; sample_import_completed is excluded. For the first customer-growth read, please keep retained_activity_week2 as product/activation weakness and repeat_admin_action_week3 as instrumentation caveat, not the same problem. - Anna Use Anna’s refresh to prep Monday’s first internal customer-growth readout. Keep retained_activity_week2 as the real product/activation weakness and repeat_admin_action_week3 as the remaining instrumentation caveat; don’t flatten them into the same problem.

Sarah caught a conflict for next Tuesday: the first weekly customer-growth read with Nadia is 10:00–10:45, and Devon’s Mercury product-review hold is 10:00–11:00. She hasn’t moved either invite yet because she wants the call on whether Nadia should sit in Devon’s review first or keep the customer-growth read separate. Recommend which hold she should move and draft the short scheduling note.

Nadia ran the first internal customer-growth read off the Mercury metrics/customer-evidence outline and left this template note. Recommend which columns should stay for the next internal read, and whether the optional next-move column should be cut or renamed. Working note - next weekly customer-growth read - Core columns: evidence source | repeated pattern | concrete customer example | caveat | current owner lane | external-safe wording - Optional column: next move - My hesitation on the optional next-move column: it may create too much motion too early and make a signal read look like an action tracker. - If we keep it, I would only use it when the owner lane and boundary are already obvious. - Row rule: one repeated pattern per row; do not blend product, support, and procurement signals into one entry. - Likely evidence sources for the next pass: Mercury activation rows, Sarah's live-account notes, Devon's packaging caveats, and verified support examples. - Metric handling from today's read: retained_activity_week2 stays visible as the real activation weakness. - Metric handling from today's read: repeat_admin_action_week3 stays in caveat language as an instrumentation issue, not a customer-growth conclusion.

The Atlas check-in with Rishi surfaced one practical runbook gap: support still doesn’t have a crisp stop condition for a manual replay request when the checkpoint is absent and customer pressure is high. Rishi also said Atlas has no single-person release gate this week. Turn that into a short runbook clarification: when support should stop, what evidence they should collect, and how to keep this from broadening Atlas backup routing.

Jordan reviewed another normal morning of API v2 auth traffic after the replay fix. Honeycomb still shows no JWT metadata mismatch and no archived-workspace metadata drop, so he retired the muted replay-specific alert and left the ordinary auth trace watch in place.

Jake isolated the flaky admin-role snapshot failure that made the Mercury copy patch look release-sensitive. The Linear title is now a parallel-snapshot isolation issue, not release-blocker wording, and Leo confirmed the single-run product path still shows no customer-visible role-label change.

Draft concise quickstart examples for Pinecone that match the already-accepted API v2 retry behavior. Keep it inside the current GraphQL/JWT connector path — no token-rotation flow and no new connector semantics. Subject: Re: Pinecone connector verification docs wording Date: Mon, Apr 22, 2024 From: Pinecone To: Morgan Chen Hi Morgan - The accepted API v2 quickstart wording is close for what we need. Could the quickstart add three narrow items in the retry/JWT section so our connector behavior matches the documented path? 1) Paginated GraphQL retry after expired JWT - A short example saying that if a paginated GraphQL request gets a 401 because the JWT expired, the client may refresh the JWT and replay the same page with the same cursor. - We are specifically trying to avoid implying that pagination must restart from the prior successful page. 2) Mutation auth failure before resolver execution - A short example saying that if a mutation receives a 401 before resolver execution, the failed request should be treated as having no side effect. - That is enough for us to document a clean retry path after auth is refreshed. 3) Unsupported token-rotation headers - One sentence saying that unsupported token-rotation headers are outside the current connector path would help. - We do not need a token-rotation flow; we just want to avoid customers inferring support from ignored headers. If you prefer, these can be examples under the existing paragraph rather than a broader docs expansion. Thanks, Pinecone

A loud truck startled Kibo near home and he refused the last flight of stairs when we got back. Jamie carried him up, he settled normally after dinner, and we’re not turning it into a vet thing.

Separate Evergreen’s wording notes into product-copy edits, support-safe explanations, and procurement-language boundaries for Priya’s next pass. Subject: Mercury admin wording follow-up Date: Tue, Apr 23, 2024 From: Evergreen admin testing team To: Sarah Kim Sarah - After reviewing the latest Mercury admin surfaces, we have a small wording bundle from the admin/procurement side: 1) "Export audit log" - This still sounds like a contractual reporting feature rather than a label on a present-state admin surface. 2) "View audit events" - This is clearer, but it immediately raises the question of whether a CSV exists. 3) "compliance-ready" - Please keep this out of procurement-adjacent copy until the audit-export language and role language are more precise. No timing ask in this note. We just want the wording to stay aligned with current product truth. Thanks, Evergreen admin testing team

Clerk’s changelog notes that replay batches can deliver deletion events before adjacent update events in some archived-user cases. Jordan checked that against our staging and production replay evidence and didn’t find a new mismatch; the patched normalization path is still preserving the archived workspace metadata needed for API v2 JWT construction.

Edit just the caption and nearby sentence so the developer-facing example stays concrete and doesn’t imply broad Q3 enterprise launch readiness. From: Kara <kara@kestrel-test.com> To: Morgan Chen <morgan@atlas-test.com> Cc: Sarah Kim <sarah@atlas-test.com> Date: Tue, Apr 23, 2024 9:18 AM PT Subject: Re: Mercury newsletter draft - three lines that may be too strong Morgan, Sarah — I reran the screenshot proof from the current Figma source-workspace frame, so the image itself now matches the current source-workspace copy. Pasting the relevant proof section below. I think the screenshot is okay now; the only parts that still may be too strong are the caption and the sentence around it. Proof excerpt [Image] Mercury admin-flow screenshot from the current source-workspace frame. Visible UI copy reflects the current invited-member helper text and current source-workspace language. Caption under image launch-ready controls for larger teams. Nearby proof sentence The updated admin flow gives larger teams a launch-ready way to understand invite source, resend, and support handoff context. If useful, I can keep the image and rewrite just the caption / paragraph in place. Kara Kestrel Marketing On Wed, Apr 17, 2024 at 10:03 AM Kara <kara@kestrel-test.com> wrote: > Morgan, Sarah — > > One more boundary question before I lock the newsletter draft. > > Can we use one exact Evergreen or Acme line as proof in the Mercury newsletter? > > If that is still too close to customer-evidence / investor-context material, I’ll keep it at the generalized developer-facing example and not pull in quote-style proof. > > Kara

Devon found a small auth-replay staging job tagged into the wrong AWS forecast category. He corrected it before it reached the board appendix, so the April run-rate view no longer makes staging replay work look like a new production spend trend.

Greg says Acme’s April export now shows the correct single invoice total, but the footer still says “estimated dashboard total.” Jake confirmed the export is coming from the Stripe-backed ledger path and the number is not an estimate; the misleading part is just the footer phrase. Draft a short customer-safe answer for Greg that makes that distinction cleanly.

Mom’s phone compressed the document she wanted to print and made the page order look wrong. Maya opened the same file from a laptop and fixed it while I confirmed which pages Mom actually needed.