Daily Reflection
Tuesday, November 3, 2026
10
Decisions
77%
Avg Confidence
medium
Avg Importance
Summary
Four-day window, Friday afternoon 10/30 to Tuesday morning 11/3, dominated by the Google GDC kernel scramble and the strain of running Core42 datacenter work out of engineering. Ten decisions logged, all reviewed one at a time and all confirmed without correction. Google: what fits the existing build machinery ships Monday and everything else is quoted work, slips are written down against Google's own constraints, a rough weeks-not-days estimate is enough, and Google's narrow previously-identified-CVEs contract wording is accepted as future scope leverage. Core42 and datacenter: Stratum stays out of the IRIS deployment to protect a closed win and avoid a first large deployment failing in public, Grenoble can take it later; and Peter told Bjorn ahead of the C-suite that datacenter work needs its own team and deals that pay now, with engineers no longer signed up for out-of-remit work. Also: public CVE claims must match what is actually in customers' hands; the Linux and RAT teams get a sequenced award with Greg's public thanks and Peter's in-person praise at the on-site, plus a public line from Greg that Ginux is no replacement; the Fuzzball/Stratum doc is signed off with a reserved right to shape a unified UX; customer-specific fixes on 10.2 LTS need universal LTS demand to protect bug-for-bug compatibility; a kernel-security candidate in India is evaluated on fit with Nathan running it; and Skip is repaid for working the deadline from a family vacation. Coverage: the 10/30 Fuzzball/Stratum alignment meeting fell outside the Fathom list cap and the 11/2 Google sync in a Google room was unrecorded; both were covered from Slack and email.
Wins
CIQ had the MPTCP CVE fix built before the public disclosure and promoted to customers within about 36 hours while major competitors had not shipped. CIQ took its first self-serve purchase. The Fuzzball/Stratum strategy got sign-off without waiting on Peter. Google accepted contract wording that bounds CIQ's obligations to an identified CVE list.
Challenges
Core42 IRIS is back to square one on networking after roughly 30 days of daily monitoring, with Ryan's team and others burned out and Fuzzball work displaced, while Grenoble arrives in 8 to 9 weeks with a wider scope. Google's Anthropic-driven asks replaced its roadmap and kept moving across the 1.16, 1.17 and 1.18 lines. The TPS board barely moved: every Oct 30 item is now four days overdue, LTS 10.2 GA was re-dated to Nov 5 without a confidence change, and the Errata API is 18 days late.
Learnings
Peter draws commitment lines at CIQ's machinery and controls rather than at effort: what the existing pipeline produces is free to promise, anything else is quoted. He separates ready-to-sell from ready-to-deploy-into-a-live-deal. He holds public claims to the customer's point of view. Earlier conditions such as scope arriving with headcount get called in when the load returns, not restated.
What I Learned About Your Decision-Making
Peter treats CIQ's existing build pipeline as the boundary of what can be promised on a date; anything needing a new path becomes quoted work regardless of how small it sounds. He accepts a customer's own restrictive wording when it bounds CIQ's obligations, because a boundary in the customer's language is harder to dispute later. He separates a product being ready to sell from it being ready to drop into a deal still delivering its base scope, and chooses where a first large deployment absorbs its inevitable failures. He calls in an earlier condition (scope arrives with headcount) when the load returns rather than restating it, and carries a direct report's burnout upward even when the report asked for nothing. He sequences recognition so each layer reinforces the last, and pairs it with removing a morale threat in the same exchange. When someone pays a personal cost for a customer fire, he looks for a repayment that matches the cost and asks the person directly what that would be.
Team Status
View TPS ReportAlmost nothing moved since yesterday: every Oct 30 item is now four days overdue and no target or status changed, only two confidence bumps (Fuzzball v4.4 image definition in workflow to 90% for Nov 4, ISO 27001 surveillance to 80%). The live risk is Core42 GB300, three days past its Oct 31 target at 50% with the customer network still down. RLC Pro LTS 10.2 GA was re-dated on Nov 2 from Oct 21 to Nov 5 with no confidence change, the Errata API for Veeam is 18 days late in acceptance testing, the two RLC release-file items sit near 50% four days past due, and 10.2 FIPS to atsec stays Blocked for Nov 17. Due this week: Fuzzball v4.4 image definition and NVIDIA driver automation (Nov 4), LTS 10.2 GA (Nov 5), Fuzzball DGX Spark and RLC Pro AI 10.2 (Nov 6) - the last two carry 30- and 7-day slips with no confidence update.
Tomorrow's Focus
Derived, not stated by Peter: send the Google resource head count to Bjorn and Kelly Hall by Wednesday night; confirm the CVE award details before Wednesday's on-site praise; watch whether the datacenter-team position landed at C-suite before the Grenoble SOW goes out; follow up on what Skip asked for.
Decisions Made
Keep Stratum out of the Core42 IRIS deployment - take the win we have; Stratum can go into Grenoble
strategy · high
Datacenter business needs its own team and deals that pay now - stop signing engineers up for work far outside their remit
strategy · high
Public CVE claims match what is in customers hands - fixed internally is not fixed; engage fast with accurate words
operational · medium
GDC targeted-CVE builds - what fits the existing build machinery ships Monday, everything else is quoted work, slips are attributed to the customer constraint, and a rough estimate is enough
strategy · high
Accept Google previously-identified-CVEs contract wording - it is not what they really want, and their own language becomes scope leverage later
strategy · medium
Award the Linux and RAT teams for the CVE onslaught - deliberately sequenced: Greg thanks them publicly, Peter praises them at the Wednesday on-site, award announced then; Greg disowns Ginux-as-replacement
people · medium
Sign off on the Fuzzball/Stratum strategy doc while reserving one right - shape the UX so there is one unified experience regardless of what underpins the cluster
technical · medium
Customer-specific fixes on 10.2 LTS only if every LTS customer wants them - otherwise it is a multi-year branch and breaks bug-for-bug compatibility with RH
technical · medium
India location does not disqualify the kernel-security candidate Greg referred - take it seriously, Nathan runs the conversation
people · low
Make it up to Skip for working the GDC fire from a family vacation - money is a given, pair it with time away or something better, and ask him directly
people · low
Reflection ID: 6b6d7c2b-b5a7-4fec-8498-4c02e9931738