← Back to Reflections

Daily Reflection

Friday, October 23, 2026

10

Decisions

76%

Avg Confidence

medium

Avg Importance

Summary

Two-and-a-half-day window, Wednesday 10/21 after the morning reflection to Friday 10/23, dominated by the Core42 deployment turning political, the Fuzzball plus Stratum alignment session, a Google-NVIDIA supportability problem and a dense run of Thursday 1:1s. Ten decisions logged, all reviewed one at a time. Core42: when Core42 pulled Fuzzball out of the Oct 31 plan, CIQ holds to the SOW - stock Kubernetes plus Fuzzball as soon as the network is up, no CIQ-installed CozyStack, and technical pushback routes to Peter rather than to engineers on the ground. Fuzzball plus Stratum: under the product split Bjorn set, Peter fixed engineering invariants - one identity and one pane of glass, nothing built twice, and Engineering owns the developer experience. Customer-specific asks, including a large AWS private-offer deal and a field-built OCI image, go through Work Ingestion and RAT build and validation with no side channels. Peter took personal ownership of getting NVIDIA to commit in writing to Google that it supports CIQ-built RLC. Apptainer is held to the zero-engineering-hours promise so the maintenance cost reaches Product. The program-manager req for the PIC org is declined in favor of an engineer if headcount is spent. Groundwork starts on a contingency RESF leadership path for Tim Pepper. Coverage: eight Fathom meetings 429d in the gather; all eight were recovered by hand before analysis, so no gaps remain.

Wins

NVIDIA agreed within hours to work toward a three-way NDA and a joint-support statement to Google. Engineering shipped PHP2Shell fixes for RLC Pro LTS 9.6 and 9.2 about three and a half hours after disclosure, and Peter was quoted in The New Stack. Stratum is on track for open source and launch next week. The Fuzzball plus Stratum session produced a locked product split and a concrete next step (written vision, diagrams, use-case matrix).

Challenges

Core42 removed Fuzzball from the Oct 31 plan, the network is still not complete, and a single Core42 network engineer is carrying a timeline nobody believes. A Google kernel team is forming in-house while an NVIDIA support clause undercut CIQ supportability. A large customer private offer stalled for a week because the ticket lacked business context. Recurring TPS slippage: 10.2 GA and Pro H packages overdue, FIPS still blocked on ARM hardware, NVIDIA driver automation moved again.

Learnings

Peter treats the SOW as the shield in a politically divided customer and routes technical pushback to himself so engineers never argue politics. He sets architectural invariants that must hold whichever way an open product debate lands, and leaves the product question with its owners. Intake and validation rules are held most visibly exactly when a big deal and an angry customer push against them. When a partner policy undermines CIQ with a customer, he goes after the policy at its source, in writing, rather than asserting CIQ own supportability.

What I Learned About Your Decision-Making

Peter separates engineering invariants from product choices in the same meeting: he will agree with Bjorn on the product split and then state what must be structurally true regardless of how Greg and Bjorn settle the open question. Peter will let a promised-free cost become visible rather than absorb it, accepting real degradation so the owner of the promise prices it in. A reinforcement is still a decision when someone reopens a settled question and Peter has to choose to hold the line; explaining a past decision to a new audience is not. Peter goes to the source of a third-party policy that hurts CIQ with a customer and asks that party to fix it in writing, holding the CEO in reserve as closer.

Team Status

View TPS Report

Little moved overnight; the story is slippage, not new work. RLC Pro LTS 10.2 GA (Justin) is now two days past its Oct 21 target and RLC Pro H 10.2 package updates went overdue yesterday, both after date moves that came with no confidence change, and Koji production signing (Nathan) is two days late while the Veeam Errata API is a week late in acceptance testing. The NVIDIA driver automation pushed again, Oct 23 to Oct 27, and three RAT items (SBOM for rlc-bootc, Depot/Portal reliability, retiring Jenkins from image builds) are all due today. 10.2 FIPS to atsec stays Blocked: Brady says an ARM box arrives next week but Sapphire Rapids x86 is up to four weeks, which puts the Oct 30 target out of reach unless cloud bare metal is used. Core42 still reads 90% for Oct 31 even though the out-of-band network remains the critical path and CozyStack validation cannot finish before Oct 28, so that confidence number is stale.

Tomorrow's Focus

Derived, not stated by Peter: follow up with Scott Hara on the NVIDIA engineering alignment and the statement to Google; confirm the Fuzzball install stays in the Oct 31 scope with someone at Core42 who has signing authority; review the written Fuzzball plus Stratum vision and diagrams at next week follow-up.

Decisions Made

Reflection ID: 49e542c7-a7dc-4dda-b96e-40ede962106b