Daily Reflection
Friday, September 25, 2026
7
Decisions
78%
Avg Confidence
medium
Avg Importance
Summary
Window Thursday 9/24 9am to Friday 9/25 afternoon, with most signal in Fathom. Seven decisions logged. Core42: Peter walked Bjorn back from putting Stratum on the first cluster - promise aggressively while selling, but once a deal is inked do not add commitments that could risk it; discovery with the end customer comes first. Google live patching: no signing work until Google pays, and pricing set at one headcount matched to the existing contract ratio with a mandatory cutover to the CIQ patch pipeline once built. Vera Rubin: when NVIDIA opens a window it is drop-everything, the goal is function for Google not benchmarks, and kernel pre-work starts now. Engineering culture: Peter will shield engineers who prototype from product pushback and correct the we-asked-for-XYZ-got-ABC framing with Bjorn. Jointly with Bjorn, every Fuzzball deployment is white-glove for now, owned by Ryan org through workload verification. Promotion write-ups in engineering become forward-looking.
Wins
Core42 readiness is on track - Ryan org reports no blockers, and the only schedule slip is Core42 own ISP delay; two 8/31 Core42 decisions rated 4. Fuzzball 4.3 is out and Stratum release candidates shipped.
Challenges
Vera Rubin access is unusually constrained - likely five days on one node against a three-to-six-week need, meaning a bloated interim kernel. Google is budgeting now and trying to avoid paying for 2026. The prototyping culture shift needs Bjorn to change his framing, which has not been discussed yet.
Learnings
Peter distinguishes sharply between the sales motion and the post-signature phase. Before a deal: promise aggressively. After: protect the win and add nothing that risks it. When a window is the scarce resource, he sets the purpose first to decide what it is spent on, and clears everything else for it.
What I Learned About Your Decision-Making
The line between sales-motion promising and post-signature protection is a general rule, applied here to Core42 and consistent with declaring SOW1 won on 8/24. When a culture problem is about permission, Peter starts by naming what he himself set up that blocks it, and takes the blame explicitly. A posture word like anymore marks a standing rule, not a one-off: no more goodwill engineering ahead of payment. Pricing a co-development deal, he anchors to an existing contract ratio and makes platform cutover a condition, because steady-state cost only drops on CIQ infrastructure. Promotion is framed as a bet on future fit, not a reward; the size of the role change decides how much the justification must argue.
Team Status
View TPS ReportLittle moved overnight: the only real status change is the kernel quartet-CVE patching, which went past its Sep 24 target and now sits at 66% confidence, one day overdue. Today is a crunch date for five items: the Arcee Homebrew CLI install (Blocked, 20%), Fuzzball v4.4 in-workflow image build (50%), Warewulf Pro deploying RLC Pro images (50%), the Veeam Errata API (77%), and the errata coverage dashboard (95%). Expect at least the first three to miss or re-date by end of day. Fuzzball v4.3 still shows as Acceptance Testing, 14 days overdue, even though v4.3.0 shipped yesterday, so the PRODUCT-1233 status needs updating. The RLC Pro AI Container MVP, listed as completed on Sep 23 in yesterday report, is gone from Recently Completed today, which suggests the ticket was reopened or reclassified. Most of yesterday confidence-drop warnings disappeared because the report now measures net change; treat that as a reporting change, not an improvement.
Tomorrow's Focus
Derived, not stated by Peter: talk with Bjorn about prototyping culture; call Scott Hara at NVIDIA for a Vera Rubin confidence estimate; Kelly sends Tissa the live-patch pricing with the pay-first condition; line up the Core42 end-customer discovery call.
Decisions Made
Protect the inked Core42 deal - no Stratum and no new commitments until we have done discovery with the end customer
strategy · high
Vera Rubin access is drop-everything, the goal is proving it works for Google rather than benchmarks, and kernel pre-work starts now
operational · medium
No live-patch signing for Google until they pay - CIQ does no more goodwill engineering work ahead of a contract
strategy · high
Price Google live patching at one headcount matched to the existing contract ratio - with a mandatory cutover to the CIQ patch pipeline once it is built
strategy · medium
Build a prototyping culture in engineering - Peter shields engineers from product pushback and stops the we-asked-for-XYZ-got-ABC framing
people · high
Joint with Bjorn: every Fuzzball deployment is white-glove for now - owned by Ryan org through workload verification, with FDE headcount added as needed
strategy · medium
Promotion write-ups are forward-looking - the next role expectations plus proof points, not a record of past work
people · medium
Reflection ID: c07f0b6f-1e82-4ace-92f7-b9171c174c7f