Started 2026-07-27 · living profile, updated as patterns are actually observed
Not a technical concept note — this tracks how I reason and make decisions, across whatever project I’m working in at the time. Requested this deliberately as a self-improvement tool: I want direct, specific feedback on my thinking, not just on my code. See a project’s own thinking-patterns.md (e.g. projects/gees-arise/thinking-patterns.md) for the specific instances behind entries here — this file is the cross-project pattern, not the story each time.
Recurring strengths
Catching real problems in my own proposals, unprompted, before anyone else flags them.
First clear instance (gees-arise, 2026-07-27): proposed defaulting every circle’s reset time to 23:59 “to match when the cron runs,” then in the same message caught that this breaks across timezones — 23:59 UTC isn’t midnight for a user in a different timezone, so a “universal” default doesn’t actually deliver the “resets at your midnight” experience it sounds like it does. Explicitly chose to defer rather than force a shaky fix through. This is a strong instinct — most people ship the naive default and only discover the timezone problem when a real user in another timezone complains.
Recurred 2026-10-01 (SWE 101). After a review showed the course was overloaded, said unprompted: “20 something weeks is a lot” and proposed splitting it into 6–12-week courses (101, 102, 103). A real structural catch — shorter courses give a nearer finish line, which is the condition his own JAMB result came from.
Recurred 2026-10-01 (vault work), process this time. Mid-way through a long batch of generated lessons — seven companions, about 50 labs, nothing committed — asked unprompted: “don’t you think this should be done in batches… let’s commit before going on.” The risk was real (one bad step could have tangled a large uncommitted change) and the AI doing the work hadn’t raised it. Same instinct as the timezone catch, applied to a workflow rather than a design.
New strength, 2026-11-02 (NextVibe): domain knowledge that overrode a technical recommendation, correctly. Twice in one session he disagreed with the AI’s design and was right: wishlist items should stay open past 100% (“people wanna give”, and the money all lands in one account anyway), and spin the wheel belongs inside the games system as a round, not as a separate module. Both times he asked “what do you think” instead of just overriding. Details: nextvibe/thinking-patterns.
Recurring blind spots / things to watch
Proposing a new default without fully tracing through an architectural decision made moments earlier in the same conversation.
Same instance as above: the reset-time proposal (coupling every circle’s reset to the cron’s schedule) ran against something we’d just built — the whole reason the scheduled-jobs migration used a 15-minute pg_cron sweep instead of once-daily Vercel Cron was specifically so each circle’s reset time (and audit deadline) could be independent of the sweep’s own schedule. The sweep catches each circle’s deadline within ~15 minutes of whenever that circle’s deadline actually is — there was never a need to couple the two. Worth watching for: after a design decision lands, check a next proposal against it before floating the idea, not just against the original requirement.
Third instance, 2026-10-01 — a strategic decision re-opened by a feeling, not by its reasoning. In the same message as the good catch above: “I also think I need the fundamentals to apply for jobs. I don’t know where I was rushing.” That reverses the central decision of SWE 101 — apply from week 7, in parallel — which his own plan had argued for in writing: “Start before it feels ready. This is the instruction most likely to be ignored”, and “treating [fundamentals] as a prerequisite for applying is what pushed the original plan to 52 weeks before a single CV went out.” It arrived around week 6, the week before applications were due to start — exactly where the plan predicted the urge to wait would appear. And it merged two separate findings: the review said the study load was rushed, not that applying was premature. What better looks like: before reversing a decision, restate the reason it was made, and say what new evidence defeats that reason. “I feel unready” is not new evidence; the plan expected that feeling. Correction, same day: there was evidence I didn’t know about. He had already been applying, got interviews, and failed the graph questions in the DSA rounds. So the reversal was driven by a real result, not only a feeling, and that part of the critique was wrong. What still stands: the evidence was “graphs are a gap”, and the conclusion jumped to “fundamentals before applying”. The narrower conclusion — “fix graphs while still applying” — is what the evidence supports. The lesson for both of us: state the evidence behind a reversal, not only the conclusion. “I failed graph questions in two interviews” would have made the case on its own.
Recurred 2026-11-02 (NextVibe), the build version. A new game idea arrived in the same message as a bug report, before the previous feature (the wishlist) had been run end to end, and was designed and built the same day. Same shape as above: new work starts before the last decision or build is checked. See nextvibe/thinking-patterns.
Notes on communication pattern (not reasoning quality, but adjacent)
Messages sometimes bundle several independent, high-effort asks together (UI gaps + a new default + “continue the build plan” + a large new meta-request, all in one message, 2026-07-27). Not wrong, but worth naming: it risks any single one getting shallower treatment, or the most time-sensitive one getting lost among the others. Not a “bad thinking” pattern on its own — a pacing thing to be aware of.
Recurred, larger, 2026-08-23. One message contained nine asks: is the Python course complete · cross-reference everything against roadmap.sh · finance/tax/entrepreneurship · the systems-engineering gap · other kinds of software engineering · game development · desktop apps · data centres and infra careers · manufacturing as a business. Closing line: “I’m thinking of so many things I can do to make a future-proof career man.”
Two things worth separating, because they’re different and only one is a problem.
The bundling was partly rational this time. The stated reason — a subscription ending — is a genuine, time-bounded constraint, and “extract durable reference material while I can” is a correct response to it. Six domains were mapped in one session precisely because they were batched. That’s the pattern working for him.
The scatter underneath is the thing to watch. Every item was framed as a possible career direction, and they arrived forty-eight hours after deliberately parking Java, mobile, embedded and robotics on the reasoning that “a CV aimed at four roles reads as aimed at none”. The parking decision was sound, was his own, and was re-opened almost immediately — not by revisiting the reasoning, but by curiosity arriving from a different angle.
The same shape as the 2026-07-27 entry above: a decision lands, and a later proposal isn’t checked against it. There it was one architectural decision within a conversation; here it’s a strategic decision across two days.
What worked: the interests became notes, not courses, and went into the parking lot — which is the mechanism his own system already contains for exactly this. The distinction he was already making implicitly, and worth making explicitly: writing a map is cheap and reversible; starting a course is neither.
Worth watching: whether “future-proof career” is doing the work of anxiety rather than planning. Six directions is not a hedge against an uncertain future — it’s the thing that makes the current one take longer.
Software engineering
Infrastructure & security
Computer science
Hardware & engineering