Opti-Hour
Sole engineer. Concept, build, ship, iteration.
Timeframe
v1 shipped, then 4 weeks of live iteration. Offline now.
Stack
Claude Code
Google for timing.
A personalized astrological timing calculator.
Best times for important activities, per user, per week.
Problem.
Timing questions get generic answers.
The personal answer is real math: 98 techniques, 18 transit modifiers.
Per user, per week. Nobody runs that by hand.
Constraints.
One engineer, end to end.
Negatives must veto. A bad hour cannot average into a good one.
A full week of analysis has to land in seconds.
Birth data and timezone must be exact. Everything downstream depends on it.
Approach.
- 1
Score 6 dimensions per hour. Veto logic on top.
- 2
Precompute the week as a heat map. Drill down to the hour.
- 3
Onboard once: birth data, timezone autodetected.
- 4
Score in real time. Export winners as ICS, where plans live.
- 5
Ship v1 with Claude Code. Iterate on user feedback for 4 weeks.
Artifact.
98
techniques per weekly analysis
18
transit modifiers on top
15-20s
full weekly analysis, per user
6
scored dimensions, veto logic
Dimensions: luck, mental, physical, emotional, social, spiritual.
v1 features: activity search, weekly heat map, hourly drill-down,
real-time scoring, ICS calendar export.
Results.
15-20 sec
One user's full week: 98 techniques, 18 modifiers, ranked.
4 weeks
Live iteration driven by user feedback after v1 shipped.
Status.
v1 is offline. It worked — and it wasn't verifiable.
Rebuild in progress under the playbook. Log coming.
[Pending]
Rebuild log link.
Learnings.
Speed without verification builds code you cannot own.
It shipped, it worked, and I could not prove it. So it is dead.
The rebuild pays the verification cost up front.
[Pending]
Three concrete unmaintainability examples. The kill note earns them.