The course
Shipping AI output you can trust. Four weeks. A cohort, not a video course.
This course is not open for enrollment yet. It launches on Maven only after two free Lightning Lessons build a real waitlist, per the same discipline the course itself teaches: do not ship before the evidence says you are ready.
Technical PMs, engineers, and data people who already use an LLM to get to an answer and need to know when to trust it. Zero math prerequisites. This is not a research-mathematics course, and math credentials are not the point; catching a wrong answer before it ships is the point.
Four weeks. One live session plus async exercises each week. A beta cohort of 8 to 15 seats, priced at $950 to $1,500, with a discount for the first cohort in exchange for feedback. Each student ships one verified public artifact as the capstone, graded on its audit trail, not on whether the answer turned out to be right.
Four weeks, built from the same ladder this site's guide teaches.
Self-review, internal consistency, independent re-derivation, external authority, hostile expert. Lab: break a plausible wrong answer someone else built.
Pre-registered anchors, kill criteria, reproduce-before-extend. Lab: pick a real target (a dataset, a public table, an existing sequence) and write its anchors file before touching it.
Two-implementation traps, external authorities, exact versus float, validated-at-n is not validated-at-n+1. Lab: hunt a seeded bug that two agreeing programs both share.
Provenance (models, prompts, transcripts), disclosure norms, the submission checklist. Capstone: ship one verified public artifact with its full audit trail.
Two free 30 to 45 minute sessions before any cohort opens. Both are how the waitlist gets built, and both are worth attending even if the course never happens for you.
The reproduce-before-extend method, live: matching a published census to four decimal places, then chasing down the rows that would not match.
PMs, engineers, data people who use AI and need to trust the output. No math background required.
Counting by brute-force listing is exponential death. Why exact arithmetic and pre-registered anchors are non-negotiable once a shortcut that fast is on the table.
Engineers and quants.
34 contributions live in the OEIS under Tyler's name, each verifiable by A-number. The teaching material is drawn from a public research practice, not a curriculum written for the course: pre-registered anchors, dual independent implementations, checkpoint and resume with bit-identical verification on long-running computations, and an append-only audit ledger, all visible at Checkable. The record is public before the course is.
Before committing to a waitlist, read the how-to-vibe-math guide this course is built from.