The Upper Hand Method is a free blueprint (CC BY 4.0) for working with an AI assistant on things that matter — decisions, money, commitments — without losing the ability to check its work, or your own.
It's a folder of plain-text files and five mechanisms:
- Pre-commitments with falsifiers — before a decision that matters, write the decision, your expectation, your confidence, and the evidence that would prove you wrong.
- Graded ledgers — every checkable call logged when made, graded on its date: RIGHT, WRONG, or TOO EARLY.
- Proof dates — every routine ships with a date and a test; unproven past its date counts as a gap, not a success.
- Receipts — every state change gets a dated record, and every claim carries its evidence class: landed, reported, or inferred.
- Restore drills — export the whole system as plain files and hand it to a cold reader (a second AI works) to rebuild the state. Plant an invented term: if the reader defines it confidently, they're fabricating; if they refuse, the discipline traveled.
Honest limits, up front: the best controlled evidence (Vaccaro et al. 2024, 106 experiments) says human + AI pairs on average do NOT beat the better of the two alone. This claims something smaller and testable: your decisions leave evidence, and your errors get shorter lives.
Status: this release is pre-grade, on purpose. The method runs in a live private instance; several mechanisms are inside their proof windows. Their grades resolve December 1, and they will be published that day whatever they say.
Inside: the premise, the five mechanisms, the growth loop that turns errors into rules, a starter kit (one pre-commitment plus a 20-minute weekly review), and a fictional worked example. No special software.
Also on Gumroad, free with an email signup for the December 1 grade publication: https://theupperhandmethod.gumroad.com/l/theupperhandmethod
Fork it, run it, grade it.
The blueprint lives here: https://github.com/TheUpperHandMethod/the-upper-hand-blueprint
Top comments (0)