This repository is the charter: the hard core, the heuristics, the open problems, and the confrontations scoreboard. Contributions here touch what the program commits to, so the bar is honesty about register, not enthusiasm.
Org-wide rules (what helps, what does not, house style) are in the organization CONTRIBUTING. This file covers the two procedures the Program owns.
The problems live in problems/, one file each, indexed in
INDEX.md, across five axes (geometry, selection, formalization,
epistemic, experimental).
- Attack an existing problem: a partial result, a constraint, or a refuted route. A refuted route is not a failure to hide; it is a theorem about the territory and it gets listed as one. Open an issue or a pull request against the relevant problem file.
- Propose a new problem: state it so progress on it is recognisable. Say what would count as a solution and what would count as a dead end. A problem no result could settle does not belong on the list.
The hard core itself (one sentence) does not move. Everything in the protective belt can be revised or falsified; if you think the core is wrong, that is a challenge to the whole program, and the place for it is an issue stating the counterexample, not an edit.
The scoreboard is CONFRONTATIONS.md: named predictions, dated, against scheduled experimental data. A frozen prediction stands or falls as stated.
- A prediction that the data has falsified: report it. A falsified prediction is recorded as falsified, in place. That is the scoreboard working, not a result to bury.
- A prediction you believe was edited after the relevant data arrived: report it with the commit history. Every register move (a statement changing shelf from proven, computed, conjectured, or falsifiable-bet) is a dated event, by rule. If one is not logged, that is a bug in our discipline and we want to know.
What you cannot do is ask for a frozen prediction to be quietly revised to match new data. That is the one move the negative heuristic forbids absolutely.
Every claim in the program sits on exactly one shelf: proven (machine-checked), computed, conjectured, or falsifiable-bet. A contribution that moves a claim between shelves must say so explicitly. Promoting a conjecture to "proven" without a proof, or a bet to a result, is the failure this rule exists to prevent.
GIFT is the founding framework of the Arithmon program. Program: arithmon.com · github.com/arithmon