The equities are incidental.

The reframe

The documents give it away. Year one's stated output is "a compliance record and a forward performance record — not P&L". Phase progression is gated on compliant cycles, not on returns. The reward for good behaviour is more capital; the reward for good returns is nothing. That's operant conditioning pointed at process rather than outcome — the correct direction, and almost nobody builds it.

Bias by bias

The disposition effect — realising gains, riding losses — is structurally inverted: the 200-day exit fires mechanically on the losers and stays silent on the winners, and the sell decision arrives pre-made, never chosen while feeling it.

Myopic loss aversion is regulated at its source. The most important rule in the whole rulebook is the one that sounds like laziness: no price monitoring of any kind between rebalances. Twelve observations a year instead of two hundred and fifty changes the felt volatility of an identical portfolio enormously. Most systems regulate trading frequency; this one regulates looking frequency, which is upstream of it.

Attention-driven buying has no channel: no watchlist, no news input, no ideas pipeline through which salience could enter. Mental accounting — normally a bias — is deliberately weaponised: the ring-fence is an economically irrational partition of fungible money, and it is exactly what makes the experiment psychologically survivable. The zero-balance test pre-experiences the worst outcome in advance, which blunts the panic response when a version of it arrives. Escalation of commitment is countered by the pre-written, non-editable retirement rule.

The three genuinely sophisticated pieces

"No override in both directions." Almost every retail rule system prohibits adding risk; very few prohibit skipping a printed trade. But the scary-looking buy you talk yourself out of is where the return distribution lives — momentum entries feel worst exactly when they work best. Banning the omission is the harder and more valuable half.

Amendments must be written before back-checking what they'd have earned. This severs the link between outcome and rule change — the exact mechanism by which systematic strategies die. Not through one dramatic override, but through a sequence of individually reasonable revisions, each prompted by a recent bad month. Hindsight bias needs the result to operate on; this denies it the result.

The temptations log. A running record of urges not acted on. There is no financial reason for this file to exist. Behaviourally it is the highest-value artefact in the repository: it converts an impulse from something you're inside of into something you're looking at, and it generates the only leading indicator available. The breach log says the system already failed; the temptations log says pressure is building, months earlier.

Where it's structurally weak

Self-enforcement. Ulysses had rope and a crew with wax in their ears; this has a markdown file and a date. Every rule is self-imposed, self-monitored, and — critically — self-revocable. And the failure mode isn't violation, it's evaporation: rule systems overwhelmingly die from quiet abandonment, not dramatic breach. A missed cycle is logged and is explicitly not a violation — a humane rule worth keeping — but it means the system can stop being run without ever generating a single breach entry. An abandoned repository and a compliant one look identical from the outside.

The month that actually breaks it

Not the drawdown — the −25% month has a written continuation case waiting for it, and shared pain is easier. The dangerous month is the one where the gates fail broadly, the account sits mostly in cash, and the market rips upward for ten weeks. No loss to justify anything. Just a correct, obedient system underperforming visibly and boringly while everyone else makes money, with a printed sheet that says do nothing. If it dies, that's the month — and the mechanism won't be an override. It will be an amendment proposal that sounds completely reasonable.

Two series are worth recording alongside the returns: temptation entries per cycle, and hours elapsed between sheet generation and execution — a near-perfect proxy for reluctance that costs nothing to measure. The returns will be statistically meaningless in 2029. Those two series won't be.