Evidence up, authority down, context sideways — and the audit that checks it.

The three directions

Each direction carries something different, and only one of them carries permission.

Evidence moves up: a trade produces an outcome, the outcome becomes a journal row, rows become weekly hit rates, hit rates become expectancy, expectancy becomes a structural read. Nothing in that ascent grants anyone the right to do anything — it is pure observation accumulating into interpretation. Authority moves down: capital state defines the arena, the arena selects a row, the row caps the tier, the tier resolves to a pool, the pool sizes the trade. Context moves sideways: volatility reads, regime labels, and coefficient guidance arrive from the side and inform how a permitted trade is managed. The asymmetry is the whole design — evidence earns nothing by itself, and context authorizes nothing by itself.

The audit

Two violations worth checking for by name.

The fastest structural audit available is a search for wrong-direction flow, and the page's own examples are the canonical pair. A sizing decision quoting floating P&L is authority reaching sideways into live evidence — the deployment layer consulting a number that has not settled, which is exactly how open positions start influencing the size of the next one. A diagnostic writing into a journal is evidence flowing backwards — the interpretation layer editing the record it was supposed to interpret, which destroys the journal's status as system of record. Both look locally reasonable when they happen. Both are the beginning of the everything-modifies-everything state the architecture exists to prevent.

• Sizing that quotes floating P&L → authority consulting unsettled evidence.
• A diagnostic writing into the journal → interpretation editing its own source.
• Context granting permission → a good volatility read expanding risk past the gate cap.

Why direction, not content

The audit is structural, which is why it works when the numbers look fine.

Most system checks ask whether the values are right — is this expectancy plausible, is that drawdown accurate. A direction audit asks something different and considerably cheaper: is this number arriving from a place it is allowed to arrive from. The advantage is that it detects problems that produce entirely reasonable-looking output. A sizing decision informed by floating P&L will return a number, and the number will be in range, and nothing about it will look wrong. What is wrong is the dependency, and dependencies are visible only when you audit direction rather than value.

The subtle one

Context is the direction most likely to leak, because it feels like evidence.

Evidence and authority are easy to keep separate — they originate in obviously different places and travel obviously different routes. Context is the difficult one, because a strong volatility read genuinely is information about the world and it genuinely does bear on the trade. The temptation is to let it do slightly more than inform: a confirmed ATR expansion becoming a reason to size up, a clean regime label becoming a reason to override a gate constraint. The architecture's answer is categorical rather than proportional — context never grants permission regardless of how strong the context is, because a rule that bends under sufficiently good conditions is not a rule.

What the flows preserve

Directional discipline is what keeps the system changeable.

There is a practical payoff beyond correctness. When every flow is legal and declared, the consequences of changing any one module are boundable — you can trace what it feeds and what feeds it, and the trace terminates. When flows have leaked, that trace stops terminating: every module potentially touches every other, and the honest answer to 'what does this change affect?' becomes 'possibly everything.' At that point improvement stops, not because the operator lacks ideas but because no change can be reasoned about safely. Directional discipline is what a system pays now so that it can still be modified later.