The rail is seven mechanisms, not a metaphor.

Not interchangeable

Seven mechanisms listed together are read as one idea.

The page names seven guardrail components in a single sentence, which compresses well and hides that they operate on different inputs, at different moments, with different consequences. A reader takes away 'there are guardrails' and misses that a badge and a System Lock have almost nothing in common except membership in the list. Separating them is what turns the rail from a reassuring noun into something an operator can reason about. Compression is useful in copy and costly in comprehension, and the trade is worth making explicit: a reader who cannot distinguish the seven cannot reason about which one failed when something goes wrong.

What each intercepts

Seven mechanisms, seven interception points.

Each component sits at a distinct point in the deployment chain:

• Gates — intercept total authority, on drawdown from peak.
• Tier caps — intercept per-trade magnitude, on evidence quality.
• Throttle directives — intercept the current deployment, on open exposure.
• Brakes — intercept trade frequency when conditions warrant slowing.
• Badges — intercept nothing; they mark state so the operator can see it.
• Percentile bands — intercept interpretation, by supplying normal ranges.
• System Lock — intercepts everything. The terminal state.

Two that do not constrain

Badges and bands are informational, and that is not a weakness.

Two of the seven govern nothing. A badge marks a state; a percentile band supplies a range. Neither prevents any action, and both belong in the list because interception is not the only way a guardrail works — much of the damage the rail exists to prevent comes from misreading a situation rather than from deliberately breaching a limit. A trader who sees a drawdown correctly labelled as inside the expected band does not need to be stopped from panicking, because the panic does not start. This is also the cheapest part of the layer to implement and the part most often omitted from competing systems, which tend to build constraints and leave the operator to interpret the situation unaided.

The terminal one

System Lock is a different kind of thing from the other six.

Six mechanisms adjust; one stops. System Lock is not the bottom of a gradient but a discontinuity — the state in which no deployment is authorised at any size, and the account's only permitted activity is review. Its existence is what allows the other six to be graduated rather than paranoid: because a hard floor exists and is specified, the intermediate bands can afford to compress smoothly instead of hedging against catastrophe at every level.

Why so many

Each mechanism reads an input the others cannot see.

Seven mechanisms is only justified if each consumes information the others lack, and they do. Drawdown from peak, evidence depth per branch, open exposure right now, condition-driven frequency limits, current state, distributional context, and a terminal threshold are seven genuinely different inputs. A single unified control would have to reduce all of them to one number, and that reduction would necessarily discard the distinctions the individual mechanisms exist to act on. The seven also fail independently, which is a property worth having: a mechanism that is misconfigured or switched off degrades the layer rather than disabling it, and the remaining six still observe the situation.

Reading the list correctly

The rail is a layer, and layers have holes unless they overlap.

The right mental model is overlapping coverage rather than a single barrier. Any one mechanism can be inapplicable in a given situation — gates say nothing about a normal-state account taking too many concurrent positions, and throttle directives say nothing about a compressed account holding a single oversized one. Overlap is what makes the layer hold: several mechanisms observe most situations, and the ones that do not apply cost nothing by remaining silent.