Retrieval mode says where a value came from this second. Reliability says how well the underlying quantity is observed at all, which is a property of the source and its method rather than of this request. petral uses four tiers, and the ordering matters more than the labels.
`settlement` is an exchange's official mark, struck on a published methodology, identical for everyone, and revised only in rare announced circumstances. It is the strongest thing available. `official` is a number published by a statistical agency or a recognised assessor: a government inventory survey, a price reporting agency's assessment. Well-constructed, methodologically documented, and an estimate.
`proxy` is a series standing in for the quantity you actually wanted, because the real one is unavailable. A tanker equity index used as a freight indicator is a proxy. `estimate` is a modelled or inferred figure with no direct observation behind it.
The tier is a constraint on what can be claimed, not a quality score. A proxy is not a bad number; it is a number that supports a narrower claim. Tanker equities genuinely do respond to freight conditions, so they support a statement about direction and tell you nothing about the level of a specific route rate. Using it for the second claim is the error, and the tier is what warns you.
This is also where the crude-and-products unit problem interacts with reliability in a way worth noticing. A comparison that crosses the $/bbl and $/gal boundary needs a conversion, and a conversion is arithmetic that does not degrade reliability. A comparison that crosses from a settled contract to a surveyed grade does degrade it, and the arithmetic looks identical in both cases. Reliability is not visible in the formula.