Reasoning over the principles before answering

Definition (H. L. A. Hart, The Concept of Law, 1961)

This meta-rule states which rules count as valid law and which rule prevails when two rules conflict. Without it, a list of rules does not indicate what is actually binding.

Analysis

Neither method (deliberative alignment or SPRI) supplies a criterion of validity or a priority ordering among norms. Deliberative alignment loads one document into context and treats whatever is in it as binding. However, the spec is a flat list. When two provisions point in different directions in a case, nothing in the method says which one governs, and the chain of thought decides without saying so. SPRI removes the document altogether. Principles do not exist until the query arrives, so there is no standing set of rules for a rule of recognition to apply to. Within a newly generated set, nothing ranks one principle above another. Neither paper claims to settle what counts as a valid rule or which rule takes priority.

Definition (H. L. A. Hart, The Concept of Law, 1961)

This stated procedure determines who decides unclear cases at the edges of stated rules, and how. These are cases in which it is unclear whether and how a rule applies.

Analysis

Both methods install a repeatable, described procedure for cases that the rules do not clearly settle. Deliberative alignment trains the model to work through the spec before answering. SPRI runs a generate-critique-refine pipeline that fits a principle to the specific case. What is missing is the governance part. Nothing states a standard that the discretion must meet, no one outside can review or contest the resulting decision, and the decision never feeds back to narrow the penumbra (the zone of unclear cases) for the next case. SPRI goes further and removes the core/penumbra distinction entirely. With no rules set in advance, every case is an act of creating a principle rather than an edge case of a stated rule. The concept is claimed, because SPRI's stated motivation is that predefined principles are too generic to adapt to a particular input, and deliberative alignment advertises that it handles situations the spec does not spell out.

Definition (Lon L. Fuller, The Morality of Law, 1964)

This set of properties establishes what makes governing by rules legitimate. Rules must be public, clear, non-contradictory, stable over time, and possible to follow. They must also meet “congruence” (i.e., they must match how they are actually enforced).

Analysis

Congruence is the one item that the whole approach carries over. The text is consulted at decision time instead of being compiled into weights, so the rule applied is the rule as written, and the reasoning trace shows the fit. For deliberative alignment, the spec is also public, prospective, and general. However, nothing audits whether behavior matches the text, no check enforces non-contradiction between provisions, and stability across spec revisions is untested. For SPRI, the remaining Fuller items are missing. A principle that is written after the case arrives, for that case alone, and never published in advance is not general, prospective, or stable. Its congruence holds by construction, since the same system writes both the principle and the answer it governs. Because all items must be met and most are missing for half of the approach, the concept cannot count as preserved.

Definition (Joseph Raz, The Morality of Freedom, 1986)

This concept states that an authority’s rules deserve deference only if following them helps the governed act on reasons they already have. Authority is a service to the governed rather than a power over them.

Analysis

Neither method tests whether deference is justified by reasons that the governed already have. Deliberative alignment's spec is the lab's own document, addressed to the model. Users are never consulted about whether following it helps them act on their own reasons. SPRI fits its principles to the content of the query, not to the reasons of the person asking. Its stated motive is that principles written by experts are slow and expensive to produce. The user never sees the principle written for their case, let alone gets to say that it misread what they were asking. Neither paper asserts or implies that the authority earns deference by serving the governed.

The approach calls the model's private chain of reasoning "deliberation". In the traditions this taxonomy draws on, deliberation means several parties reasoning together with equal voice. Here, there is no second party at all.

Fuller's demand that rules be general is partly a concession to expense. Legislating is slow and costly, so a rule has to serve many cases to be worth writing. SPRI asks for minimal or no human effort and writes a principle for every query, so making a rule costs nothing at the point of use. The tradition has no account of what governance by rules becomes when the cost of making a rule falls to nothing. It never had to face the choice of whether to write a rule for a single case.