{"path":"research/wild-weighing-dialects-and-the-question-set.md","content":"# Wild Weighing Dialects, and the Four Questions They Added to the Descent\n\n**Date**: 2026-08-25 · **Type**: adjudication-review analysis (the blind-label corpus read as a requirements document) + a founder ratification. **Status**: the four questions are BLESSED and SHIPPED (weighing question set 10 → 14, `deliberus/weighing.py`); the Chang small-improvement probe was explicitly held back — the founder is pondering it separately, and nothing here presumes its adoption.\n\n## Where this came from\n\nThe blind ground-truth labeling of 72 wild Swedish election sentences ([weighing-eval/ground-truth-public.md](weighing-eval/ground-truth-public.md)) produced ten hard cases — four committed borderlines and six wide-margin negatives — and reflecting on *what kind of weighing each could be* turned the case list into a species taxonomy. Checking each species against the existing ten questions showed which wild forms the descent could already interrogate and which it could not. The gap analysis became four new questions. This is the induction provenance, stated per the three-runs law: **one register (campaign politics) induced these; expect the next register to break the set again.**\n\n## The species found in the wild (none of them the textbook form)\n\nAlmost nothing in 72 wild sentences was \"X väger tyngre än Y.\" The forms that actually occur:\n\n- **Agenda-weighing** (the poster's \"the question is not just whether... the question is what\") — ranking *questions* by decision-relevance rather than values by weight; the suppressed \"my axis dominates\" premise half-said. The form that carries every framing battle.\n- **Proportionality weighing** (the bridger's \"played up the differences too much and too long\") — second-order: a judgment that the weight *given* to something exceeded its proper level, presupposing an unstated calibration.\n- **Deontic comparative** (the Green op-ed's \"must take greater responsibility\") — obligation measured against the status quo with the counterweight (cost, capacity) deleted; half a scale with one pan loaded.\n- **Urgency weighing, both signs** (\"we cannot afford to wait\" / the train-timetable mockery of making something an issue *now*) — temporal weighings; the same axis with opposite signs, so admitting one admits both.\n- **Betrayed-priority accusation** (\"a betrayal of all crime victims\") — a weighing attributed to the opponent's *failure*; stance toward a weighing rather than a weighing asserted.\n- **Triage under scarcity** (\"prisons are overfull, priorities are required, therefore begin with sexual offenses\") — an explicit rank-ordering conditional on a resource constraint; the compression-check question probes whether it survives the scarcity vanishing.\n- **Depth-ordering of evaluations** (\"dysfunctional, legally insecure — and fundamentally inhumane\") — ranking one's own criticisms; climax wearing a ranking's syntax.\n- **Enthymematic weighing** (\"preventing family reunification creates insecurity, hampers integration, damages mental health\") — three causal claims whose argumentative *force* is a weighing completed entirely by the reader; run 3F's implicit-crux shape at sentence scale.\n- **Anti-weighing / Pareto claim** (\"stronger action against welfare fraud could fund higher benefits without any tax change\") — the claim that an assumed tradeoff is escapable, with a smuggled deservingness ranking (\"those who *truly* need help\") inside the no-tradeoff move.\n- **The inert inputs** (trend facts, cited numbers, capacity facts) — not weighings but the premises weighings consume one sentence later; a detector must leave them alone even adjacent to the weighing they feed (proximity contamination as a named failure mode).\n\n## The gap analysis, and the four blessed questions\n\nCovered by the existing ten: triage (compression check), risk posture, scope/deservingness, state-deficit, satisfier-belief. Not covered — and now shipped as questions 10–13 (the typing question stays last):\n\n1. **The counterweight question** — *\"What is on the other side of the scale — what is being given up or outweighed, and is it named?\"* The wild weighing's signature failure is not bad weighing but a **scale with one pan deleted** (deontic comparative, enthymematic form, and the corpus's most repeated finding: the crux stated on one side, implicit on the other). This operationalizes the load-bearing-unsaid finding at the weighing layer — it forces the enthymeme's completion as a tap.\n2. **The temporal-scope question** — *\"Is this a priority for now or for always — does the weight change if the time horizon shifts, or if the window of reversibility closes?\"* The descent's first temporal question; two of ten wild cases were purely temporal. The weighing layer's piece of the temporal rung.\n3. **The option-set question** — *\"Is the tradeoff real — is there an option that serves both considerations, making the weighing unnecessary?\"* Catches anti-weighings and false dilemmas; Fairclough's deliberation scheme (test alternative means before weighing goals) is the academic warrant. Distinct from specification (which qualifies norms); this finds a third option. A false dilemma should die before it is priced.\n4. **The level question** — *\"Is this a weighing of the considerations themselves, of which question deserves attention, or of someone else's weighing?\"* Three of the four borderlines were META-weighings (agenda-ranking; proportionality; betrayed priority). A meta-weighing's descent looks different — its covering value is decision-relevance — so it re-levels before the first-order questions are pressed. The discriminator test (supply the suppressed premises and watch what conflicts: topic-drift / unstated-crux / unsurfaced-weighing) is this question mechanized.\n\n**Held back, founder pondering separately**: wiring Chang's small-improvement probe into the typing question (perturb one option slightly; if the improved option still fails to beat the rival, equality is refuted and parity is on the table — the stress test called it mechanizable). Not adopted; not to be presumed.\n\n## Why fourteen questions no longer spam anyone\n\nThe worth-asking layer prices every question by whether a decisive answer could move the badge across a display band; below-the-line questions collapse to a state line. Hinge-ordering surfaces whichever question actually bites for *this* claim. So the marginal display cost of a richer set is near zero — the set can grow toward completeness while the surface stays one-question-at-a-time.\n\n## Cross-references\n\n[weighing-eval/ground-truth-public.md](weighing-eval/ground-truth-public.md) (the blind corpus these were induced from) · [weighing-scheme-and-terminus-classifier-design.md](weighing-scheme-and-terminus-classifier-design.md) (the descent's design lineage; the 2026-08-25 addenda) · [the-load-bearing-unsaid.md](the-load-bearing-unsaid.md) (the counterweight question's warrant) · [election-2026-bloc-debate-cluster.md](election-2026-bloc-debate-cluster.md) (the thread the borderlines came from; the discriminator test) · [taxonomy-gaps-and-the-closed-enum.md](taxonomy-gaps-and-the-closed-enum.md) (the three-runs law this set's provenance note obeys) · [philosophical-foundations-stress-test.md](philosophical-foundations-stress-test.md) §4 (the small-improvement probe, held back)\n"}