Least-worst: choosing what to sacrifice

In Meridian's R3m scenario, mines block the strait. No plan can satisfy all four commitments. A conventional system would either fail silently — returning no result — or rank by a weighted score that hides which commitment was traded away and why.

ASSAY does neither. It turns infeasibility into an argument surface the commander weighs: here is what each sacrifice costs, here is what it buys, and here is the tie_break convention that ordered same-tier alternatives — because encoding a civil-harm-vs-force-protection ranking is exactly the value judgement ASSAY-DEC-19 forbids.

The three least-worst candidates

The relaxation service generates R3m-responsive candidates that enter the mined water, scores each through the existing SPEC-07 scorer (reused, never re-implemented), keeps only the inclusion-minimal sacrifice sets, and ranks them least-worst first by ordinal tier cost — musts last, never dropped.

Candidate 1 — sacrifice C4 (capacity)

sacrificed: C4 tier: prefer C1 robust C2 robust C3 robust

Route transits the mined channel with reduced capacity. The prefer-tier sacrifice trades tonnage for clearance — the cheapest relaxation in the commander's priority order.

Candidate 2 — sacrifice C3 (clearance)

sacrificed: C3 tier: should C1 robust C2 robust C4 robust

Route holds capacity but accepts reduced clearance. A should-tier sacrifice — more costly than C4, but preserving the logistics commitment the force needs.

Candidate 3 — sacrifice C2 (width)

sacrificed: C2 tier: must C1 robust C3 robust C4 robust

Route narrowing through the strait. The must-tier sacrifice — ranked last, never dropped. The commander sees it because silence on a must-sacrifice is more dangerous than showing it (G4).

Illustrative widget, fixture data. The cards show the three inclusion-minimal candidates the relaxation service produces for Meridian R3m. The ranking, sacrifice tags, and tier labels follow the real s3Cards component's rendering. Frozen as shipped on 2026-07-13.

Least-worst, never silence

Three design constraints, each load-bearing. Inclusion-minimal: a strict-superset sacrifice (e.g. {C2, C4}) is silent waste and is dropped — the commander does not need to see dominated options. Least-worst first: prefer-tier sacrifices before should-tier, should-tier before must — no weights (DEC-19), just the ordinal priority the commitment itself declares. Never empty: the report always contains at least one candidate (G4), and the must-sacrifice is always present, never hidden behind a ranking threshold.

Assessment, not fact — and demonstrator, not product. Every number here renders the Meridian Archipelago, an engineered fiction (ASSAY-DEC-8). Nothing on this page reflects any real operational picture. A thesis in ASSAY is only ever explored, never claimed proven.

Sources & trace

Relaxation service — src/relax.ts · src/tiers.ts
Cards component — src/components/s3Cards.ts
Research note — 04-relaxation.md (the lexicographic goal-programming decision)
No-weight rule — ASSAY-DEC-19 in the decision register
Fixtures — R3m (mined-water scenario) in the Meridian vignette