nx_compare_rank_refresh_test_20260907.nx

nx_compare_rank.nx -- THE ROADMAP RANKER: priority computed from the boards, never chosen by a seat. Operator 2026-08-18: "we need sota best of breed prioritization and ranking from the foundation first bit up in our compares to drive our roadmaps ... it needs to push back on your random doing of low priorities" and "we cant keep having you do random shit and say its effective". A seat picking rungs by whatever is cheapest with parts in-tree is the defect; this organ makes the compare COMPUTE the order from data the boards already carry and PUBLISH it where every seat, MCP caller and RACI row reads it. It decides nothing itself: the math is nx_dr_ocm (DR-6 opportunity-cost ranker, Priority=(V*M)/C, integer deterministic -- REGISTERED-DARK until today, adopted here) and the plane write is nx_store_put; this organ only DERIVES the three inputs per open rung, all from data: THE MATHEMATICS (v5, 2026-08-18 -- operator: "a full mathematical system on opportunity cost, clustering, foundation-up capabilities, best-of-breed research as of August 2026", not a seat's coefficients). Every term is COUNTED or a RATIO of board data; the only chosen constant left is RANK_COST_UNKNOWN and it fires only on a plan row that declared no cost (a labeled data defect). value v = deficit + CoD + option + enables (all from matrix codes + dep rows) deficit = sum of rival codes on the rung's row (2 leads, 1 present, 0 absent): value at stake CoD = deficit x (rivals LEADING / rivals total): WSJF cost-of-delay -- the fraction of the field already ahead is the time-criticality; every unit open, that many rivals widen it option = 1, ONLY when every rival is absent: the lead itself, held by nobody -- its unlocking value is already the enables leg below (one leg each; a v5 draft double-charged it) enables = number of OTHER open rungs whose transitive dependency closure contains this rung: the foundation-first / opportunity-enablement leg (WSJF) and the compound-option leg (real-options portfolio theory), counted from dep rows momentum m = 1 + declared dependencies already MEASURED present (substrate landed; never 0 because nx_dr_ocm multiplies and a zero would hide the rung) cost c = the plan's own est u in tenths, DIVIDED by the rung's cluster size (open rungs sharing the same organ file share substrate: read once, one build lane, one gate to extend -- graph clustering on the organ column, measured never labelled); unestimated -> labeled sponsor = the operator's sponsor- plane coefficient (PR6), applied to value, every firing row named priority = nx_dr_ocm( v x sponsor, m, c ) = (V x M x 1000)/C with forgone-best-alternative and budget-aware greedy selection (DR-6, integer deterministic) critical path: the FIRST unmet product version's milestone rungs + transitive open deps rank before everything else (ver| rows) -- staged delivery 0.1 -> 1.0 -> 2.0. Provenance: WSJF/cost-of-delay (SAFe LPM), real-options R&D portfolio valuation with project interdependencies (Blau et al.; OptFolio-class DSS), integer-programming portfolio selection under budget (knapsack form -- dr_ocm's greedy is the bounded-effort stand-in), dependency-graph enablement. pmdash.refs carries the citations. What is NOT here, honestly: uncertainty distributions (PTRS-style success probabilities per rung) -- the estate has no measured per-rung success rate yet; PR7's actuals-write-back is the data that makes that term possible, and it will be added when the data