nx_debt_recurrence_lib.nx

nx_debt_recurrence_lib.nx -- CE3 (codeeffectiveness ce_recurrence): RECURRENCE PER DEFECT CLASS over the debt plane. Resolution effectiveness is not a close count: the estate's own record has one defect filed five times in three weeks. This is the ONE ruler for that class, composed by nx_debtmine (the incumbent bugs-class miner, which already held per-scope eat-rate and aging) and driven in-process by its gate. Over the plane rows (id TAB sev TAB scope TAB status TAB desc) it interns scopes and, per scope, records the FIRST EATEN filing as the prior and counts every row filed AFTER it as a recurrence of the class: recur_total (history, non-decreasing by construction) and recur_open (the live re-filings, which FALL when they are eaten). The ids are named -- the prior id and a bounded prefix of the recurring ids beside their total -- because a bare number is not the measure the rung asked for. CLOSE TIME IS UNRECORDED ON THIS PLANE (ids are file-time), so a row filed after a prior that was eaten later still counts: the count is an UPPER BOUND and says so in the document. A run with no rows writes no spine row; the spine path is derived from the source (the production spine iff the source is the live plane, else a <source>.recurrence.spine sidecar), so a fixture can never forge the production trajectory. The scope table has NO PICKED BOUND: its capacity is drc_capacity(q, n), derived from the rows, so scopes_capped is 0 by construction and stays in the document as an invariant the gate asserts (the first live beat, 2026-09-06, hit a 256-slot bound on a 4,353-row plane and published a floor wearing a total). Ids are epoch-seconds, so an autofile burst shares one id across several rows: those are distinct filings and count as such, and the adjacent-burst count is published beside the ids (adjacent only, a declared floor) so a reader knows why an id can repeat. The row walk mirrors nx_debtmine's inline loop (sj_le / sj_col / comment skip); extraction of that walk into a shared reader is the declared follow-up, not a second parser by intent. license_tier: ORIGINAL No hw writes.