nx_srcreach.nx

nx_srcreach.nx -- THE SOURCE REGISTRY READER AND THE PACE BUDGET, for /compare/researchreach rungs RR1 and RR3. Operator 2026-09-06: "we need to measure and make sure we never have any unreachable data whether by search or api we need to be able to gather it all" and "a regular cadence for all domains without smashing our nas resources". nx_srcreach load <conf> -- read and classify the registry, print the partition nx_srcreach plan <conf> <reqs_per_source> <max_requests> -- ALSO print the request budget, and REFUSE over the ceiling WHY plan PRINTS BEFORE IT WOULD FETCH: a sweep that cannot state its own request count is a load incident waiting to happen. The budget is computed and announced with NO network touched, so the decision to go wide is made on an arithmetic a person can check rather than discovered from the array afterwards. RR2 LANDED 2026-09-06 AND THE PROBE IS REAL. An earlier draft of this header explained why the probe entry point was deliberately absent; that note is retired because the capability now exists rather than the name. It COMPOSES the estate's fetcher over sovereign TLS instead of re-implementing a client, probes into ONE fixed scratch name so a forty-source sweep leaves one file rather than forty mirrors, paces per host, and reports a partition over EVERY row. A HISTORICAL NOTE WORTH KEEPING: while that paragraph existed, nx_domain_admit reported rule-disagree wordboundary=1 declared=0 against this very file -- the naive scan found the symbol inside the comment explaining its absence, which is the estate's own prose-is-source-bytes trap firing on the organ that documents it. BITE-PROVEN LOCALLY 2026-09-06 on a fixture assembled at runtime with FOUR planted defects: each fired, each was NAMED by its key, and the partition summed (rows=6 ok=2 bad_fields=1 bad_kind=1 bad_proto=1 bad_pace=1 sum=6, verdict=REFUSED exit=1). The POSITIVE CONTROL is the one that matters most: a clean three-row registry read rows=3 ok=3 sum=3 verdict=OK exit=0, so this is not a guard that refuses everything and passes every absence test. The budget was proven in both directions too: 3 sources x 10 requests = 30 planned, ceiling 100 -> OK exit 0, ceiling 20 -> REFUSED-OVER-CEILING exit 1 with nothing fetched. exit: 0 OK | 1 REFUSED (a bad row, a partition that does not reconcile, or a budget over the ceiling) | 2 usage | 3 NO-REGISTRY (unreadable or no src rows) license_tier: ORIGINAL No hw writes (Rule 26).