nx_research_fetch_cli.nx

nx_research_fetch_cli.nx -- the RUNNABLE DRIVER for rf_fetch_bank. WHY THIS EXISTS: runtime/nx_research_engine.nx carries a complete `rf_fetch_bank(url, name, store, out, cap)` -- the exact symbol knowledge/compare/deepresearch.matrix names as grounding its "Web fetch and bank pipeline" claim -- but the file's own entry point is `func main() -> i64 { return 0 }`, a COMPILE SMOKE. So the capability is real, gated, and grounded, and yet cannot be RUN: no promoted binary anywhere exposes a fetch+bank verb. Same shape as nx_https_get (a smoke main that needed a real CLI beside it) and the ~dozen other build-what-already-exists cases banked this session. ★★★★★A LIBRARY FUNCTION NOBODY CAN INVOKE IS A CAPABILITY ON PAPER: the claim grounds, the gate passes, and the world is never actually touched. CONSEQUENCE THIS UNBLOCKS: deepresearch holds quorum + provenance + freshness + non-vacuity (redseen 1/1) and is short exactly ONE method class. An EXPERIENTIAL row supplies both that class and the require_human leg at once -- but only if the pipeline can be RUN against a real third-party server, because a row whose `observed=` was never observed is prose. This driver is what makes that row honest. nx_research_fetch_cli <url> <name> Fetches over our own TLS + Mozilla trust store, banks raw -> knowledge/fetched/<name>.raw and text -> knowledge/library/<name>.txt, and prints a MACHINE-READABLE observation line for the attestation: RFETCH url=<url> name=<name> rc=<n> banked_bytes=<n> verdict=OK|BANKED-SKIP|FAIL ★EXIT CODE IS THE VERDICT (0 fetched+banked, 1 already banked, 2 usage, 3 trust store, 4 fetch failed) so a caller never has to parse prose -- and, per this session's own law, two different failures never share a code. expect_exit: 0 license_tier: ORIGINAL No hw writes (Rule 26).