nx_spendgate.nx

nx_spendgate.nx -- ASK BEFORE YOU SPEND. The gate between "I should build X" and building X. WHY IT EXISTS. On 2026-08-07 I built `nx_strdiff` when `nx_contentdiff` already existed and did the same job better, wrote `ss_khash` when `nx_web_shard_compact` already carried the identical hash dedup, and ran a 649-row `nx_stalecensus` that DUPLICATED a sibling seat's completed run -- the third of those saturated the mgmt API for ~25 minutes and bought nothing. Every one of those was preventable by a single query against tooling that already existed and worked. ★★★★★★"I DIDN'T KNOW IT EXISTED" IS A RETRIEVAL FAILURE, NOT A DISCOVERY. A line in a doc did not stop me. An exit code is branched on; an advisory is skipped. COMPOSES THE INCUMBENTS, never reimplements them (building a fourth duplicate to prevent duplicates would be its own punchline): nx_presubmit <name> -- is the NAME taken or dual-copied? MECHANICAL, decidable nx_capsearch <intent words> -- what already does something like this? RANKED, NOT decidable ⛔WHY THERE IS NO SIMILARITY THRESHOLD, AND WHY ADDING ONE WOULD BE A LIE. Measured over four queries whose ground truth I knew, because I had just lived each one: DUPLICATE (nx_contentdiff existed) top=nx_route_diff score=499998 NOVEL (fixture ratchet) top=nx_govern_sweep score=421048 DUPLICATE (nx_capsearch existed) top=nishi_search score=470584 NOVEL (no pre-spend gate) top=nx_adnet_recbeat score=500000 <-- HIGHEST OF ALL FOUR The highest score in the set belongs to a NOVEL case, and in BOTH duplicate cases the true incumbent ranked SECOND, not first. Any cutoff would have blocked the novel work and waved the duplicates through -- the precise inversion of the tool's purpose. ★★★★★★A RULER THAT SCORES A NOVEL THING HIGHEST OF ALL CANNOT BE THRESHOLDED -- PUBLISH THE LIST, NOT A VERDICT. ★★★★★TOP-1 IS NOT THE ANSWER; THE LIST IS. So this gate automates ONLY what is mechanically decidable and REFUSES to synthesise the rest. It prints the candidates and makes you look. That is the honest maximum. It does NOT create anything. A gate that performs the act it authorises cannot be re-run to audit its own decision -- the same rule nx_adoptgate states for promotion. nx_spendgate <proposed-name> <intent words...> -> SPEND-GATE name=<n> presubmit=FREE|TAKEN verdict=REVIEW|BLOCKED exit 0 REVIEW (name is free; READ the candidates before spending) | 1 BLOCKED (name taken) | 3 usage license_tier: ORIGINAL No hw writes (Rule 26). expect_exit: 0