nx_barfresh_lib.nx

nx_barfresh_lib.nx -- THE ONE RULER FOR BAR AGE ON A /compare BOARD: a bar is the CURRENT MONTH'S, or it is refused. Operator 2026-09-06, verbatim: "make sure its a hard fail to use anything less than current month unless there is nothing current month for these what does x month of x year say is sota" and "this needs to become the way that the logic forces you from the /compare domain and nishi estate as youve invested significantly in 12 year old tech". Measured cost the same day: the intelmine ladder (a tagger, a parser, a clusterer, a CRF, an embedding reader) climbed for a day toward the SemEval-2014 winners' F1 because that paper was the one source the domain had mirrored and the bar carried no month. Nothing in the estate asked how old the bar was: nx_capsearch and nx_spendgate guard ORGAN duplication, nx_sota_status guards CLAIM binding, and bar AGE had no ruler. This lib is that ruler. It is composed by nx_compare_rank, so a stale bar refuses the board's work queue instead of being remembered by a seat. THE DATA (rows in <domain>.plan, pipe-separated, NO pipe inside a field): bar|<id>|<YYYY-MM>|<subject>|<what that month says is the state of the art>|<ref key>|<note> the source MONTH of the bar and the .refs key of the listing it was read from (a mirror of the bytes read) barscan|<YYYY-MM-DD>|<ref key>|newest=<YYYY-MM>|<note> an ATTESTATION: on that date the listing at <ref key> was fetched and read, and its newest entry was in <newest>. This is the only way a bar older than the current month survives: the current month was LOOKED FOR and had nothing. THE STATES, per bar, against the current month (derived from the clock, or handed in by a gate): FRESH the bar's month IS the current month ATTESTED older, and a barscan dated THIS month for the SAME ref reports newest == the bar's month: the fallback is named, "what does <that month> say is sota" is exactly what the row carries STALE older, and a barscan dated this month for the same ref reports a NEWER month than the bar cites: the bar ignores evidence the board already holds UNATTESTED older, and no barscan dated this month for its ref: nobody looked this month. "Nothing newer" is a claim that must be re-earned every month, so this is the hard fail the operator asked for MALFORMED a month that does not parse, or a month after the current one DUPLICATE a sotabar row whose id an EARLIER sotabar row on the same plan already carries: one declaration made twice. MEASURED 2026-09-14 on search.plan: a whole intake landed twice across a transport drop (two bars, two scans, two targets, one rung id reused) and only the rung id was refused, by the plan-graph rule; two rows for one id hand every reader a coin flip, so this ruler refuses them the same way. AND PER ATTESTATION: a barscan row repeating an earlier row's date AND ref is a DUPLICATE SCAN, counted (scans_dup) and refused like a duplicate bar; two scans of one listing on DIFFERENT days are two attestations, not duplicates. THE VERDICT: FRESH (exit 0) when every bar is FRESH or ATTESTED; STALE (exit 1) when any bar is STALE, UNATTESTED, MALFORMED or DUPLICATE or any scan is a duplicate (bf_refusal_rule names WHICH, so a consumer prints the rule from this lib instead of paraphrasing it); NO-BAR (exit 3) when the plan declares no bar rows. A domain opts in by declaring its first bar; a domain that has not is LABELED by its consumers, never acquitted and never refused (a fleet-wide refusal on day one would be the permanently-red detector everyone learns to ignore). The partition bars = fresh + attested + stale + unattested + malformed is printed by every consumer and must sum. COMPOSES, never re-implements: sys_read_file (whole file, cannot short-read), sys_clock_gettime_real (the clock), civil_from_days (nx_timefmt, Hinnant) for the month of an epoch, nx_comparetree_lib for the two-tree plan path.