Feeder protection: fault slide

DRAFT, UNVALIDATED harness. Illustrative 12.47 kV feeder (a uniform conductor whose per-mile Z is the calculate_line_constants output for the harness test's inputs, 8 mi long, utility source 250 / 500 / 1000 MVA at X/R 10, relay at the source bus with no CT ratio): not any real feeder. Three-phase bolted faults only. The functions compute values and never judge; 'exceeds_withstand' / 'within_withstand' is this demo's own comparison (src 'demo-logic'). FAULT CURRENT: internal self-consistency only. Tested for internal self-consistency only — no independent reference value yet. `calculate_available_fault_current` (hand-calc tests recompute the same symmetrical-components formula with round input numbers — no independently published worked fault-current example was used) CABLE WITHSTAND, ICEA K factors UNVERIFIED: Elevated to a likely defect, NOT fixed, 2026-07-17 — an AI-assisted pre-review pass found the K factors below don't reconcile with independent published sources, by a source-independent check. For a 350 kcmil copper conductor at 0.5s, this module's formula gives ~40% higher withstand capacity than Southwire's published 90/250C copper figure — the non-conservative direction. Neither the K factors nor the citation are changed in this pass — this repo does not guess replacement values for a safety-relevant constant without a primary source. `withstand_amps`/`withstand_ka` should be treated as UNVERIFIED, not just "not yet reviewed," until a human confirms the K factors and citation against the actual ICEA text. The harness cable function takes no insulation class or temperature rating: This function's single per-material K silently assumes one unstated temperature pair for every cable of that material. RELAY CURVES: A conflicting secondary source exists for IEEE Moderately Inverse and Very Inverse's A/B constants — the two curves this file's own tiering already flags weakest. An AI-assisted pre-review pass found a source (industrialmonitordirect.com's curve table) giving materially different constants for these two curves (Moderately Inverse: A=0.0055, B=0.0254 vs. this module's A=0.0515, B=0.1140; Very Inverse: A=3.0457, B=0.1023 vs. this module's A=19.61, B=0.491) while agreeing with this module on Extremely Inverse and all three IEC curves. The same pass found other independent sources (an Eaton Digitrip TCC-curve datasheet, a third-party relay-calculator table) matching this module's values for all six curves. Net read from that pass: majority support for the current values, but a real, unresolved conflict on exactly the two curves already tiered "literature-only" — do not treat these two curves' constants as settled without a primary IEEE C37.112-1996 check. The IEC curves are independently corroborated at M=2/TMS=1 only; the IEEE extremely inverse curve is corroborated against a secondary web table. Axes: dist_mi (fault distance from the source bus), source_mva, curve (string), td (time dial: IEEE TDS or IEC TMS, one harness parameter), pickup_a. Main cols: i_a (fault current), m (multiple of pickup), trip_s, m_gt_ceiling (1 when m exceeds 18.5, the harness's own calibrated ceiling; the formula is still evaluated). Table cable: withstand_a at that cell's trip time, margin_a = i_a - withstand_a, exceeds (exporter comparison, demo-logic). Tables tcc (trip_s per curve, td and m; current = m x pickup is a plot scaling) and withstand_line (withstand_a per cable and time) draw the log-log chart. Cable areas are exporter inputs; the harness has no conductor catalog.