analog M5d: nand, nor, tgate and tracking-switch rows of the external team's checks computed from the PSP103 tables, cross-checked with ngspice #11263

Part of the analog epic (Refs #10486). After M5c (#11212, PR #11215) the external team's per-block checks of 2026-10-16 run as a .t27 evaluator on our models: 3 PASS, 1 FAIL, 9 NOT MODELLED of 13 blocks. A row we do not compute is never a pass, so the NOT MODELLED rows are the largest gap to the v0.1 acceptance. This slice computes the rows our existing tables can already answer. Only .t27 (skill only-t27).

Requirements

• FR-001: The nand and nor output levels (a 1 within 1.45 to 1.55 V, a 0 within 50 mV of 0 V, into 10 fF) MUST be computed from the tt PSP103 tables (mos_psp103.t27 and its data) by a DC solve of the two-input gate, for all four input codes.
• FR-002: The transmission gate's Ron at 0.75 V, its worst Ron over 0.1 to 1.4 V and its off leakage MUST be computed from the same tables (the leakage through the log-domain weak-inversion read of analog M3c-2: a log-domain PSP103 current read with an integer exp meets the 1 percent bar in weak inversion (worst cell 0.705 percent), inverter delays within 0.025 percent of ngspice #11146).
• FR-003: The tracking switch rows (tracking error at 0.3 V and 1.2 V into 2 pF, the hold step, the droop) MUST be computed from the bootstrapped switch model (M4c) where it holds the needed quantity; a row it cannot answer stays NOT MODELLED and the reason is written next to it.
• FR-004: Every newly computed value MUST be checked against our own ngspice-47 run of the same cell on the reference host (container jobs, at most 6 of 12 cores, files removed afterwards) and the difference written in the report; a value without an ngspice cross-check MUST NOT be counted as PASS.
• FR-005: The report document MUST give the new totals (PASS, FAIL, NOT MODELLED of 13) and list each row with its measured number and limit.

User Scenarios

• Given the nand inputs 11, when the DC solve runs on the tt tables, then the output is reported in volts and judged against the 0 V +- 50 mV limit.
• Given the tgate at 0.75 V, when Ron is computed, then it is judged against 30 kohm and the ngspice value of the same cell is printed next to it.
• Given a row the models cannot answer, when the evaluator runs, then it stays NOT MODELLED with its reason.

Boundary

• specs/analog/accept_sar.t27
• specs/analog/
• docs/analog/accept-sar-2026-10-20.md

Success Criteria

• t27c test-report specs/analog/accept_sar.t27 prints FAIL 0 and vacuous passes 0
• the NOT MODELLED count in docs/analog/accept-sar-2026-10-20.md drops from 9 to at most 6 of 13
• every newly computed row has an ngspice difference printed, in mV or percent