Enhancement-634: the diagnostic findings of the 2026-09-12 hunt, D1–D24

Enhancement-634: the diagnostic findings of the 2026-09-12 hunt, D1–D24

Scope: twenty-two findings of the
2026-09-12 hunt's
diagnostic table, each a place where the simulator or the compiler did something
defensible and said nothing, or said the wrong thing. Nothing binds, draws or solves
differently; what changes is what is said, four refusals, and the record's labels.
Simulator: src/osdi/osdisetup.c (D1, D10), src/frontend/inp.c (D2, D6),
src/frontend/mcsave.c (D3, D7, D12, D19), src/frontend/com_sweep.c (D3, D4, D14),
src/spicelib/parser/inp2n.c (D8, D13), src/spicelib/parser/inpdpar.c (D9),
src/frontend/outitf.c + parse.c (D11), src/maths/misc/randnumb.c (D19). Compiler:
openvaf/basedb/src/ast_id_map.rs, openvaf/hir/src/{attributes,lib}.rs,
openvaf/sim_back/src/module_info.rs (D5, D21). Documented: D11's multiplier, D15,
D16, D17, D18, D20 (the statistics guide §7.3, §7.4, the handbook §3.2). Already held:
D24 (E-622's suffix rule matches @x1.n1[dr]). New suite
hunt12diag_examples (20 checks per solver).
Both sides.

Suites: hunt12diag_examples 20 of 20 per solver, both solvers; savemc, writemc,
osdimc, autoadapt, adaptquiet, autoopts, highsigma, mcpolicy, autobuskicad,
silentaccept, osdidist, wcd green; full sweep 507 of 507.

The findings, and what is done about each

#	was	now
D1	mcseed=4294967297 "is not an integer; using 2147483647" — an integer past int, and every seed ≥ 2³¹ aliased onto one ensemble; mcseed=-3 taken in silence	the seed is a 32-bit pattern, 0 … 4294967295: a value outside it is refused and said (seed 1); 4294967295 is a seed of its own; 1.5 is floored and said
D2	set controlswait under ngspice -b: the commands behind it ran at once, before the .op, with nothing said	a note: the wait is for a host's run under libngspice; here there is none and the commands run now
D3	writemc trial=9 analysis=2 added a second trial and analysis column	trial, analysis, status are refused as the row's fixed columns, in writemc and in montecarlo -writemc
D4	highsigma on a deck whose only statistics are uniform printed "scale = 2" and an "equivalent sigma" while running plain MC	when every weight is exactly 1 the report says nothing was inflated — a uniform has no sigma to scale — and that the figures are a plain estimate
D5	(compiler) (* std=25.0, std=30.0 *), (* trunc=3, trunc=1 *) took the last in silence	a warning names the attribute, the count and the one used (the last)
D6	a second .option savemc= card, and automc_save= beside savemc=, dropped without a word	any .option given more than once with different values is warned (the first card's is used); automc_save/osdimc_save beside savemc is warned once (savemc used)
D7	writemc after a run that made no row said "no analysis has run yet"	"the last run made no row — this circuit has no parameter with statistics to record"
D8	.option autoadapt without adapter= printed "Error: … needs an adapter model" and ran anyway	a Warning that says the option is ignored (the same for autoadapt without autobus)
D9	a model parameter on the instance line: "unknown parameter (r)"	"… it is a model parameter of this device — set it on the .model card, not on the instance line"
D10	altermod rm r=0 onto from (0:inf) was accepted; the next trial's failure blamed the draw ("value −5.11649")	the recentre is warned where it happens, naming the range (the text E-558 exports, read for numeric bounds and sets)
D11	the OSDI flow vector n1#flow(p,n) was typed voltage and its name had to be quoted in expressions	typed current; the lexer reads the parenthesised name; the per-unit/×m difference against @n1[i_p] is documented
D12	the txt writer put nan in a cell a row never got	empty, as csv and xlsx
D13	a 4-bit bus into a 2-bit port bound the low two bits in silence	pass 3 names the bits of the base beyond the port's width (/mid_2_, /mid_3_)
D14	a setseed 5 in effect was overridden by the loop's default seed 1, never said	the seed note adds that the setseed 5 is set aside, and that -seed 5 draws from it
D15	(doc) resume of a completed analysis re-runs it as a new trial	§7.3: the contract is for an interrupted run
D16	(doc) rows per sweep depend on the path	§7.4: the device path is one op row per point, the model-parameter path one dc row
D17	(doc) a host that loads the circuit twice restarts the trial counter; the analysis column says run	§7.4
D18	(doc) a .param derived from a mvnorm() one is recorded twice	§7.4: one shared draw, not inlined, recorded under its name and under every slot that reads it
D19	wcd's FORM iteration rows were indistinguishable from samples	labelled op (wcd probe) — the nominal point, the probes and the line searches, keyed on the search's own mode
D20	(doc) set mcseed=9 after .option mcseed=7 wins for the next trial	§7.3
D21	(compiler) (* std=0.5 *) on from {1.0, 2.0, 3.0} compiled without a word and nearly every trial failed the range check	a warning: a draw around a member lands off the set; declare a continuous range or drop the statistics
D24	-inflate @x1.n1[dr] "matched nothing"	held since E-622: the owner is matched on a .-suffix, so the print spelling matches the internal n.x1.n1

Verification

check	result
D1 seeds 4294967297, −3, 1.5, 4294967295, 3	the messages; 4294967297 and −3 draw as seed 1, 4294967295 and 3 each their own
D2 … D4, D6 … D14, D19	each message or label as in the table, on a deck that provokes it
D5, D21	the compiler's warnings; the object carries std 30 and trunc 1 (the last)
D24	no "matched nothing" for -inflate @x1.n1[dr]
the twelve suites named above	unchanged
hunt12diag_examples	20 / 20, both solvers
full sweep	507 of 507

Write-up: enhancements_doc/Enhancement-634.md