Undefined field access on an empty record COMPILES silently (reads garbage)

• Track: A (shared parser/symtab name resolution; Pascal-facing). Found 2026-07-18 night by fable-O — surfaced by the optfuzz reducer, whose body-stripped candidates kept compiling when they should not have.
• SILENT WRONG VALUE class — highest severity per the compat escape rule.

Repro (current master)

type
  TR0 = record
  end;
var
  r0g: TR0;
  x: longint;
begin
  x := r0g.someundefinedfield;   { compiles! x reads 0/garbage }
  writeln(x);
end.

pascal26 lax2.pas out → ok: (35KB binary), prints 0. FPC rejects with "identifier idents no member". An undefined VARIABLE errors correctly (undefined variable); it is the FIELD path that falls through — likely the member-resolution failure path defaults to offset 0 / silent-Integer instead of erroring (compare project_32bit_truthiness "unknown-type silent-Integer hole" — same disease, field flavor). Also observed in the same reduced file: a method (TC0.Calc) defined for a class that never DECLARED it, and an undefined local (v4) inside such a method body — verify both while in there; they may share the fallthrough.

Where to look

Field lookup (FindUField / record member resolution in parser.inc/symtab.inc) — the not-found branch. Gate: make test + self-host + the fuzz reducer's candidates now FAILING to compile (they became the accidental conformance probe).

Log

• 2026-07-19 — resolved, commit 4d46a7ad.