The bodyless-type exposed subset is four names of 66, 1045 of it Symbol — and a nullary alias rendered with an applied type argument

[gunbai-bot]

The headline is four names, not sixty-nine

#9879 filed the third inhabitance seam — any bodyless type X admits any argument in parameter position, silently — and stated that the exposed subset was smaller than the syntactic population and unmeasured. This measures it.

4 of 66 distinct names are reachable in parameter position, where a non-inhabiting value can actually flow in:

name	module	parameter sites
Symbol	v2.std.node	1045
SourceSpan	v2.lens.application	129
Bytes	std.types	4
ConfigPatchRecord	v2.std.patch	1

33 names appear only in field position. 27 appear in no parameter, field, or return position at all. So the class is small and its exposure is dominated by a single declaration — Symbol, the substrate's own identity field — by an order of magnitude over everything else combined. A census reported as "69 exposed positions" would bury that.

The two censuses join at identity, not at count

Re-derived independently at origin/main 6741a3953, per #9879's own instruction to re-derive rather than quote either number: 69 declarations / 66 distinct names, which agrees with that row's 69 exactly.

The reconciliation is the part worth keeping. A first pass matching ^type X$ found 62. The missing 7 are parameterized bodyless declarations — Compose<carrier, dimension>, MachineWidth<bits> (twice), ReferenceModel<T>, CppTargetDataModel<…>, CppTargetProfile<…>, ConfigPatchRecord<Config>. 62 + 7 = 69, so the two lanes agree on membership, not merely on a total.

Compose and MachineWidth turn out to be unexposed: type constructors applied as type arguments, never parameter types. Adding all 7 to the population moves the exposure count by one.

The naive count reproduces the broken instrument. Matching ^type X$ without reading the following line returns 5194 — within 1.5% of tools.opaque_realization_census's 5269 — because 5132 declarations carry = … or | … on the next line. I hit this before catching it by reading the hits, which is independent corroboration of that instrument's defect by accident rather than by design. Repairing the instrument is a separate PR (different subject; the prefix-boundary bug ranks first since it survives a fix of the other).

The ceiling is not one value

Reporting a single ceiling would overstate the row, because the population splits and the parts differ:

• (a) Declared-abstract by ruling — Symbol, Unit. The attainable ceiling is not structural impossibility via arms; it is a construction discipline at the boundary, values entering only through a declared minting authority. symbol_intern_lexeme is such an authority, which is why XL-0-INFER: classify and repair inference-failure survivors #9854's repair of target_value_expression_char_unicode_scalar has the right shape where minting a host String via from_code_point did not. R3 at sites that go through one, R2 corpus-wide.
• (b) The 27 type-level tags (cpp_abi widths, std.measure rate units, std.constructors' Product/Coproduct/Phantom). Nullary markers used as type arguments, not value carriers. Hole unreachable, honest ceiling no climb needed — filing them as exposure would inflate the class and bury the one member that matters.
• (c) SourceSpan and Bytes — genuinely exposed, unruled, recorded so they stay countable rather than forgotten.

Class rung is the minimum across parts: below floor. The trigger names a capability, per §4b(3) — a bodyless declared type can be inhabited only through a declared minting authority, and the direct-call inhabitance judgment consults that authority. #9854's witness char_unicode_scalar_expression_carries_a_symbol_identity is a wall for one site and does not retire this row; the invalid state stays writable at the other 1044 Symbol parameter sites.

Method limits, carried in the row rather than left to be discovered: parameter sites are counted only on lines beginning fn /func , so multi-line signatures are undercounted; the field-position figure uses a looser match and may over-count. Both are bounds. The shape — four exposed names, Symbol dominating — is robust to both errors.

Second change: a new class

A nullary alias rendered with its body's type argument applied to it. Two instances, different aliases, one shape:

• type LexText = FreeMonoid<Char> is nullary, yet a refusal at a LexText-declared parameter printed expected 'Node(LexText<Product(Char)>)'.
• A repair to is_host_text_carrier_type emitted pub fn string_is_empty(s: String<i64>) into v2_std_text.rs — not valid Rust.

The harm is not the malformed spelling; it is that the malformed type reads as a real refusal. The LexText arm produced 129 refusals across three files where the qualified-spelling arm produced 30, and that 129 was reported and endorsed as a stricter wall over a larger population before the diagnostic text was read closely enough to notice the alias cannot take an argument. A population inflated by a printer defect is worse than a missing measurement, because it gets quoted.

It is distinct from the neighbouring non-confluence rows: those are two authorities disagreeing in two places; this is two decisions inside one rendered type expression — base name from one authority, type argument from another — which is why it survives every emitter-vs-resolver question currently open. It reaches both a resolve-stage diagnostic and an emitted Rust signature, so it is not confined to reporting.

The row deliberately does not name a repair site. The defect is in the type applier or the diagnostic printer, three lanes are converging on 05_emit_rust.dag, and naming a site would create a second authority for a fix nobody has scoped.

Provenance

Both instances reproduced on a binary built from a pristine origin/main export (6741a3953, modified-mirror-files=0, the is_host_text_carrier_type repair verified absent from the compiled mirror), each arm carrying a firing positive control — so neither is an artifact of the measuring tree. Earlier figures from this lane were produced by a binary built with that repair in the compiled mirror and were re-run before any of this was treated as a fact about main.

Neither row is enrolled: no probe asserts either, so nothing reds if the applier or the exemption changes. Both say so.