Go encapsulation counts every lowercase symbol: 72% of a Go DOOM port lands in the encapsulation archetypes (C original 2%) #5897

Finding (DOOM study calibration, #5620)

Go's encapsulation rule counts every lowercase top-level symbol (go.py: ^func\s+(?:\([^)]*\)\s+)?[a-z]… and ^(type|var|const)\s+[a-z]…). keyword-rosetta ledgers this as intended visibility morphology (visibility-morphology-encapsulation, engine-semantic, validated 2026-08-31) and warns that "cross-language encapsulation comparisons must expect these to run structurally hot". On real translated code the effect is no longer a footnote. It decides the function archetypes the engine reports for Go.

Measurements at engine 9f213ab83, licensed DOOM ports, functions of at least 3 lines:

¹ Encapsulated Accessors + Stateful Encapsulated Methods (k=14 function model).

gore keeps linuxdoom's call wiring (77% of linuxdoom's calls survive between the corresponding functions; structural-fingerprint-scanner subway maps, population-analyses examples/doom/), so the algorithm is the same. Yet nearly three quarters of its functions, renderer and playsim included, are labelled as encapsulation archetypes. Seven of gore's nine package clauses are package main, and 950 of its 1,025 func declarations are lowercase. In a main package nothing can be imported, so a lowercase name is not a visibility decision at all. The keyword-rosetta control shows the same skew on the identical planted program: def_encapsulation C 0, Java 0, Dart 0, C++ 0, Go 3.75, Rust 3.

Fixture (our own code, not from any port):

package main

func renderColumn(x int) int { // lowercase: counted as encapsulation today
	return x * 2
}

func main() {
	renderColumn(3)
}

Options

1. Rule: in a package main file, lowercase declarations carry no visibility meaning, so don't count them. Optionally count only unexported methods and types in library packages. This is the smallest semantic fix.
2. Model: keep the rule, but have the function-archetype classifier down-weight def_encapsulation for languages ledgered under visibility morphology (go, perl, and rust if confirmed), so the archetype labels stop tracking spelling.
3. Leave it: document in the archetype docs that Go archetype mixes are not comparable with other languages.

Recommendation: option 1, which stays faithful to the ledger's "correct morphology" (library packages still count), then re-measure. If Go archetype mixes still diverge on translations that keep the original's wiring, do option 2 as well. Check Rust's rule against the same question (rosetta 3 vs 0).

Acceptance

• The fixture above: 0 encapsulation hits in package main; a library-package variant still counts its unexported functions (test).
• gore's encapsulation-archetype share re-measured and posted here (target: within ~10 points of the C original's, or the residual explained).
• keyword-rosetta visibility-morphology-encapsulation updated (Go expected counts re-locked), and goldens re-blessed with the drift explained.

Part of #5457; calibration finding for #5620. Data: structural-fingerprint-scanner per-subsystem view (gitgalaxy-population-analyses PR #19).