shell: reflection_metaprogramming matches plain ${var} expansion (indirection marker is optional) and every $(...) -- 1,019 hits in 2,823 LOC on configure-helper.sh

[squid-protocol]

Found while attributing #2719's golden-master diff: with per-file dynamism (reflection_metaprogramming) feeding documentation risk, 15 of the crucible's 46 shell files jump ≥ 25 points, to ~99% -- configure-helper.sh reads 1,019 reflection hits in 2,823 coding lines, ansible-galaxy__runme.sh 149 in 333, xcrun_cache.sh 122 in 225.

The shell rule (gitgalaxy/standards/language_standards/languages/shell.py):

\$\((?:[^()]|\([^()]*\))+\)|`[^`]+`|\b(?:awk|sed|perl|python[23]?|ruby)\s+['"][^'"]{0,500}|\beval\s+\$|\$\{!?[a-zA-Z0-9_]+\}

Two of its alternatives count ordinary shell as metaprogramming:

1. \$\{!?[a-zA-Z0-9_]+\} -- the !? makes the indirection marker optional, so plain ${var} matches. ${!var} (indirect expansion) is reflection; ${var} is a variable. This looks like a typo for \$\{![a-zA-Z0-9_]+\}.
2. \$\((?:...)\) and backticks -- every command substitution. test_shell_strict.py L105 pins output=$(date) as a positive, so this one is a deliberate decision by the rule's author; whether "runs a command" is the same thing as "dynamic dispatch a reader can't follow" is the question. On real shell it is one hit per ~3 lines, which is the language's vocabulary, not its dynamism -- the same shape as the control_flow_ratio vocabulary leak in length leaks: func_internal_density, avg_func_loc and control_flow_ratio read file length with content and tier held equal (rosetta F.1 design pass) #2705.

Suggested fix: (1) is a one-character change (!? → !) and should be uncontroversial; for (2), keep eval, ${!var}, declare -n, source/. of a computed path, and the inline awk|sed|perl|python programs, and drop bare $(...)/backticks -- or move them to a signal that means "runs a command". reflection_metaprogramming is unplanted in keyword-rosetta, so the change is not gate-visible there; it is golden-master-visible (documentation and cognitive load on shell).

Related: #2716 (three-input model), #2719 (per-file dynamism), #2705 (vocabulary leaks).

[squid-protocol]

Fixed in #2725 (branch fix/2722-shell-reflection, off 4c283edf).

Measured before touching anything, over the crucible's 255 shell files / 26,274 coding LOC, each alternative counted independently:

alternative	hits	share
${var}	4,117	84.7%
$(...)	789	16.2%
`...`	164	3.4%
awk|sed|perl|python|ruby '...'	152	3.1%
eval $	5	0.1%
${!var}	0	0.0%

So (1) is even more lopsided than it looked: ${!var}, the construct the alternative exists for, does not occur in the corpus at all -- every one of those 4,117 hits was a plain variable.

On (2) the registry settles it rather than the test pin. Every other language's rule counts name-based dynamic dispatch and introspection -- python getattr/setattr/inspect., go reflect./unsafe./cgo, ruby method_missing/define_method, js Reflect/Proxy/__proto__. None counts invoking another program; python's does not fire on subprocess.run. $(date) is that same call and its result is data, not code. test_shell_strict.py L105 now pins it as the NEGATIVE half of the pair, with eval "\$cmd" as the positive.

Took the suggested fix in full: dropped $(...)/backticks and the !?, kept the inline sub-language programs and ${!var}, added eval unanchored (the old \\beval\\s+\$ missed eval "\$cmd"), namerefs declare/typeset/local -n, and source/. of a computed path. 4,863 -> 227 hits, 73 of 255 files still carrying one; configure-helper.sh 1,019 -> 15.

Golden master re-blessed both venvs: 508 diffs, 0 topological, all shell files (the shell/ tree plus wrf's .csh/.ksh scripts and tigerbeetle/download.sh), every other language zero, and 20 files' Documentation Exposure falls by >=25 -- the reversal of the jumps #2719 flagged. Exactly one value in the corpus rises: sqlite/tclConfigShToMake.sh gains a hit for . "\$1", in a file whose header comment calls itself "a level of indirection".

Considered and left out: trap '...' SIG (code in a string, but owned by cleanup) and a "runs a command" home for $(...), which would be a new signal rather than this fix.