[yyjeqhc]

Four execution projection paths independently maintained the same rules for omitting empty stdout/stderr tails, zero line counts and false truncation flags. A private sparsify_empty_execution_streams function now owns those exact rules for terminal validation, pending validation, structured execution and shell success.

The caller-specific terminal proofs, pending checks, failure handling, recovery metadata and ordering remain in place. Null/malformed values, nonempty output and output-loss evidence are preserved. There are no schema, authority, persistence, continuation or tool-surface changes, additional allocations, generics or dispatch abstractions.

This is deliberately a small structural cleanup: 35 production Rust lines removed, with four independently maintained implementations reduced to one. Its benefit is preventing stream-projection drift; it does not attempt to unify semantically different execution responses.

Metric	Before	After	Net change
ToolRuntime production Rust (estimated separation)	92,729	92,694	-35
ToolRuntime test Rust (including estimated inline tests)	164,632	164,632	0
ToolRuntime Rust total	257,361	257,326	-35
Repository Git-tracked Rust total	928,714	928,679	-35

Counts use exact PR merge base bc63e43505f0162153950bc9ee43c3f6458b4294, physical source lines including blanks/comments, and all tracked .rs files. Test separation includes 145,402 lines in dedicated test files plus an estimated 19,230 lines in explicit cfg(test)/cfg(all(test, ...)) item spans. Partition counts are estimates; totals and deltas are exact. This branch is independent of #1245 and edits a different file.

Validation:

• cargo test --locked -p webcodex --lib tool_runtime::result_projection: 25 passed, 0 failed, 0 ignored, covering execution/pending/failure projections, shell exceptional-state preservation and validation output schemas.
• cargo fmt --all -- --check and git diff --check: passed.
• Expanding the four helper calls in a temporary source comparison yields the exact original Rust token stream, excluding comments/whitespace. Existing tests are unchanged.

Risk is limited to the final model projection; canonical evidence recording and execution are untouched. Windows/Linux and optional feature configurations were not run locally. No deployment, release, service restart or merge is included.