Without headers or DWARF a Rust cdylib run reports NO_CHANGE with Gate: ACCEPTED, and header-less additive changes get alignment noise (2026-10-15)

Status (2026-10-15): OPEN. Two related behaviors on the no-header path; both are generic ELF behavior that Rust hits by default.

1. Headline label. A default cargo build --release (no debug info) compared without headers:

Compatibility: NO_CHANGE | Policy: strict_abi | Gate: ACCEPTED (exit 0)
WARNING: Compatibility is UNCONFIRMED ...
WARNING: Binary-only analysis without debug info; many ABI changes cannot be detected

The run had a real return-type change. The report carries analysis_assurance: partial and evidence_tier: elf_only, but the headline label is a clean NO_CHANGE and the gate accepts it. The two warnings say the opposite. The AGENTS.md rule is that weaker evidence narrows conclusions and never upgrades to a clean compatibility claim, so this label is a candidate violation. Changing it is a verdict-labelling decision (ADR-066/063), not a fix for one reader.

2. Alignment noise. Adding an exported function to a Rust cdylib, with no headers, produced exported_object_alignment_reduced on an unrelated exported static (RL_VERSION, 64 to 32 bytes) and COMPATIBLE_WITH_RISK. The check (diff_platform_elf_symbols._check_object_alignment_reduced) derives alignment from the symbol's link-time address, so a neighbour's addition moves it. With headers the finding does not appear, because declared alignment corroborates or refutes it. The clang variant of the same gap is pinned in tests/test_clang_header_backend_integration.py.

Fix direction: the headline should not read NO_CHANGE while the surface is unconfirmed, and the address-derived check needs declared-alignment evidence before it fires on a header-less run.

Workaround: pass headers (cbindgen) and build with debug = 2. See Rust Libraries.

Index: Known gaps