case147 — Depth ladder: the same input answered at three depths¶

Field	Value
Verdict	🟡 COMPATIBLE_WITH_RISK
Category	Risk
Platforms	Linux
Flags	Bad practice
Detected `ChangeKind`s	—
Source files	`examples/case147_scan_depth_ladder/`

Verdict: 🟡 COMPATIBLE_WITH_RISK · Cross-check: private_header_leak · Mode: single-release audit · Evidence tier: L2 (→ L5 corroboration)

What it demonstrates¶

The legibility anchor for ADR-035's honest-coverage promise: identical input scanned at increasing depth, with the report stating exactly what each depth proved and what it could not — never a bare "scan failed".

connect() takes detail::SessionState&, a private-header type.

Depth	Method	What it proves	What it cannot
S3	lexical pattern pre-scan (no compiler)	flags a risky construct: a public signature mentions a `detail::` name	cannot confirm the type is actually private — only a textual hint
S2	preprocessor (if a compile DB is present)	resolves the `#include` graph	does not parse the AST
S5	source replay + L5 source graph	confirms `detail::SessionState` originates in a private header and is reached from a public decl → `PRIVATE_HEADER_LEAK`, corroborated by the `source_index` provider	(the deepest answer)

The committed snapshot.abi.json carries the L2 header provenance and the L5 source graph, so the cross-check fires with the source_index corroboration the S5 pass would add — the endpoint of the ladder.

Reproduce the ladder¶

abicheck scan --binary libdemo.so -H include/ --audit --source-method s3   # pattern only
abicheck scan --binary libdemo.so -H include/ --audit --sources . --source-method s5  # S5 replay + graph

Fix¶

Same as a private-header leak: opaque-handle the type, or install its header.

Source files¶

snapshot.abi.json