PROBE3 debug eprintln! shipped in v0.49.1-v0.49.3 — 31k lines/day into the daemon journal #199

crates/aft/src/lsp/tsconfig_membership.rs has two bare eprintln!("PROBE3 …") calls on the membership hot path:

:72   eprintln!("PROBE3 no tsconfig root: file = {} canonical = {}", …);
:79   eprintln!("PROBE3 file = {} | canonical = {} | tsconfig_dir = {}", …);

They arrived in b0cab7b2 ("fix(context): normalize the status-bar filter root (verbatim class, #10)", Aug 6) alongside the real fix, and are present in v0.49.1, v0.49.2, v0.49.3 and main. v0.49.0 and earlier are clean.

Unconditional, no log level, no feature gate — one line per file per membership check.

Why this is worth more than "debug line left in"

In standalone mode it goes to a bridge stderr nobody reads. Under the subc daemon it lands in the shared journal, and the volume is not small. Last 24h on this box:

31,438  PROBE3 lines
     2  [aft] callgraph store refresh failed: sqlite error: database is locked

That is roughly 1,300 lines/hour of unconditional output from one module in a multi-module daemon. The practical cost is that it degrades operational visibility for everything else sharing that journal — a supervised module exited abnormally (crash) module_id=X line now has to be found among the noise. We caught a claustrum crash-loop yesterday exactly that way, by sweeping journalctl --user -u ck-subc, and that sweep is measurably harder now.

Worth noting the class: this is another defect that standalone mode structurally hides and daemon mode surfaces immediately — same shape as the six-registration problem in #195, where a month of green standalone CI concealed something that broke within hours under the daemon. Debug output is harmless when it goes to a per-session stderr and load-bearing when it goes to a shared supervisor log.

Fix

Delete both, or gate them behind log::trace! / the existing slog_* macros so they honour log level like the rest of the file's diagnostics. Happy to send a PR if you want it as one — it is a two-line delete, so it may be faster for you to just drop it in the next train.

Reproduction is a box running aft as a subc module across three sessions; the count above is from journalctl --user -u ck-subc --since "24 hours ago" | grep -c PROBE3. The consumer-side session that flagged it can add detail if useful.