A results.json the report cannot read renders as "nothing measured yet", at exit 0

[vyncint]

Today — Results::load (crates/launchbound-bench/src/run.rs:367-370) is read_to_string(path).ok()? followed by serde_json::from_str(&text).ok(), and RunDir::load (crates/launchbound-report/src/build.rs:35) takes its None to mean the measurement box has not run. So a missing results.json, a truncated one, an empty one, null, [], a document with a different schema, and a directory called results.json all render the same report. Measured against 2.0.0 on the run directory this repository ships, runs/reduce-stable-metal, with its results.json replaced:

results.json = '{'                          exit=0   launchbound report — reduce-stable · gate cc metal · unmeasured · no device — nothing measured yet
results.json = ''                           exit=0   (identical)
results.json = 'null'                       exit=0   (identical)
results.json = '[]'                         exit=0   (identical)
results.json = '{"schema":"results.v1"}'    exit=0   (identical)
results.json = <a directory>                exit=0   (identical)
stderr: empty, in every case

$ launchbound report badrun --json | jq -r .measurement_kind
unmeasured                                  # and the document validates against schemas/report.v1.json, 0 errors

With the shipped file intact the same command prints CHOSEN: c1-43e865c2e1620697 … measured.

model --results goes through the same loader, so it cannot say why either (crates/launchbound-cli/src/main.rs:602-603):

$ launchbound model reduce-flip --cc 8.6 --results empty.json
error: cannot read results.v1 at "empty.json"
$ launchbound model reduce-flip --cc 8.6 --results shape.json        # {"schema":"results.v1"}
error: cannot read results.v1 at "shape.json"
$ launchbound model reduce-flip --cc 8.6 --results /nonexistent.json
error: cannot read results.v1 at "/nonexistent.json"

Why it is worth fixing — the run directory is the hand-off between two machines. stage writes the plan here, the measurement box writes results.json, and the file comes back by whatever copies it. A partial copy, the wrong file, or a results.v2 from a newer runner all read as "the box has not run yet": exit 0, nothing on stderr, and a JSON report that is schema-valid and says unmeasured. The two conditions call for opposite actions — wait, or go and look — and nothing tells them apart. tune renders this same report at the end of a run (main.rs:533), so the last thing a run prints can be wrong in the same way.

The checkpoint writer is atomic — temp file, then rename (run.rs:373-381) — so a crash on the machine that measures leaves results.json.tmp, not a torn file. The exposure is everything after the box, which is the part no test on either machine sees. verdicts.json already gets this right two lines above: its schema tag is checked by name and a mismatch is unsupported verdicts schema …, exit 2 (build.rs:28-33). results.json deserves the same.

Fix — make Results::load return Result<Option<Results>, String>: NotFound is Ok(None); every other I/O error and every parse error is Err with the path and serde's message. report exits 2 on Err with the reason on stderr and no report on stdout; model --results prints the cause instead of "cannot read". Check the schema tag against results.v1 the way verdicts.v1 is checked, so a future results.v2 is refused by name rather than read as nothing.

Done when — a truncated, empty, wrong-schema or directory results.json makes report exit 2 with the cause on stderr and nothing on stdout; a missing one still renders "nothing measured yet"; model --results names the cause; a test in launchbound-report covers each shape; tune's end-of-run report inherits the behaviour.