PP-295 local JSON-lines log triage: earlier hands-on check record, 2026-10-03 UTC Historical record. This check ran on October 3, 2026, before Practical Picks adopted its research-only method later that day. It is kept as a dated observation of the versions named below (lnav 0.14.1, angle-grinder v0.19.6 release asset, Logdy core 0.17.1), not as current testing. The comparison was re-decided from public evidence on October 20, 2026, and now also covers the DuckDB command line and jq, which this check did not include. The angle-grinder version difference noted below is explained by the project's release history: the v0.19.6 tag points to a commit whose Cargo.toml still read 0.19.5. Task: locally inspect an invented JSON-lines app log, find the endpoint with the most valid error records, count those errors, inspect req-42, and account for one malformed line. No real logs, accounts, uploads, system install, or production server were used. Download these five public files into one directory: fixture.jsonl lnav-format.json acquire.py replay.py replay-results.json The synthetic fixture is 1,054 bytes, nine lines, SHA-256 7c93edfd8a679703433e836e16b09f951bda3021aeb01e9e589813335bc18e65. It has eight valid rows and a truncated JSON row at line 5. Among valid rows, /api/orders has three error events, /api/login has one, and req-42 has two /api/orders records. The truncated line is not a valid error event. Linux x86-64 CLI replay (Python 3 standard library): python3 acquire.py ./pp295-tools python3 replay.py --lnav ./pp295-tools/lnav --agrind ./pp295-tools/agrind --workdir ./pp295-run The acquisition script downloads official release assets into the chosen folder, pins SHA-256 before extracting, and never installs to a system path. It downloads lnav v0.14.1 Linux musl x86-64 zip (42,423,829 bytes) and angle-grinder v0.19.6 Linux musl tarball (2,139,243 bytes). Official asset URLs and digests are in acquire.py. The angle-grinder asset binary prints `ag 0.19.5`; release tag and embedded version disagree, so we do not resolve provenance beyond the pinned official asset. The replay script copies the fixture and format into its disposable workdir; it does not alter the downloaded fixture. It uses lnav's -I path to load a custom JSON-lines format without installing into the user's configuration. It runs lnav's grouped count, invalid-row count, invalid-row detail with parser location, req-42 detail, and repeat use of the same format. It runs angle-grinder's grouped count and req-42 filter. It asserts expected values and writes command argv, stdout, stderr, exit codes, hashes and assertions to pp295-run/replay-results.json. The shipped replay-results.json is our recorded output. A matching PASS only covers this nine-line fixture. Important interpretation: lnav returned one `invalid` row with a parser location, but its process exit was 0. Explicitly query `SELECT count(*) FROM pp295_json WHERE log_level = 'invalid'` before trusting a count; the tool did not fail the command. angle-grinder printed `Expected JSON` to stderr, excluded that line from its count, and also exited 0. Check stderr; exit status alone is insufficient. Neither is claimed to validate every JSONL format or operate as an unattended production gate. Generic JSONL needs a source-specific lnav format; our file maps only this invented schema. Newcomer screen: Logdy core v0.17.1 was also run locally on the same fixture. Official binary: https://github.com/logdyhq/logdy-core/releases/download/v0.17.1/logdy_linux_amd64 (44,794,274 bytes), SHA-256 e51ba54043efd052ed7aca4d6db1695825bae294c65f64794c20243ae05bae7c. Task-local command: ./logdy_linux_amd64 follow fixture.jsonl --full-read --ui-ip 127.0.0.1 --port 18795 --no-analytics --no-updates Verify `ss -ltnp` shows 127.0.0.1:18795 only, then visit http://127.0.0.1:18795/ in a local browser. The filter `data.level == "error" and data.path == "/api/orders"` showed 3 out of 9 logs; `data.request_id == "req-42"` showed 2 out of 9. The malformed input remained visible as a raw row, but the default UI did not explicitly mark a JSON parse failure in this test. Stop the process after inspection and confirm the listener disappears. We used explicit opt-out flags for documented anonymous analytics and update checks. This is not a network/privacy audit or proof that the product is fully offline. No Rising Star badge is awarded for sensitive-log use. Limits: one small invented file, Linux x86-64 on one host, no scale test, adversarial log corpus, packet capture, real secrets, production monitoring, collaborative access or support test. We did not test custom formats for other logs. The browser UI was loopback-only and stopped. Preserve copies of real logs and consider their sensitivity before running any viewer or sharing output.