[nc-intezra]

2.0.2, stacked on #2 (the guided setup). Base is setup-wizard; GitHub retargets it to main when #2 is merged.

Changes

• parallel_files per folder module (1–16, default 1): that many files of a batch in flight at once, each its own lidi transfer, so a batch of many files uses several RaptorQ threads/ports instead of one file at a time. Outcomes are still collected in order; after a failure nothing new starts, in-flight files finish, the rest wait for the next pass.
• lidi.max_clients default is now the modules' concurrent transfers (parallel files included) + 2. Derived from config.yaml, so it stays identical on both boxes.
• lidi ports by CPU count: the default stays [5000] (it must be the same on both boxes, so it cannot follow each box's CPU count automatically). Instead dyode setup suggests a count and dyode check prints a hint with a ready-made ports: list.
• dyode check vs the installed unit: warns when dyode-in/out.service waits for a different NIC than config.yaml (it would never start), runs the wrong command for the side, or points at another config; also warns when an explicit max_clients is below what the modules need.
• Reference units in packaging/systemd/reference/ (v1/v2, both sides), with a test that keeps them identical to the generator.
• Version 2.0.2, CHANGELOG, README quick start (now led by the guided setup), config.example.yaml comments.

The "installer warns when the unit was built from the example config" item from the 2.0.2 list already landed in #2.

Testing

• 54 unit tests + 2 loopback tests through real embedded lidi engines (new: 9 files with parallel_files: 4 over 2 ports, all intact and retired).
• New unit tests: unit/NIC check cases, reference units match the generator, advanced defaults accepted by the loader.
• dyode check exercised against a unit generated for the wrong NIC.
• fmt and clippy -D warnings clean. Not yet run on diode hardware; no throughput measurement of parallel_files yet (the sandbox has 2 CPUs).

---

Generated by Claude Code