Nice — I’ll give you a developer-focused, under-the-hood architecture + concrete implementation plan you can execute in 4–5 days. I’ll keep it practical, with the exact layers you’ll touch, data formats, algorithms, failure modes, tests, and short code skeletons/hints (not full copy-paste app, since you prefer understanding the mechanics). Ready? Let’s go.

✅ High-level summary (1 line)

Use WebTransport (HTTP/3 / QUIC) as the transport; split the app into Chunker → FEC → Priority Scheduler → WebTransport Streams → Receiver reassembly + Verify + Resume. Add a small control channel for metadata/commands and a monitoring stream for live stats.

1) System components & responsibilities

1. Client (browser / CLI)
   • Chunk files, compute per-chunk hash, send metadata.
   • Apply FEC encoder producing parity shards.
   • Open multiple WebTransport streams (control, priority, data, fec, stats).
   • Scheduler decides chunk → stream mapping and retransmits.
   • Persist resume metadata locally (bitmap of received/sent shards, Merkle root).
2. Server / Receiver (edge node or relay)
   • Accept WebTransport sessions.
   • Manage streams: accept control, data, FEC shards.
   • Validate chunk hashes; store chunks temporarily or pipe to storage.
   • Run FEC decoder when enough shards arrive.
   • Handle resume requests; serve missing chunks.
   • Emit realtime stats over a dedicated stream.
3. Optional Relay / Store-and-forward node
   • Buffer in lossy networks; acts like a tolerant relay (useful for intermittent connections).
4. UI / Dashboard (React)
   • Show stream-level throughput, retransmit counts, FEC recovery stats, ETA per file.
   • Visualize health (packet loss estimate, inferred RTT).
5. Persistence / Metadata store
   • Minimal: file manifests (Merkle root, chunk count/size, FEC params). Could be memory for short runs.

2) Protocol & stream layout (logical)

Use 3–4 logical WebTransport streams (mapped to QUIC streams):

• Control stream (bidirectional) — metadata exchange, manifest, resume handshake, ACKs (small, reliable).
• High-priority stream — small chunks/headers/manifest chunks (low latency).
• Data streams (one or more unidirectional/bidirectional) — bulk chunk payloads.
• FEC stream(s) — parity shards (can be unidirectional).
• Stats stream (optional bidirectional/unreliable datagrams) — live telemetry (byte/sec, packet loss estimates, decode events).

WebTransport supports datagrams (unreliable) as well — use datagrams for live metrics and low-priority non-critical messages; use streams for chunks that must be reassembled.