# status and evidence

## what was verified

All commands run from the repository root with Go 1.25+ and TinyGo 0.43
(Go 1.27 from `~/.local/goroot` where TinyGo needs it).

| Check | Result |
|---|---|
| `go test ./...` | pass |
| `go vet ./...` | pass |
| `go test -race ./pkg/tunnel ./pkg/neutrino ./internal/e2e` | pass |
| `tinygo build -target wasm -o bin/nostrino.wasm ./cmd/nostrino-wasm` | pass |
| `tinygo build ./cmd/nostrino-bridge ./cmd/nostrino-neutrino` | pass |
| wasm sync through torly + bridge + test node | tip 50, 51 filters; ~2.7 s relay, ~20 ms WebSocket |
| IndexedDB reload (two wasm instances, fake-indexeddb) | second sync adds 0, same tip, local match still works |
| wasm cancel mid-sync | settles as `cancelled`, bridge closes the session |
| frozen reader / zero credit | 4 MiB write blocks, then unblocks on reset |
| NIP-42 on a torly with auth required | endpoint-to-endpoint sync passes |
| two relays, one dropping every event | stream continues on the other |
| WebSocket data plane (bridge=) | e2e sync over the WS stream; both wasm smokes assert `transport=websocket` |
| reorg to a different branch at height 6 | old filters replaced, new branch matches |
| checkpoint at height 10 | starts at 10, verifies 11..tip, no filter below |
| gateway presence + `nostrino-discover` | finds the URI, chains and node tip |
| file store resume | second sync adds 0; tip and filters preserved |
| dropped relay events | 512 KiB echo survives 1-in-9 drops |
| profile JSON + difficulty/retarget/MTP | unit tests pass |

Reproduce:

```sh
make build wasm
./scripts/wasm-smoke.sh

(cd /tmp && npm install fake-indexeddb)   # once
NODE_PATH=/tmp/node_modules ./scripts/idb-smoke.sh

bin/nostrino-discover -relay wss://relay.example -chain btc
```

The smoke scripts start torly, the deterministic test node, the bridge and the
wasm bundle under Node. They are the acceptance tests for the browser target.

## known gaps

Ordered by what matters next.

1. **Pinned network data.** Mainnet checkpoint values and the real XBT
   parameters are not in the binary. `-profile` and `-checkpoint-*` accept
   them; the operator must obtain them from a trusted node and keep them
   current. A stale checkpoint only costs sync time; a wrong one is rejected
   on the first filter header.
2. **Streaming browser storage.** The IndexedDB store loads its indexes into
   memory on open. That is fine for a regtest chain and a recent-range wallet,
   but a full mainnet filter set (GB scale) needs a store that reads filters
   on demand. The `Store` interface is the seam; the sync loop should not
   change.
3. **First retarget after a checkpoint.** The window's first block is not
   stored, so that one boundary is accepted from the peer. Every later
   boundary is verified. A checkpoint that also pins the window's start would
   close this.
4. **Deep reorgs.** Fork search is bounded at 10,000 headers. A deeper reorg
   is refused rather than silently accepted, which is the safe failure; a
   mainnet wallet also needs the height it scans from.
5. **TLS termination for the bridge.** The WebSocket data plane is
   implemented, but the reference build listens in plain `ws://`. A public
   `wss://` endpoint needs a reverse proxy (or a later TLS listener) and the
   operator supplies its advertised URL with `-bridge-url`.
6. **Registry abuse and accuracy.** Presence is an unauthenticated public
   advertisement; the tip is from a handshake every five minutes and can be
   stale. There is no TTL enforcement beyond the relay's addressable-event
   replacement, no per-publisher rate on announcements, and discovery trusts
   the payload (the event signature and `d` tag are checked, but a gateway can
   lie about its tip).
7. **Quotas and accounting.** The bridge has a per-stream byte rate and a
   per-client session cap. There is no per-day quota, no per-IP limit, and no
   accounting to bill or block a heavy client.
8. **Relay auth for reads.** NIP-42 is answered and publications wait for the
   auth result. A relay that requires auth for REQ is not specially handled;
   torly gates writes only.
9. **Wallet and swap layers.** Out of scope for this repository: BIP32/BIP84,
   UTXO selection, HTLC scripts, signing, the swap state machine, the
   settlement deadline recomputation on resume, and the server-side swap
   record. The registry and the tunnel deliberately do not carry swap state.

## operational notes

- The gateway key lives in `DATA/gateway.key` (0600), the URI in
  `DATA/gateway.uri`.
- Torly's WebSocket idle timeout is 150 s; the native transport pings every
  45 s and the listener reconnects with backoff.
- Torly's maximum event is 128 KiB; the default 32 KiB frame produces about
  55 KiB of encrypted content.
- The bridge defaults to 4 MiB/s per direction and two streams per client;
  adjust with `-max-bytes-per-sec` and `-max-sessions-per-client`.
- `nostrino-bridge -check -node ...` reports whether the node advertises
  `NODE_COMPACT_FILTERS`; an operator must still confirm
  `blockfilterindex=basic` and the filter index height.
- The browser WebSocket stream has no portable socket read timeout; the
  bridge's idle bound and the WebSocket close/error events bound liveness.
  The P2P read deadline is advisory on that path.
- The test node is a fixture, not a Bitcoin node.