kixelated · Oct 17, 2026 · Oct 17, 2026 · Oct 17, 2026 · Oct 17, 2026 · Oct 17, 2026
diff --git a/.github/workflows/nightly.yml b/.github/workflows/nightly.yml
@@ -97,7 +97,7 @@ jobs:
             recipe: pub check-bbb
             uncached: true
           - name: Browser backend baseline
-            recipe: test wasm-baseline
+            recipe: test wasm-baseline --uniffi
             chromium: true
             artifact: wasm-backend-baseline
             artifact-path: target/test

diff --git a/bench/ffi-gen/README.md b/bench/ffi-gen/README.md
@@ -131,8 +131,8 @@ BoltFFI's JS glue is its generated web module plus `@boltffi/runtime`, unminifie
   `.github/workflows/platform.yml` (`macos-latest`, beside `just swift check`). The
   harness has no Swift runner yet.
 - **UniFFI's JS glue.** ubrn, UniFFI's JS generator, does not support 0.32 yet, so only
-  its `.wasm` is sized. Browser per-call cost is
-  [#5478](https://github.com/moq-dev/moq/pull/5478)'s job.
+  its `.wasm` is sized. The browser comparison, including its per-call cost, is
+  [`test/wasm/generator.md`](/test/wasm/generator.md).
 - **The full moq-ffi through BoltFFI.** BoltFFI cannot bind moq-ffi without the API
   changes listed below, so its numbers come from the slice alone.
 - Go, Dart, C++, and C, which BoltFFI lacks or has only in part. Startup and memory.
@@ -233,4 +233,5 @@ bulk copies), not from its core. A UniFFI fork can adopt each one backend at a t
 using these numbers as targets, while keeping all eight targets working throughout.
 
 Pick BoltFFI instead if call overhead outweighs everything else and writing the Go and
-C++ backends is acceptable. The browser half is decided separately in #5478.
+C++ backends is acceptable. The browser half was decided separately: UniFFI over ubrn
+([#5838](https://github.com/moq-dev/moq/pull/5838), [`test/wasm/generator.md`](/test/wasm/generator.md)).
diff --git a/bun.lock b/bun.lock
diff --git a/quest/m1/README.md b/quest/m1/README.md
@@ -110,6 +110,7 @@ lands before the cut.
 - [Close tail drain](/quest/m1/close-tail-drain.md) - after a peer close, subscribers read the group streams their transport already delivered instead of aborting them, so a finished tail arrives on every version
 - [Publisher linger](/quest/m1/publisher-linger.md) - every finishing publisher keeps its broadcast up for the media max-age so subscribers can drain it
 - [Exporters follow a rendition to a new track](/quest/m1/export-rendition-rebind.md) - the fMP4, TS, MKV and FLV exporters rebuild a rendition whose resolved `track` or `broadcast` changes under the same ID
+- [lite-06 max sequence](/quest/m1/lite06-max-sequence.md) - group sequence 2^62-1 reaches a lite-06 subscriber or is refused loudly, instead of vanishing
 
 - [C++ and OBS package proof](/quest/m1/cpp-package-check.md) - the installed artifacts pass before their first release
 

diff --git a/quest/m1/lite06-max-sequence.md b/quest/m1/lite06-max-sequence.md
@@ -0,0 +1,23 @@
+# [S] lite-06 drops group sequence 2^62-1 silently
+
+## Goal
+
+A group whose sequence is 2^62-1, the largest QUIC varint, either reaches a moq-lite-06
+subscriber through the relay or is refused loudly where it cannot be carried. Today the
+publisher accepts it and the subscriber waits forever, while moq-lite-02 and
+moq-transport-19 deliver it and 2^53+1 crosses all three.
+
+## Plan
+
+Found by the browser binding spike (#5838): `test/wasm/uniffi/cases.ts` publishes the
+ladder 5, 2^32, 2^53-1, 2^53+1, 2^62-1 from one moq-ffi session to another through each
+relay flavour. On lite-06 it records 2^62-1 as a known gap, and the case fails once the
+gap closes.
+
+Reproduce natively first, with a moq-net test over the memory transport and then through
+moq-relay, to find whether the publisher, the relay, or the subscriber drops it. Fail
+loud if the value is illegal on lite-06; otherwise deliver it. Land a regression test
+with the fix, then remove the browser case's known-gap branch.
+
+Public API: none expected. Wire: none if the value is legal; otherwise the refusal
+matches the published draft.
diff --git a/quest/m2/ffi-bytes.md b/quest/m2/ffi-bytes.md
@@ -11,6 +11,10 @@ wrapper (Python, Go, Kotlin, Swift, Dart, and C++).
 
 Decided with the maintainer on 2026-10-13:
 
+- **The browser too.** Since 2026-10-17 (#5838) the browser binds moq-ffi through
+  UniFFI over ubrn, and pays the same per-byte path: about 7 ns per byte written and
+  8 ns read at 128 KiB in Chromium, against BoltFFI's 0.5 and 1.3. Re-measure with
+  `test/wasm/uniffi/cases.ts`'s per-call timings once the fork's core changes.
 - **Benchmark first.** `just bench-ffi` ([bench/README.md](/bench/README.md))
   measures every wrapper before and after. Add a Rust microbenchmark of
   UniFFI's lift and lower for `Vec<u8>` here, beside the patch it proves.

diff --git a/quest/m2/rs2ts/README.md b/quest/m2/rs2ts/README.md
@@ -44,6 +44,13 @@ for a source change. A batched boundary is an optimization candidate, not a
 reason to require a full sans-IO rewrite. Controlled time remains independently
 useful; source-level `Instant` mappings depend on the chosen backend.
 
+Decided with the maintainer on 2026-10-17
+([#5838](https://github.com/moq-dev/moq/pull/5838)): the WASM binding is moq-ffi
+generated by UniFFI over ubrn's `wasm2` player. It won over hand-written moq-wasm and
+BoltFFI on maintenance (one surface for every language, no API break) and parity. Its
+measured costs, size and payload copies, have their own quests. The report is
+[`test/wasm/generator.md`](/test/wasm/generator.md).
+
 Evidence from [#5337](https://github.com/moq-dev/moq/issues/5337) motivates
 reopening the options, not choosing a winner. Its attachment reports the
 original async Rust consumer passing nine browser tests in both WASM and
@@ -61,8 +68,9 @@ any necessary breaks. Wire: no changes intended.
 
 ## Required
 
-- [Binding generator](/quest/m2/rs2ts/generator.md) - the maintainer picks UniFFI- or BoltFFI-generated moq-ffi, or hand-written moq-wasm, from measurements
 - [Complete WASM bindings](/quest/m2/rs2ts/wasm-bindings.md) - Rust publishing and consumption cover the browser protocol and model surface
+- [Net-only moq-ffi](/quest/m2/rs2ts/net-only.md) - the browser `.wasm` drops the media layer, near moq-wasm's size
+- [ubrn reads UniFFI 0.32](/quest/m2/rs2ts/ubrn-032.md) - waiting on upstream, so the browser TypeScript generates without the measurement pin
 - [Adopt WASM in @moq/net](/quest/m2/rs2ts/wasm-adopt.md) - the TypeScript interface uses Rust, with measured tradeoffs accepted before replacement
 
 ## Related

diff --git a/quest/m2/rs2ts/generator.md b/quest/m2/rs2ts/generator.md
diff --git a/quest/m2/rs2ts/net-only.md b/quest/m2/rs2ts/net-only.md
@@ -0,0 +1,30 @@
+# [M] A net-only moq-ffi build for the browser
+
+## Goal
+
+The browser's moq-ffi `.wasm` carries only the network surface `@moq/net` wraps, so its
+shipped size lands near moq-wasm's (about 740 KB gzip on 2026-10-16) instead of the full
+build's 1.70 MB, stripped. Measured with `bun test/wasm/uniffi/size.ts` and the
+baseline sweep's asset sizes.
+
+## Plan
+
+Decided with the maintainer on 2026-10-17 (#5838): the browser binds moq-ffi through
+UniFFI, and size was the choice's largest cost. The report
+(`test/wasm/generator.md`) attributes about 30% of moq-ffi's code to the media layer:
+mp4\_atom, moq\_mux, `hang`'s regex crates, and the JSON crates. A slice that consumes and
+publishes over WebTransport landed within 3% of moq-wasm under UniFFI, so the generator
+itself adds little.
+
+Use a cargo feature on moq-ffi, not a new crate: #5460 rejected per-layer crate splits.
+The UniFFI exports the linker keeps are the ones the feature compiles in. Toggling
+`default-features` changes nothing on wasm32, because `audio` and `video` are already
+native-only. Also strip the `name` section when staging, worth about 235 KB gzip.
+
+Public API: a new moq-ffi cargo feature; the bound surface is unchanged where it is
+enabled. Wire: none.
+
+## Related
+
+- [Adopt WASM in @moq/net](/quest/m2/rs2ts/wasm-adopt.md) - weighs the shipped size
+- [Bindings size profile](/quest/m2/ffi-size-profile.md) - the native builds' size question
diff --git a/quest/m2/rs2ts/ubrn-032.md b/quest/m2/rs2ts/ubrn-032.md
@@ -0,0 +1,30 @@
+# [XS] Waiting: ubrn generates UniFFI 0.32 TypeScript
+
+## Goal
+
+uniffi-bindgen-react-native (ubrn) ships a release whose `wasm2` flavour reads UniFFI
+0.32 metadata, so moq-ffi's browser TypeScript generates from the published crate
+without `test/wasm/uniffi/pin.patch`. Check: whether
+[ubrn#468](https://github.com/jhugman/uniffi-bindgen-react-native/pull/468) has merged and
+a release after 0.31.0-6 lists it. Delete this quest once the pin is gone.
+
+## Plan
+
+Decided with the maintainer on 2026-10-17 (#5838): the browser binds moq-ffi through
+UniFFI over ubrn's `wasm2` player. ubrn pins UniFFI 0.31; moq-ffi is on 0.32, and the
+harness pins a scratch checkout to 0.31 for measurement only.
+
+If ubrn has not moved by the time adoption is otherwise ready, carry ubrn's TypeScript
+generator in [the UniFFI fork](/quest/m2/uniffi-fork.md) instead, and delete this quest.
+
+When raising it upstream, also report two costs the spike measured. Posting outside
+`moq-dev` and `kixelated` needs the maintainer's approval.
+
+- `@ubjs/core` arms and clears a `setTimeout` keep-alive on every async call, which only
+  Node.js needs.
+- The generated `*Like` return types omit `uniffiDestroy`, so freeing a returned handle
+  early needs a cast to the class.
+
+## Related
+
+- [Adopt WASM in @moq/net](/quest/m2/rs2ts/wasm-adopt.md) - ships the generated TypeScript this unblocks
diff --git a/quest/m2/rs2ts/wasm-adopt.md b/quest/m2/rs2ts/wasm-adopt.md
@@ -30,10 +30,15 @@ Update watch, publish, hang, room, demos, and other affected consumers together.
   in existing API docs and package READMEs, together with regeneration.
 - Retain baseline revisions and shared tests after removing the handwritten
   implementation. Keep the WASM backend while the JS-only stage is evaluated.
+- Delete `rs/moq-wasm` with the handwritten code: on 2026-10-17 (#5838) the
+  maintainer chose UniFFI-generated moq-ffi as the browser binding, so moq-wasm
+  is only the baseline's reference until then.
 
 Public API: `@moq/net` remains the package; enumerate any approved signature,
 initialization, lifetime, or deployment changes. Wire: none.
 
 ## Required
 
 - [Complete WASM bindings](/quest/m2/rs2ts/wasm-bindings.md) - the complete Rust-backed candidate and parity coverage
+- [ubrn reads UniFFI 0.32](/quest/m2/rs2ts/ubrn-032.md) - the TypeScript generates from the published moq-ffi, without the measurement pin
+- [Net-only moq-ffi](/quest/m2/rs2ts/net-only.md) - the browser `.wasm` drops the media layer, near moq-wasm's size
diff --git a/quest/m2/rs2ts/wasm-bindings.md b/quest/m2/rs2ts/wasm-bindings.md
@@ -16,8 +16,18 @@ close/error behavior, cancellation, and backpressure. Validate supported
 versions and transports with the browser fixtures and `just test interop --all`.
 Wire new tests into CI, at least nightly.
 
-Complete the binding the [generator quest](/quest/m2/rs2ts/generator.md)
-chose: `moq-ffi` through UniFFI or BoltFFI, or the consume-only `moq-wasm`. Measure boundary calls, Promise allocation,
+Decided with the maintainer on 2026-10-17 (#5838, report in
+[`test/wasm/generator.md`](/test/wasm/generator.md)): complete `moq-ffi` through UniFFI
+over ubrn's `wasm2` player, not the hand-written `moq-wasm` and not BoltFFI. This quest
+completes moq-ffi's browser parity and generates its TypeScript in CI, extending
+`test/wasm/uniffi/` (`just test wasm-baseline --uniffi`, nightly). `moq-wasm`'s deletion
+moves to [adoption](/quest/m2/rs2ts/wasm-adopt.md). Until
+[ubrn reads UniFFI 0.32](/quest/m2/rs2ts/ubrn-032.md), generation keeps applying
+`test/wasm/uniffi/pin.patch` to a scratch checkout.
+
+Gaps the spike found against `@moq/net` and moq-wasm: moq-ffi exposes no negotiated
+version and has no WebSocket fallback on wasm, and payload copies cost about 7 to 8 ns
+per byte (see [Payload bytes](/quest/m2/ffi-bytes.md)). Measure boundary calls, Promise allocation,
 payload copies, and retained memory. Compare a batched poll/event-drain
 boundary if these dominate, including latency and backpressure effects;
 do not require a sans-IO rewrite or impose batching latency to win throughput.
@@ -34,7 +44,3 @@ Update the bindings' README and existing API docs for the surface added.
 
 Public API: extends the unpublished browser bindings; any changes needed to
 `@moq/net` are proposed in the adoption quest. Wire: none.
-
-## Required
-
-- [Binding generator](/quest/m2/rs2ts/generator.md) - which binding this completes
diff --git a/test/tsconfig.json b/test/tsconfig.json
@@ -5,5 +5,5 @@
 		"noEmit": true
 	},
 	"include": ["**/*", "../js/common/worklet.d.ts"],
-	"exclude": ["**/node_modules", "**/dist", "**/target"]
+	"exclude": ["**/node_modules", "**/dist", "**/target", "wasm/uniffi"]
 }
diff --git a/test/wasm/README.md b/test/wasm/README.md
@@ -172,3 +172,15 @@ WASM-to-JS measurements are separate, unverified results here. Full network
 sequences above 2^53 need a native publisher and a JS model that can represent
 them before adoption. WASM frame timestamps and complete publish/backpressure
 API parity remain explicit gaps too.
+
+## Generated bindings
+
+```bash
+just test wasm-baseline --uniffi
+```
+
+`--uniffi` adds moq-ffi, bound by UniFFI over ubrn's `wasm2` player, as a third
+baseline backend. It also runs parity cases and per-call timings in `uniffi/` against
+every relay, once normally and once under a strict CSP. It builds from a scratch
+checkout, since ubrn needs a UniFFI pin that does not land.
+[generator.md](generator.md) is the report that chose this binding.