Current System Map

Mimir is the public Face and product name for this repo. A few lower native ABI names still use localcast until a deliberate rename cut exists.

The live target is the native rolling field machine described in docs/native-rebuild-plan.md, docs/perfect-machine.md, and config/perfect-machine.example.json.

Perfect Machine Target

flowchart TD
    A["direct camera drivers"] --> R["Mimir.Runtime rolling buffers"]
    B["mic/loopback drivers"] --> R
    C["Leap timing/IR driver"] --> R
    D["network feed producers"] --> R
    R --> N["native reservoir handles"]
    N --> E["Fensalir GPU fusion + UI"]
    N --> F["Faust/native DSP"]
    E --> G["Spout2/program video"]
    F --> H["program stems + spatial bed"]
    G --> I["OBS"]
    H --> I

Ownership:

Mimir owns configuration, calibration truth, launch, status, persistence, and runtime contracts.
Mimir.Runtime owns app-level stream buffers and synchronization.
Native capture workers own device reads.
Fensalir owns dense visual fusion, material/brush/splat reconciliation, D3D12 interop, runtime UI, and Spout2 publication.
Faust/native DSP owns hot audio alignment, suppression, separation, spatialization, and stem generation.
OBS owns broadcast controls.

Invariant: the live window is bounded, in memory, and has one timing authority. No private history outlives the rolling buffer.

Viable Stream App

docs/viable-stream-app.md defines the near-term app target. Fensalir hosts the running Mimir app, keeps the default five-second runtime in memory, exposes debug/settings/output controls, and emits synchronized OBS program video plus separately controllable audio stems.

Mimir.Runtime currently provides:

MimirSynchronizationHub;
MimirRollingStreamBuffer;
stream descriptors/settings;
IMimirStreamSource;
MimirNativeIngestStreamSource;
MimirProcessStreamSource;
MimirFrameEventProcessStreamSource;
MimirVideoFrameDescriptor;
MimirAudioSynchronizationAnalyzer;
IMimirVideoCaptureDriver;
MimirVideoCaptureDriverSource.

Process-backed sources are bridge/network edges. Local cameras should feed native descriptors with device timestamps and optional native/GPU handles. The frame-events/json-lines adapter is a diagnostic witness only: native probes can emit per-frame JSON metadata so Fensalir sees real sensor cadence in the rolling buffers while the direct ABI driver is being cut. It does not carry pixels and does not own the final six-camera hot path. Multi-camera probes are one process with declared accepted source ids, not one process per camera.

OBS Bridge Utility

flowchart TD
    A["config/localcast.json"] --> B["sender-start.ps1"]
    C["FFmpeg on sender"] --> D["Windows desktop capture"]
    C --> E["DirectShow audio capture"]
    D --> F["NVENC encode"]
    E --> G["audio encode"]
    F --> H["SRT video endpoint"]
    G --> I["SRT audio endpoint(s)"]
    H --> J["OBS Media Source"]
    I --> K["OBS Media Source per audio source"]

The bridge is useful because it is inspectable and already speaks OBS. It does not become the synchronized program authority.

Audio Field

The six-microphone path is separate from the bridge. Scarlett speaker loopback is the current timing authority when calibration chirplets are playing, but Scarlett production capture belongs on ASIO rather than WASAPI shared mode. MimirChirpletTimeline owns the birdsong-like timeline fingerprint: an order-3 de Bruijn sequence over 32 time/frequency constellation symbols. Start band, glide shape, duration, and following inter-chirp gap all carry code. Any three consecutive correctly detected symbols identify the event index inside the current operating horizon. Mimir continuously queues that timeline through Fensalir audio and decodes timing through the constrained chirplet-transform path. MimirAudioSynchronizationSettings.Mode chooses whether active calibration is allowed: chirp-only emits the witness, passive stays silent and uses program-audio phase correlation, and hybrid uses passive evidence by default while emitting active pilot chunks only when passive confidence is weak. Hybrid does not inherit the passive two-second analysis floor: passive still needs a longer program-audio window, while chirp-bin fallback can decode short pilot windows once at least a code-valid triplet is present. The chirp-bin detector now proposes events from dechirped bin energy rather than broadband RMS and uses a widened symbol spacing to keep sub-frame timing error from becoming symbol substitution. Chirp-bin frame timing is refined with a local matched-peak fit below one sample, and reports/states expose delayUs next to fractional sample delay. Mimir.BufferSmoke --hybrid-sync-self-test proves the analyzer path with a one-second rolling-buffer window and a recovered 317.375-sample delay at 0.369 us error. Reports now carry fractional delay and per-band matched energy. The first MimirChirpletSymbolCodebook owns separable symbol definitions; every symbol has a unique chirp shape, with rhythm as additional evidence. MimirChirpletStreamDecoder is the constrained chirplet-transform receiver: it turns a bounded PCM window into transform frames with multiple phase-invariant symbol candidates and refined per-candidate sample offsets, code-valid triplet anchors selected through gap/clock coherence, and an affine source clock fit. The synthetic Mimir.BufferSmoke --chirplet-self-test path renders canonical timeline audio into memory and currently recovers events 0-12 with a 47999.999990 Hz clock fit and 0.000014 sample MAE. The analyzer only emits timing reports from matched canonical anchors. The hub also owns cached reports plus smoothed per-source sync state with delay-slope/SRO in ppm. MimirRuntime runs analysis online as a bounded rotating service and can print live telemetry with MIMIR_SYNC_TELEMETRY_SECONDS. UI and telemetry are passive readers of cached sync state; they must not invoke the analyzer. Current app testing proves loopback wakeup, live mic buffers, and confident online sync states, but the next proof is stable canonical anchors through real mic streams. Camera mics are spatial/context witnesses; Focusrite devices are dialogue anchors. Fractional delay and the hot resampler belong in Faust/native DSP.

The WASAPI cadence probe is now a format/state diagnostic: it can request shared or exclusive sample rates, bit depths, channel counts, and float/PCM formats, then report the selected or closest format. The Focusrite driver stack is now installed and registers Focusrite USB ASIO. The Scarlett Solo 4th Gen is attached to Starfire and exposes 4 ASIO inputs / 2 outputs: Input 1, Input 2, Loopback 1, and Loopback 2. native/probes/asio_audio_cadence can instantiate that driver, verify 44.1-192 kHz support, and capture nonzero 4-channel Int32LSB callbacks at 192 kHz with 192-frame preferred buffers. The runtime analyzer accepts Float32, Int16, Int24, and Int32 PCM windows so ASIO/native capture can preserve the interface format. Raven also has a loopback-capable Scarlett ASIO path at 192 kHz for co-streamer/game timing evidence; Starfire still owns the heavy soundfield and sensor-fusion work.

Visual Fusion

Visual fusion belongs in Fensalir over current reservoir claims. Native capture workers provide frames; Fensalir owns feature extraction, matching, material fitting, render budgeting, and publication.

Known Risks

Windows device names vary by driver and localization.
Some FFmpeg builds omit SRT or NVENC.
Separate bridge endpoints can drift.
OBS SRT reconnection behavior can be fussy.
Direct driver work must prove sustained cadence before it becomes timing authority.
PS3 Eye audio endpoints are enumeration/runtime-fragile. A later replug made both mic buffers emit 480-frame WASAPI blocks again while both PS3 Eye camera buffers were live.

7.8 KiB Raw Blame History