NRSC-5 bounded fixture and partial live evidence¶
These are the 2026-08-29–31 source/artifact-specific checkpoints, condensed from the former roadmap. They are not new results or qualification of current source. The decoder roadmap retains all open 6B.5 gates.
The operator authorized continuous implementation through 6B.5E on 2026-08-29. The checklist remains open until repository validation and live operator acceptance both pass. All decoding remains browser-local and consumes only the bounded decoder stream supplied by the existing lane worker.
Recorded 2026-08-29 candidate: pinned
nrsc5 3.2.0 at a5c09723b135327aeef157770caae9aebf597547, hash-verified
FFTW 3.3.10 and FAAD2 2.11.2 inputs, audited FM/AM artifact
0557a72170c65a87..., and an isolated five-function runtime import surface.
The provenance-known upstream FM capture acquired sync, produced 45/46
validated events and 503,808 finite stereo 44.1 kHz audio frames with a
nonzero measured left/right difference, and processed at approximately 4.28
MS/s in 256 MiB on the Pi. All 221 browser regressions, exact-rate
channelizer checks across every capable capture rate, native gateway tests,
the production build, and the repository validation/secret scan pass. That FM
fixture contains no broadcast image, and no provenance-known positive AM
fixture is retained, so received imagery and positive AM remain explicit live
acceptance gates. No 6B.5 item is struck until the operator accepts the deployed
candidate.
The acceptance candidate preserves the NRSC-5 decoder's distinct left/right channels through the exact 44.1-to-48 kHz converter, stereo worklet ring buffers, and browser output node. Mono analog sources continue to be duplicated to both outputs, and source handoff remains exclusive and reset at boundaries.
A follow-up diagnostic candidate adds separate, never-automatic NRSC-5 FM/AM AddressSanitizer artifacts. The adapter, upstream decoder, FFTW, and FAAD2 are instrumented together; the final import audit still permits no tuner, network, or filesystem capability. Decoder faults retain bounded runtime output, artifact/build provenance, memory/input/timing/continuity facts, and recent protocol events in an expandable and exportable operator report. An explicitly selected sanitizer job also retains a 20-second decoder-rate IQ ring only for post-fault download and deterministic offline WASM replay. These items remain unstruck until live operator acceptance.
The diagnostic implementation candidate now binds independent SHA-256 records for the WASM, hosted Emscripten loader, and privacy-normalized source map. The map covers the adapter, NRSC-5, FFTW, and FAAD2/HDC without embedded source or operator home paths. An exact import audit admits only the reviewed hosted ASan/runtime surface; filesystem-like shims are verified fail-closed and no hardware, tuner, socket, or arbitrary network capability reaches the WASM. The worker verifies all three support artifacts before instantiation and keeps runtime output bounded to 256 KiB.
The fault workspace exposes artifact identity, current/maximum memory, input rate/bandwidth/domain, capture clock epoch and uncertainty, continuity/drop counters, expandable source-oriented runtime output, and the last bounded status plus protocol/metadata context. JSON export is separate from the 20-second raw-IQ replay. Replay refuses a mismatched WASM, loader, or source map. Raw captures and reports are ignored by Git and carry an operator privacy warning. Optimized jobs allocate neither the capture ring nor the hosted ASan runtime.
Repository qualification proves the hosted diagnostic path with the complete
50,331,648-byte upstream FM fixture, 45/46 validated events, 503,808 true-stereo
audio frames, and a test-only heap-use-after-free that resolves to the wrapper
source line. The sanitizer replay processed approximately 2.36 MS/s and grew
to 658.3 MiB; that diagnostic cost never reaches normal decoder jobs. Both
shipped FM/AM diagnostic artifacts instantiate through their content-addressed
loader/map pair, and deterministic .rfiq replay validates artifact provenance
before consuming samples. The deliberate fault trigger is fixture-only and
cannot appear in a published decoder. Live reproduction of the original fault
and confirmation of receiver/audio/display isolation remain the acceptance
gate, so no 6B.5 checkbox is struck yet.
Partial live validation on 2026-08-30 sustained HD audio for more than three to five minutes and allowed the operator to select between received HD programs. Subsequent live checkpoints through 2026-08-31 corrected decoder restart and input-continuity faults, preserved distinct stereo output, and exercised the normal NRSC-5 FM decoder at 100.7 MHz while station metadata, HD1/HD2 services, track art, station logos, TTN traffic objects, and Doppler radar objects were received over the air. The operator accepted the focused media and geographic presentation corrections, not the entire 6B.5 phase.
The operator workspace now de-duplicates received objects by SHA-256 while
retaining an expandable arrival history. Images and bounded text objects have
inline previews, explicit download, and enlargement. Now-playing metadata and
art remain scoped to the selected HD programme; station logos remain distinct
from track art and appear on their corresponding HD tiles. TMI and TMT objects
assemble the current 3-by-3 TTN traffic mosaic. Matching DWRI coordinates
register the newest DWRO image over that mosaic. Because a radar return can be
outside the smaller traffic-map footprint, the workspace measures legend-
matched pixels within the registered crop, reports Radar received · no
returns within map area when appropriate, and shows the full radar field with
the traffic footprint outlined instead of implying a failed overlay.
These are accepted checkpoints, not final Phase 6B.5 acceptance. Positive AM, the complete failure/reacquisition and rate/lane matrix, live diagnostic fault capture/replay, upstream-output comparison, resource benchmarks, and a final operator declaration remain open. Every 6B.5 checklist item therefore remains unstruck.