spike(v4): block transport selection pending native evidence

docs/decisions/0001-v4-transport-blocked.md

@@ -0,0 +1,72 @@
+# ADR 0001: V4 transport selection is blocked
+
+Status: blocked — no transport selected
+Date: 2026-08-02
+
+## Decision
+
+Neither custom-framed SRT message mode nor RTP/RTCP/SRTP is selected. Product
+transport implementation must not begin from this spike.
+
+The required comparison cannot be run reproducibly from this repository and
+host state. Selecting from the Phase 0 metadata impairment simulator would
+violate `docs/v4-transport-benchmark.md`, which says simulated impairment is
+harness validation rather than transport-selection evidence.
+
+## Reproduced blockers
+
+The checked-in audit reports these unmet gates:
+
+1. No pinned Windows x64 libsrt development package.
+2. No pinned Windows x64 libsrtp2 development package.
+3. No native custom-AU SRT and RTP/RTCP/SRTP probe exists.
+4. No identical real H.264/HEVC AU corpus preserves source/decode timestamps,
+   sequence, codec-config generation, orientation, payload bytes and CRC.
+5. No target Wi-Fi impairment topology/configuration was supplied. Local JSONL
+   loss injection is not a substitute.
+
+The V4 envelope documented before this spike also omits orientation even though
+the timestamp and master architecture require it. The native probe/corpus must
+resolve and version that field before measurements begin.
+
+## Commands and raw evidence
+
+Run from the repository root:
+
+```powershell
+python spikes/transport/audit.py --output out/transport-audit
+python -m pytest tests/test_v4_transport_audit.py
+```
+
+The first command intentionally exits 2 while blocked and writes:
+
+- `out/transport-audit/environment.json`: versions, queue declarations and
+  individual prerequisite verdicts.
+- `out/transport-audit/matrix.jsonl`: all 192 candidate/codec/profile/case/seed
+  cells, explicitly `NOT_RUN` with blocker IDs.
+
+The checked-in reproduction from this host is under
+`docs/evidence/v4-01-transport-audit/`. These files are prerequisite inventory,
+not raw transport measurements. The audit deliberately cannot turn a cell into
+a pass: native linking, real payload/CRC validation and target-network behavior
+must be established by the future benchmark runner.
+
+The matrix is the documented superset: clean, random loss, burst loss+jitter,
+explicit reorder, 500 ms outage, 5 s outage, bitrate step and isolated reconnect;
+both codecs; 4x1080p30 and 2x1080p60; three seeds.
+
+## Acceptance gate to unblock
+
+Pin the two native libraries; check in a spike-only native probe; provide the
+versioned real-AU corpus and target topology; then run every raw cell. Only the
+measured leader may run the required one-hour H.264, HEVC and 1080p60 soaks,
+including isolated camera failure. A leader may be selected only if timestamps,
+sequence, codec config, orientation and CRC are preserved; all queues remain
+within declared capacity/deadline/policy; gaps are explicit; memory slope is at
+most 1 MiB/hour; and copy, CPU, private-memory and maintenance evidence passes.
+
+## Scope and rollback
+
+This change adds audit/test/documentation files only. It does not implement a
+transport, re-encode media, or touch Android, OBS, engine, or legacy production
+paths. Rollback is removal of `spikes/transport`, its test, and this ADR.

docs/evidence/v4-01-transport-audit/environment.json

@@ -0,0 +1,78 @@
+{
+  "schema_version": 4,
+  "commit": "6c65476a7245e479a0cc9c2bfa6bd167d6cbd875",
+  "platform": "Windows-10-10.0.26200-SP0",
+  "python": "3.11.9",
+  "ffmpeg": "ffmpeg version 8.0.1-full_build-www.gyan.dev Copyright (c) 2000-2025 the FFmpeg developers",
+  "cmake": "cmake version 4.2.3",
+  "vcpkg_root": null,
+  "queues": [
+    {
+      "name": "encoded_au_ingress_per_session",
+      "capacity": 120,
+      "deadline_ms": 150,
+      "overflow_policy": "reject_past_deadline_and_declare_gap"
+    },
+    {
+      "name": "received_datagrams_per_session",
+      "capacity": 2048,
+      "deadline_ms": 150,
+      "overflow_policy": "discard_past_deadline_and_declare_gap"
+    },
+    {
+      "name": "fragment_reassembly_per_session",
+      "capacity": 120,
+      "capacity_bytes": 67108864,
+      "max_fragments_per_au": 4096,
+      "deadline_ms": 150,
+      "overflow_policy": "expire_incomplete_au_and_declare_gap"
+    },
+    {
+      "name": "assembled_video_per_session",
+      "capacity": 120,
+      "deadline_ms": 150,
+      "overflow_policy": "drop_oldest_non_key_request_keyframe_declare_gap"
+    },
+    {
+      "name": "retransmit_cache_per_session",
+      "capacity": 4096,
+      "deadline_ms": 120,
+      "overflow_policy": "expire_oldest_past_deadline_and_count_range"
+    },
+    {
+      "name": "telemetry",
+      "capacity": 1024,
+      "deadline_ms": 1000,
+      "overflow_policy": "coalesce_gauges_retain_fault_transitions"
+    }
+  ],
+  "prerequisites": [
+    {
+      "gate": "pinned_windows_libsrt",
+      "pass": false,
+      "detail": "VCPKG_ROOT is unset"
+    },
+    {
+      "gate": "pinned_windows_libsrtp2",
+      "pass": false,
+      "detail": "VCPKG_ROOT is unset"
+    },
+    {
+      "gate": "native_both_candidate_probe",
+      "pass": false,
+      "detail": "native probe absent"
+    },
+    {
+      "gate": "identical_real_au_corpus",
+      "pass": false,
+      "detail": "no --corpus supplied"
+    },
+    {
+      "gate": "target_wifi_impairment_topology",
+      "pass": false,
+      "detail": "no --topology supplied"
+    }
+  ],
+  "inputs_declared": false,
+  "decision": "NOT_RUN"
+}