OKFS: KAI-4 P1 V=170 ear live + data-scale verdict

Author: nihilistau

memory-okf/LUT.md

@@ -204,3 +204,4 @@ Lookup before you build. `python tools/okf_mem.py lookup --root <root> <kw>`
 | c8b97c157c1d9c9e | agent | live-play 4,toolset swap aborted,system prompt stability law,spine_toolset off,coding tools index,token 0 divergence,one stable system prompt | SP_SPINE_TOOLSET per-turn swap rewrote the system prompt -> persist cache diverged at token 0 -> whole-conversation re-prefill on any 'build/code' mention. Profile: spine_toolset=false; coding tools now in the all_tools load-on-demand index (deduped). Adaptive things go in the message stream or tool index, never the system prefix. | GREEN | sum/c8b97c157c1d9c9e.md |
 | fdebbd495a530e25 | agent | KAI-4,latent voice,GNA ear,inject_frames,audio placeholder 258881,Audex teardown,discrete token reject,voice head,GAN vocoder,V_sub vocab,G-VOICE-0,voice channel GUI,voxtral reuse,E=3840,WSL no GNA MMIO | Sovereign voice: KAI-3 GNA ear (silicon-gated) wired to mic + /v1/voice + console -> inject_frames -> 12B (G-VOICE-0 parity 8/8, live e2e proven). Audex: steal unified-input + 2-stage output split, reject discrete-token output. P0=plumbing GREEN; P1 vocab scale-up makes free speech legible; P3 voice head = text bypass. E=3840 not 1024; GNA is frozen INT16 conv-only leverage. | GREEN | sum/fdebbd495a530e25.md |
 | 05e1cca71bf544b5 | agent | KAI-4 P1,ear bake,voice corpus,SAPI render,CPU render,vsub 170,held-out by sentence,data-scale ceiling,safetensors pagefile,cuda contention CPU default,voice_train,voice_bake,G-VOICE-1,hey shannon wake | Full ear-scaling pipeline built+automated CPU-only (2060 stays Gemma's): voice_corpus (templated + wake -> sp_tok_enc -> vsub<=508), voice_render_sapi (6-voice CPU, resumable, 2058 wavs/40s), voice_frames, voice_train (CPU default; NO safetensors mmap - blows pagefile; cuda contends with daemon). Vocab 32->170; unseen-sentence exact recog 6->22%. Held-out-by-sentence is data-bound (KAI-3 pattern). Bake running -> voice_ctc.pt. Remaining: scale corpus + ONNX/POT/GNA re-gate (G-VOICE-1) + re-point ear.py + wake head. | ACTIVE | sum/05e1cca71bf544b5.md |
+| c768dec51af383b7 | agent | KAI-4 P1 live,ear V170 serving,voice_export_ir,ONNX OV IR,ear auto-select,acoustic 0.73,unseen 0.21,data-scale bake,G-VOICE-1 remaining,onnx weights_only | 343-sentence bake done (acoustic 0.73/unseen 0.21). ckpt->ONNX->OV IR path built; ear.py auto-selects voice_ctc.xml; /v1/voice/status now V=170 SERVING. Full chain automated corpus->live inject. Legibility = pure corpus-size bake (KAI-3 pattern). Remaining G-VOICE-1: scale corpus + GNA POT re-gate + wake head. | ACTIVE | sum/c768dec51af383b7.md |

memory-okf/full/c768dec51af383b7.md

@@ -0,0 +1,19 @@
+---
+type: memory
+title: ADR-KAI4 P1: V=170 conversational ear LIVE (voice_ctc IR); engineering done, legibility=corpus-size bake
+description: 343-sentence bake done (acoustic 0.73/unseen 0.21). ckpt->ONNX->OV IR path built; ear.py auto-selects voice_ctc.xml; /v1/voice/status now V=170 SERVING. Full chain automated corpus->live inject. Legibility = pure corpus-size bake (KAI-3 pattern). Remaining G-VOICE-1: scale corpus + GNA POT re-gate + wake head.
+timestamp: 2026-07-10T17:29:03Z
+resource: kairos 0ee8721
+sp_status: ACTIVE
+sp_gate: V=170 ear live; G-VOICE-1 legibility bake pending
+sp_commit: kairos 0ee8721
+sp_repro: voice_bake.bat; voice_export_ir.py; /v1/voice/status
+mem_kind: agent
+mem_addr: c768dec51af383b7
+mem_verified: unverified
+mem_lifecycle: active
+tags: [KAI-4 P1 live, ear V170 serving, voice_export_ir, ONNX OV IR, ear auto-select, acoustic 0.73, unseen 0.21, data-scale bake, G-VOICE-1 remaining, onnx weights_only, agent, tier-2]
+mem_tier: full
+---
+
+ADR-KAI4 P1 continued (2026-07-11): the 343-sentence bake FINISHED (var/voice/voice_ctc.pt, best held-out-by-sentence tok_acc=0.210, ACOUSTIC seen-text tok_acc=0.730). Built the checkpoint->live path: tools/voice_export_ir.py (ckpt -> ONNX opset13 at static Tmax -> OV FP32 IR voice_ctc.xml; needs onnx pkg + torch.load weights_only=False for our numpy-vsub ckpt); harness/voice/ear.py now AUTO-SELECTS voice_ctc.xml over the KAI-3 POT IR; voice_export_wsub.py defaults to var/voice/vsub.npy. LIVE /v1/voice/status now reports V=170 (was 32) - the conversational ear is SERVING. Decode spot-check on a training wav ('hey shannon what time is it'): 5 on-manifold frames, partial token match - the honest data-scale ceiling (0.73 acoustic / 0.21 unseen-sentence at 343 sentences). VERDICT: the ENGINEERING is done end-to-end (corpus->render->frames->train->ONNX->OV->live ear->inject_frames->12B, all automated); LEGIBILITY is now a pure corpus-size bake (KAI-3 61->924 samples = 0.44->0.87; scale to thousands). REMAINING G-VOICE-1: (a) grow corpus to thousands + rerun voice_bake.bat; (b) ONNX->POT-i16->GNA_HW re-gate (audio_port ov scripts) to put the bigger ear back on silicon; (c) G-VOICE-WAKE dedicated 'Hey Shannon' always-on head. Then P2 vocoder, P3 voice head.

memory-okf/sum/c768dec51af383b7.md

@@ -0,0 +1,24 @@
+---
+type: memory
+title: ADR-KAI4 P1: V=170 conversational ear LIVE (voice_ctc IR); engineering done, legibility=corpus-size bake
+description: 343-sentence bake done (acoustic 0.73/unseen 0.21). ckpt->ONNX->OV IR path built; ear.py auto-selects voice_ctc.xml; /v1/voice/status now V=170 SERVING. Full chain automated corpus->live inject. Legibility = pure corpus-size bake (KAI-3 pattern). Remaining G-VOICE-1: scale corpus + GNA POT re-gate + wake head.
+timestamp: 2026-07-10T17:29:03Z
+resource: kairos 0ee8721
+sp_status: ACTIVE
+sp_gate: V=170 ear live; G-VOICE-1 legibility bake pending
+sp_commit: kairos 0ee8721
+sp_repro: voice_bake.bat; voice_export_ir.py; /v1/voice/status
+mem_kind: agent
+mem_addr: c768dec51af383b7
+mem_verified: unverified
+mem_lifecycle: active
+tags: [KAI-4 P1 live, ear V170 serving, voice_export_ir, ONNX OV IR, ear auto-select, acoustic 0.73, unseen 0.21, data-scale bake, G-VOICE-1 remaining, onnx weights_only, agent, tier-1]
+mem_tier: summary
+mem_full: c768dec51af383b7
+---
+
+# ADR-KAI4 P1: V=170 conversational ear LIVE (voice_ctc IR); engineering done, legibility=corpus-size bake
+
+343-sentence bake done (acoustic 0.73/unseen 0.21). ckpt->ONNX->OV IR path built; ear.py auto-selects voice_ctc.xml; /v1/voice/status now V=170 SERVING. Full chain automated corpus->live inject. Legibility = pure corpus-size bake (KAI-3 pattern). Remaining G-VOICE-1: scale corpus + GNA POT re-gate + wake head.
+
+Full context: [full/c768dec51af383b7.md](../full/c768dec51af383b7.md)