OKFS: HINDSIGHT console-latency resolution (spine authority + session transcript + boot prewarm; 4.9s greet measured)

Author: nihilistau

memory-okf/LUT.md

@@ -198,3 +198,4 @@ Lookup before you build. `python tools/okf_mem.py lookup --root <root> <kw>`
 | 6baa9fa3efef7b2a | agent | audit-2026-07-10-pt2,agent stack,gateway 8800,tool calling dead,fence drift,tool_code space,python fence,run_gateway_system,run_console_agent,fastmcp,mcp server,mcp bridge,SP_MCP_TOOLS,gateway autodetect,persona editor,ADR-012 committed,numeric transcription residual | Agentic abilities dead = gateway :8800 not in the path (console posts :3000; run_gateway.bat armed nothing) PLUS reason-model fence drift ('` tool_code' space / '`python' wrappers) the parser rejected. Fixed both; console now autodetects the gateway + has a persona editor; FastMCP server+bridge shipped (G-MCP-SERVER 3/3); live-verified {tool}/{recall} events + real execution. ADR-012 committed ordered (36c2da3->4d307fd). Residual: model garbles numeric tool results when paraphrasing. | GREEN | sum/6baa9fa3efef7b2a.md |
 | 8ed4d23c31cab75d | agent | hindsight,kairos,SP_PERSIST_B4,persist under B4,agent turn latency,minutes to seconds,prewarm,prefix snapshot,numeric fidelity,rebuild,gemma4_tail_cpu link,build-cpu first,production repo,migration map,G-KAIROS-P0 | SP_B4_NIGHTSHIFT force-disabled persist-KV = the 2-5min agent turns. SP_PERSIST_B4=1 fix measured: 6.7-9.2s extensions, cache survives capture turns, gateway prewarm now functional. Post-tool low-temp verbatim rule fixes number garbling. Daemon clean-rebuilt (build-cpu FIRST for submodule changes). kairos repo born: HINDSIGHT charter + migration map, G-KAIROS-P0 GREEN. | GREEN | sum/8ed4d23c31cab75d.md |
 | 025b3ccc1c2fe113 | agent | hindsight execution,kairos P2,kairos P3,serve.py,profiles toml,launcher zoo replaced,cutover verified,MTP corrected,drafter blocked,prefix snapshot scoped,10 gates green in kairos | Harness+console+MCP migrated to kairos, 10/10 gates GREEN there; serve.py profile launcher replaces the bat zoo (effective-config echo, one-authority schema); cutover verified with persist timings holding (6.7-9.3s). MTP honest correction: real-text 0.87x, spec.rs unwired - drafter is the missing piece (P5). P1 prefix-snapshot scoped for a fresh session. | GREEN | sum/025b3ccc1c2fe113.md |
+| 3e76c3d6ca77f84f | agent | hindsight latency,console slow,auto_recall passthrough,SP_GATEWAY_AUTHORITY spine,session transcript,session_id,persist strict extension,prewarm at boot,4.9s greet,8.1s tool turn,numeric fidelity confirmed | 3 compounding causes: daemon-L5 authority cleared the persist cache every recalled turn (fix SP_GATEWAY_AUTHORITY=spine); stateless gateway diverged from the committed cache (fix session_id canonical transcript, mutate_messages); cold preamble on first msg (fix prewarm=true at boot). Measured console path: greet 4.9s, follow-up 5.8s, tool 8.1s (get_time copied exactly). PERSIST-KV reuse 1511/1530 suffix 1. | GREEN | sum/3e76c3d6ca77f84f.md |

memory-okf/full/3e76c3d6ca77f84f.md

@@ -0,0 +1,23 @@
+---
+type: memory
+title: HINDSIGHT: console latency RESOLVED live (minutes -> 4.9s greet / 8.1s tool turn)
+description: 3 compounding causes: daemon-L5 authority cleared the persist cache every recalled turn (fix SP_GATEWAY_AUTHORITY=spine); stateless gateway diverged from the committed cache (fix session_id canonical transcript, mutate_messages); cold preamble on first msg (fix prewarm=true at boot). Measured console path: greet 4.9s, follow-up 5.8s, tool 8.1s (get_time copied exactly). PERSIST-KV reuse 1511/1530 suffix 1.
+timestamp: 2026-07-10T05:22:12Z
+resource: kairos a93ae4a
+sp_status: GREEN
+sp_gate: live console-path probe: 4.9/5.8/8.1s + PERSIST-KV strict-extension logs
+sp_commit: kairos a93ae4a
+sp_repro: serve.bat agent (prewarm at boot); POST /v1/chat with session_id + auto_recall:true
+mem_kind: agent
+mem_addr: 3e76c3d6ca77f84f
+mem_verified: unverified
+mem_lifecycle: active
+tags: [hindsight latency, console slow, auto_recall passthrough, SP_GATEWAY_AUTHORITY spine, session transcript, session_id, persist strict extension, prewarm at boot, 4.9s greet, 8.1s tool turn, numeric fidelity confirmed, agent, tier-2]
+mem_tier: full
+---
+
+HINDSIGHT latency FULLY RESOLVED (2026-07-10, live-measured on the console path). The user reported minutes-then-[aborted] on EVERY console turn; transcript showed authority L5 on all turns. THREE compounding causes, all fixed:
+(1) DAEMON-L5 AS THE AUTHORITY: the console checkbox sends auto_recall:true; the gateway passed it through so daemon-side L5 delivered recall by RE-PREFILLING an augmented prompt and then CLEARING the persist committed cache (routes.rs recalled-turn clear) -> every interrogative turn cost ~2 full ~1.6k-token preamble prefills and the NEXT turn a third. FIX: SP_GATEWAY_AUTHORITY=spine (kairos agent profile) makes the HARNESS the one recall authority + refuses the auto_recall passthrough; spine injects its note BEFORE the new user msg = strict extension. Default 'l5' unchanged.
+(2) STATELESS GATEWAY: the client echoed its own history, which never matched what the daemon saw (recall notes + tool rounds are transient) -> the turn after any recall/tool turn diverged from the committed cache = full re-prefill. FIX: session_id -> gateway keeps the CANONICAL append-only transcript (agent_chat_stream mutate_messages=True); daemon sees a strict extension every turn. Console sends session_id (rotates on clear).
+(3) COLD PREAMBLE ON THE FIRST MESSAGE: prewarm was opt-in and off -> the one-time ~81s preamble prefill landed on the user's first turn. FIX: prewarm=true in the profile -> paid at BOOT in the background.
+MEASURED after all three (warm, console path, auto_recall:true like the checkbox): fresh-conversation greet 4.9s (was 105s+), follow-up 5.8s, TOOL TURN 8.1s with the get_time result copied EXACTLY ('1521.' - numeric-fidelity fix confirmed). PERSIST-KV logs: reuse 1511/1530 suffix 1 (one-token prefill on an extension). The minutes-long turns are gone. Residual: spine recall fires on chit-chat greetings and pulls low-value captured episodes (quality noise, not latency) -> add a QONLY-style abstain on the harness side (kairos P2 polish).

memory-okf/sum/3e76c3d6ca77f84f.md

@@ -0,0 +1,24 @@
+---
+type: memory
+title: HINDSIGHT: console latency RESOLVED live (minutes -> 4.9s greet / 8.1s tool turn)
+description: 3 compounding causes: daemon-L5 authority cleared the persist cache every recalled turn (fix SP_GATEWAY_AUTHORITY=spine); stateless gateway diverged from the committed cache (fix session_id canonical transcript, mutate_messages); cold preamble on first msg (fix prewarm=true at boot). Measured console path: greet 4.9s, follow-up 5.8s, tool 8.1s (get_time copied exactly). PERSIST-KV reuse 1511/1530 suffix 1.
+timestamp: 2026-07-10T05:22:12Z
+resource: kairos a93ae4a
+sp_status: GREEN
+sp_gate: live console-path probe: 4.9/5.8/8.1s + PERSIST-KV strict-extension logs
+sp_commit: kairos a93ae4a
+sp_repro: serve.bat agent (prewarm at boot); POST /v1/chat with session_id + auto_recall:true
+mem_kind: agent
+mem_addr: 3e76c3d6ca77f84f
+mem_verified: unverified
+mem_lifecycle: active
+tags: [hindsight latency, console slow, auto_recall passthrough, SP_GATEWAY_AUTHORITY spine, session transcript, session_id, persist strict extension, prewarm at boot, 4.9s greet, 8.1s tool turn, numeric fidelity confirmed, agent, tier-1]
+mem_tier: summary
+mem_full: 3e76c3d6ca77f84f
+---
+
+# HINDSIGHT: console latency RESOLVED live (minutes -> 4.9s greet / 8.1s tool turn)
+
+3 compounding causes: daemon-L5 authority cleared the persist cache every recalled turn (fix SP_GATEWAY_AUTHORITY=spine); stateless gateway diverged from the committed cache (fix session_id canonical transcript, mutate_messages); cold preamble on first msg (fix prewarm=true at boot). Measured console path: greet 4.9s, follow-up 5.8s, tool 8.1s (get_time copied exactly). PERSIST-KV reuse 1511/1530 suffix 1.
+
+Full context: [full/3e76c3d6ca77f84f.md](../full/3e76c3d6ca77f84f.md)