This earlier GEPA dashboard now points to the final cardinality-safe report.
+Open final reportOpen legacy GEPA-six dashboardOpen prop20 iteration graphOpen prompt diffs
+Promoted artifact: 2026-06-14-gepa-12b-prop20-cardinality-repair.
The optimizer artifacts contain one completed proposal step per GEPA run. The visible trajectory is therefore candidate 0, the seed/inherited prompt, to candidate 1, the reflected prompt produced by GEPA.
+These are not label-count reward graphs. The score is the localpager classifier scoring function: it penalizes false positives, false negatives, and over-labeling. Extra random labels cannot improve the score unless they match the gold labels in ds4.jsonl, and false positives are weighted as the more costly error.
This is the direct score-over-iteration view from each run's gepa-result.json. Candidate 0 is the baseline/inherited prompt; candidate 1 is the GEPA-reflected candidate when a proposal completed.
| Run | Candidate | Validation score | Best? | Discovery eval count |
|---|---|---|---|---|
| E4B smoke GEPA | 0 | 0.2500 | 0 | |
| E4B smoke GEPA | 1 | 0.6250 | yes | 4 |
| 12B six-row GEPA (selected) | 0 | 0.2857 | 0 | |
| 12B six-row GEPA (selected) | 1 | 0.5556 | yes | 10 |
| 12B twelve-row init-only | 0 | 0.4244 | yes | 0 |
| 12B twelve-row continuation GEPA | 0 | 0.5375 | 0 | |
| 12B twelve-row continuation GEPA | 1 | 0.6101 | yes | 20 |
This uses run_log.json. It shows the score on the exact subsample GEPA reflected on, before and after the proposed prompt.
| Run | Iteration | Candidate transition | Subsample ids | Before mean | After mean | Delta |
|---|---|---|---|---|---|---|
| E4B smoke GEPA | 0 | 0 -> 1 | [0] | 0.1429 | 1.0000 | 0.8571 |
| 12B six-row GEPA (selected) | 0 | 0 -> 1 | [4, 1] | 0.2679 | 1.0000 | 0.7321 |
| 12B twelve-row continuation GEPA | 0 | 0 -> 1 | [1, 10, 9, 5] | 0.6750 | 0.8750 | 0.2000 |
This is the scientific sanity check: the selected 12B-six candidate compared against the v9.1 seed as more of Shaun's 60-row set is accumulated. This graph is outside the GEPA training loop.
+ +| Rows added | Checkpoint rows | Seed segment | GEPA segment | Segment delta | Seed cumulative | GEPA cumulative | Cumulative delta |
|---|---|---|---|---|---|---|---|
| 1-6 | 6 | 0.5214 | 0.5833 | 0.0619 | 0.5214 | 0.5833 | 0.0619 |
| 7-12 | 12 | 0.4833 | 0.4917 | 0.0083 | 0.5024 | 0.5375 | 0.0351 |
| 13-18 | 18 | 0.5000 | 0.5833 | 0.0833 | 0.5016 | 0.5528 | 0.0512 |
| 19-30 | 30 | 0.3946 | 0.4552 | 0.0605 | 0.4588 | 0.5137 | 0.0549 |
| 31-60 | 60 | 0.4056 | 0.4687 | 0.0630 | 0.4322 | 0.4912 | 0.0590 |
These show which validation rows changed inside the GEPA run, not just the aggregate mean.
+| Validation row | candidate 0 | candidate 1 best |
|---|---|---|
| 0 | 0.2500 | 1.0000 |
| 1 | 0.2500 | 0.2500 |
| Validation row | candidate 0 | candidate 1 best |
|---|---|---|
| 0 | 0.2500 | 0.3333 |
| 1 | 0.2500 | 1.0000 |
| 2 | 0.1429 | 0.2500 |
| 3 | 0.2857 | 0.2500 |
| 4 | 0.2857 | 1.0000 |
| 5 | 0.5000 | 0.5000 |
| Validation row | candidate 0 best |
|---|---|
| 0 | 0.5000 |
| 1 | 0.2500 |
| 2 | 0.1429 |
| 3 | 0.0000 |
| 4 | 1.0000 |
| 5 | 0.5000 |
| 6 | 0.2500 |
| 7 | 0.5000 |
| 8 | 0.5000 |
| 9 | 1.0000 |
| 10 | 0.2000 |
| 11 | 0.2500 |
| Validation row | candidate 0 | candidate 1 best |
|---|---|---|
| 0 | 0.5000 | 0.2500 |
| 1 | 0.5000 | 1.0000 |
| 2 | 0.2500 | 0.2500 |
| 3 | 0.2500 | 0.2857 |
| 4 | 1.0000 | 1.0000 |
| 5 | 0.5000 | 1.0000 |
| 6 | 1.0000 | 0.2500 |
| 7 | 0.5000 | 0.5000 |
| 8 | 0.5000 | 0.5000 |
| 9 | 1.0000 | 1.0000 |
| 10 | 0.2000 | 1.0000 |
| 11 | 0.2500 | 0.2857 |
iteration-score-data.json iteration-score-summary.csv
+Candidate trees from GEPA's built-in visualization: 12B six, 12B twelve continuation, E4B smoke.
+ds4.jsonl. Evaluator: local 12B, concurrency 2.This summary dashboard is the final validation view. For GEPA score changes across candidates/iterations, open the focused graph page.
+ +| Candidate | Rows | Mean | Precision | Recall | F1 | TP | FP | FN | Over | Failures |
|---|---|---|---|---|---|---|---|---|---|---|
| v9.1 Seed | 60 | 0.4322 | 0.7411 | 0.5804 | 0.6510 | 83 | 29 | 60 | 2 | 3 | +
| GEPA 12B Six Candidate | 60 | 0.4912 | 0.7965 | 0.5960 | 0.6818 | 90 | 23 | 61 | 1 | 0 | +
| GEPA 12B Twelve Candidate (rows 1-30 only) | 30 | 0.4800 | 0.7377 | 0.7258 | 0.7317 | 45 | 16 | 17 | 6 | 2 | +
The selected artifact is GEPA 12B Six Candidate. It is more precise, has fewer false positives, and eliminated structural failures. The main caveat is one additional false negative.
| Rows | v9.1 seed | GEPA six | Delta | GEPA twelve | E4B smoke |
|---|---|---|---|---|---|
| 1-6 | 0.5214 | 0.5833 | +0.0619 | 0.5158 | 0.3607 |
| 7-12 | 0.4833 | 0.4917 | +0.0083 | n/a | n/a |
| 13-18 | 0.5000 | 0.5833 | +0.0833 | 0.5417 | n/a |
| 19-30 | 0.3946 | 0.4552 | +0.0605 | 0.4134 | n/a |
| 31-60 | 0.4056 | 0.4687 | +0.0630 | n/a | n/a |
| Target | Title | Seed | GEPA | Delta | Gold | Seed pred | GEPA pred | Seed FP/FN | GEPA FP/FN | Error |
|---|---|---|---|---|---|---|---|---|---|---|
| #77748 | +fix: Codex startup plugins + WhatsApp history & Docker Codex OAuth | +1.0000 | 0.2500 | -0.7500 | +codex chat_integrations | +chat_integrations codex | +codex gateway | +none FN none |
+gateway FN chat_integrations |
++ |
| #63007 | +Pass outbound session identity into message_sending and surface guarded gateway send denial | +1.0000 | 0.2500 | -0.7500 | +gateway sessions | +gateway sessions | +gateway hooks | +none FN none |
+hooks FN sessions |
++ |
| #59878 | +Session lane stuck in 'running' after run dies ā sessions.abort + gateway restart fail to clear stale state | +1.0000 | 0.2500 | -0.7500 | +sessions reliability | +sessions reliability | +sessions gateway | +none FN none |
+gateway FN reliability |
++ |
| #48940 | +ACP: add gateway-owned node-backed runtime | +1.0000 | 0.5000 | -0.5000 | +acp gateway agent_runtime | +acp gateway agent_runtime | +acp gateway | +none FN none |
+none FN agent_runtime |
++ |
| #77827 | +fix: LM Studio thinking blocks invisible with Responses API | +1.0000 | 0.5000 | -0.5000 | +model_serving local_models | +local_models model_serving | +model_serving | +none FN none |
+none FN local_models |
++ |
| #70882 | +fix(bundle-mcp): coerce stringified object/array params before MCP tool calls | +1.0000 | 0.5000 | -0.5000 | +mcp_tooling tool_calling | +mcp_tooling tool_calling | +mcp_tooling | +none FN none |
+none FN tool_calling |
++ |
| #84746 | +[Bug]: Auto-compaction crashes active responses after 5.18 transcript lock scope change (#13744) | +1.0000 | 0.5000 | -0.5000 | +reliability sessions | +sessions reliability | +sessions | +none FN none |
+none FN reliability |
++ |
| #47083 | +fix: respect totalTokensFresh flag to avoid showing stale token counts | +0.5000 | 0.2000 | -0.3000 | +sessions telemetry_usage | +telemetry_usage | +ui_tui | +none FN sessions |
+ui_tui FN sessions telemetry_usage |
++ |
| #65242 | +fix: CompletionDeliveryGate to prevent duplicate ACP completion delivery | +0.5000 | 0.2000 | -0.3000 | +acp coding_agents reliability | +acp reliability | +acp notifications | +none FN coding_agents |
+notifications FN coding_agents reliability |
++ |
| #69256 | +fix(cron): prevent premature session cleanup when subagents are running | +0.5000 | 0.3333 | -0.1667 | +cron_automation sessions reliability | +cron_automation sessions | +cron_automation | +none FN reliability |
+none FN sessions reliability |
++ |
| #81249 | +[Feature/Bug]: Local Ollama embeddings fail when proxy is enabled (SSRF defenses ignore NO_PROXY) | +0.2500 | 0.1429 | -0.1071 | +local_models self_hosted_inference | +local_models security | +security config | +security FN self_hosted_inference |
+security config FN local_models self_hosted_inference |
++ |
| #84477 | +Discord embedded-run prep wedge before strict-agentic, recovery skips sessionId=unknown lanes | +0.2500 | 0.2000 | -0.0500 | +sessions agent_runtime reliability | +chat_integrations sessions reliability | +gateway sessions | +chat_integrations FN agent_runtime |
+gateway FN agent_runtime reliability |
++ |
| #44379 | +fix(pi-runner): harden context-overflow recovery with one suppress-hook retry | +0.1667 | 0.1429 | -0.0238 | +coding_agents memory hooks reliability | +agent_runtime reliability | +agent_runtime | +agent_runtime FN coding_agents memory hooks |
+agent_runtime FN coding_agents memory hooks reliability |
++ |
| #40332 | +[Feature]: Per-binding and per-agent permissionMode for ACP sessions | +0.5000 | 0.5000 | +0.0000 | +acp approvals acpx | +approvals acp | +acp approvals | +none FN acpx |
+none FN acpx |
++ |
| #80255 | +fix #79026: active-memory recall subagent can deadlock on the main lane inside before_prompt_build | +0.5000 | 0.5000 | +0.0000 | +memory reliability | +memory | +memory | +none FN reliability |
+none FN reliability |
++ |
| #84670 | +[codex] fix webchat full-message reader for truncated history | +0.5000 | 0.5000 | +0.0000 | +gateway api_surface ui_tui | +gateway ui_tui | +gateway ui_tui | +none FN api_surface |
+none FN api_surface |
++ |
| #62428 | +test(exec): land exec v2 contract follow-through | +0.2000 | 0.2000 | +0.0000 | +exec_tools sandboxing approvals | +exec_tools security | +exec_tools security | +security FN sandboxing approvals |
+security FN sandboxing approvals |
++ |
| #90146 | +google-vertex: Missing gemini-3.1-flash-lite in provider catalog causes silent failure instead of error | +0.2500 | 0.2500 | +0.0000 | +local_model_providers reliability | +model_serving reliability | +config reliability | +model_serving FN local_model_providers |
+config FN local_model_providers |
++ |
| #51849 | +Docs: add freeCodeCamp OpenClaw full tutorial to showcase | +1.0000 | 1.0000 | +0.0000 | +docs | +docs | +docs | +none FN none |
+none FN none |
++ |
| #84297 | +[Bug]: Per-agent identity overlay dropped on cron --announce and heartbeat target-channel Slack pushes (announce path; reply path was fixed in #38235) | +0.2500 | 0.2500 | +0.0000 | +notifications chat_integrations | +chat_integrations cron_automation | +cron_automation chat_integrations | +cron_automation FN notifications |
+cron_automation FN notifications |
++ |
| #81957 | +ci: harden GitHub Actions supply-chain boundaries | +0.2857 | 0.2857 | +0.0000 | +security | +security packaging_deployment | +security packaging_deployment | +packaging_deployment FN none |
+packaging_deployment FN none |
++ |
| #87277 | +[Feature] Add MiMo-V2.5 to Xiaomi catalog + automatic multimodal routing when DeepSeek V4-Pro is primary model | +0.2500 | 0.2500 | +0.0000 | +local_model_providers model_serving | +model_releases model_serving | +config model_serving | +model_releases FN local_model_providers |
+config FN local_model_providers |
++ |
| #64199 | +[Bug]: ACP configured binding uses parent channel ID for session key ā all threads under same channel share one persistent Claude Code process | +1.0000 | 1.0000 | +0.0000 | +acp sessions | +acp sessions | +acp sessions | +none FN none |
+none FN none |
++ |
| #84583 | +cron announce delivery triggers EmbeddedAttemptSessionTakeoverError when user is actively chatting | +0.5000 | 0.5000 | +0.0000 | +cron_automation sessions reliability | +sessions reliability | +cron_automation sessions | +none FN cron_automation |
+none FN reliability |
++ |
| #67244 | +Explicit ACP agent runs: embedded backend visibility failure and stale final JSON state after sessions_yield | +0.2500 | 0.2500 | +0.0000 | +acpx acp | +acp sessions | +acp sessions | +sessions FN acpx |
+sessions FN acpx |
++ |
| #73910 | +BUG: OpenClaw-managed Codex ACP uses isolated CODEX_HOME without auth bridge and sends unsupported timeout config | +0.3333 | 0.3333 | +0.0000 | +codex acp acpx auth_identity | +acp codex | +codex auth_identity | +none FN acpx auth_identity |
+none FN acp acpx |
++ |
| #80008 | +feat(plugins): expose ACP spawn and prompt in plugin runtime | +0.2500 | 0.2500 | +0.0000 | +acp coding_agents | +acp skills_plugins | +acp skills_plugins | +skills_plugins FN coding_agents |
+skills_plugins FN coding_agents |
++ |
| #60979 | +feature: sessions_spawn ACP delivery to channel (stream output to Zulip/Discord topic) | +0.5000 | 0.5000 | +0.0000 | +acp chat_integrations sessions | +acp chat_integrations | +acp chat_integrations | +none FN sessions |
+none FN sessions |
++ |
| #84715 | +[Bug]: @openclaw/codex peer link failure reproduced on 2026.5.19 after update | +0.5000 | 0.5000 | +0.0000 | +codex packaging_deployment | +codex | +codex | +none FN packaging_deployment |
+none FN packaging_deployment |
++ |
| #84757 | +[Bug]: Telegram session can get stuck after compaction when encrypted reasoning content fails verification | +0.5000 | 0.5000 | +0.0000 | +sessions chat_integrations reliability | +chat_integrations sessions | +chat_integrations sessions | +none FN reliability |
+none FN reliability |
++ |
| #56442 | +feat: Add opt-in ACP parent completion notify for sessions_spawn | +0.5000 | 0.5000 | +0.0000 | +acp sessions agent_runtime | +acp sessions | +acp sessions | +none FN agent_runtime |
+none FN agent_runtime |
++ |
| #84763 | +fix(acpx): scrub provider credential env from ACP harness spawns | +0.5000 | 0.5000 | +0.0000 | +acpx acp security | +acp security | +acpx security | +none FN acpx |
+none FN acp |
++ |
| #65364 | +feat(plugins): add registerProviderRuntimeAuthOverride API | +0.2500 | 0.2500 | +0.0000 | +auth_identity api_surface | +skills_plugins auth_identity | +skills_plugins auth_identity | +skills_plugins FN api_surface |
+skills_plugins FN api_surface |
++ |
| #52747 | +fix(acp): time out stuck session lane tasks | +0.5000 | 0.5000 | +0.0000 | +acp sessions reliability | +acp sessions | +acp reliability | +none FN reliability |
+none FN sessions |
++ |
| #10467 | +[Feature Request]: Multi-lane concurrency support for sub-agents via sessions_spawn | +0.2000 | 0.2000 | +0.0000 | +queueing sessions coding_agents | +queueing agent_runtime | +queueing agent_runtime | +agent_runtime FN sessions coding_agents |
+agent_runtime FN sessions coding_agents |
++ |
| #45393 | +fix(errors): friendly message and last-message repair for tool_use/tool_result mismatch (#45385) | +0.2000 | 0.2000 | +0.0000 | +tool_calling coding_agents reliability | +tool_calling security | +tool_calling security | +security FN coding_agents reliability |
+security FN coding_agents reliability |
++ |
| #84771 | +Event loop saturation during startup: synchronous model-prewarm and session-locks block event loop for 28-64 seconds | +1.0000 | 1.0000 | +0.0000 | +reliability sessions | +reliability sessions | +sessions reliability | +none FN none |
+none FN none |
++ |
| #68187 | +SSE-backed MCP sessions can stay stale after server restart and fail with 'Session not found' | +0.5000 | 0.5000 | +0.0000 | +mcp_tooling sessions gateway | +mcp_tooling sessions | +mcp_tooling sessions | +none FN gateway |
+none FN gateway |
++ |
| #52249 | +ACP parent session stuck until refresh when yielded waiting for child completion | +0.5000 | 0.5000 | +0.0000 | +acp sessions reliability | +acp sessions | +acp sessions | +none FN reliability |
+none FN reliability |
++ |
| #44202 | +[Bug]: local memory embeddings on Apple Silicon can crash gateway in ggml-metal / node-llama-cpp; need official Metal/GPU guidance | +0.1429 | 0.2000 | +0.0571 | +local_models memory self_hosted_inference | +memory gateway reliability | +memory reliability | +gateway reliability FN local_models self_hosted_inference |
+reliability FN local_models self_hosted_inference |
++ |
| #42027 | +fix: resolve exec PATH fallback, layered browser diagnostics, and cron force-run deadlock | +0.1429 | 0.2500 | +0.1071 | +exec_tools browser_automation cron_automation | +exec_tools gateway ui_tui | +exec_tools cron_automation ui_tui | +gateway ui_tui FN browser_automation cron_automation |
+ui_tui FN browser_automation |
++ |
| #43765 | +Improve runtime recovery for heartbeat, Feishu, and exec sessions | +0.1429 | 0.2500 | +0.1071 | +reliability exec_tools cron_automation | +reliability agent_runtime chat_integrations | +reliability gateway exec_tools | +agent_runtime chat_integrations FN exec_tools cron_automation |
+gateway FN cron_automation |
++ |
| #39248 | +Bug: sandbox.mode: "non-main" silently breaks sessions_spawn subagent initialization | +0.2000 | 0.3333 | +0.1333 | +coding_agents sandboxing agent_runtime | +sandboxing sessions | +sandboxing | +sessions FN coding_agents agent_runtime |
+none FN coding_agents agent_runtime |
++ |
| #51667 | +Feature: Native Audio Input for Omni-Modal Models (skip STT transcription) | +0.3333 | 0.5000 | +0.1667 | +model_serving security config | +model_serving | +model_serving config | +none FN security config |
+none FN security |
++ |
| #68725 | +feat(amazon-bedrock-mantle): add known context windows for open-weight Mantle models | +0.2500 | 0.5000 | +0.2500 | +open_weight_models local_model_providers | +open_weight_models model_serving | +open_weight_models | +model_serving FN local_model_providers |
+none FN local_model_providers |
++ |
| #43246 | +fix(message): deny same-provider cross-context sends by default [AI-assisted] | +0.0000 | 0.2500 | +0.2500 | +tool_calling security | +none | +security config | +none FN none |
+config FN tool_calling |
+seed fail | +
| #82507 | +[Feature]: ACPX Codex sandbox should inherit user-installed plugins (e.g. Superpowers) | +0.2000 | 0.5000 | +0.3000 | +acpx codex skills_plugins | +acpx security | +acpx skills_plugins | +security FN codex skills_plugins |
+none FN codex |
++ |
| #83863 | +ACP/Codex child tasks can be marked succeeded with progress-only output and no final deliverable | +0.2000 | 0.5000 | +0.3000 | +acp codex agent_runtime | +acp reliability | +acp codex | +reliability FN codex agent_runtime |
+none FN agent_runtime |
++ |
| #48580 | +Bug: acpx codex sessions åå»ŗēä¼čÆē«å³éåŗ - stdin is not a terminal | +0.2000 | 0.5000 | +0.3000 | +acpx codex sessions | +acp sessions | +codex sessions | +acp FN acpx codex |
+none FN acpx |
++ |
| #46552 | +docs(queue): clarify steer behavior with partial streaming and tool boundaries | +0.5000 | 1.0000 | +0.5000 | +queueing docs | +docs | +docs queueing | +none FN queueing |
+none FN none |
++ |
| #63826 | +security: fix HIGH/CRITICAL vulns in skill scanner, SSRF, hook priority, and token verification | +0.0000 | 0.5000 | +0.5000 | +security hooks skills_plugins | +none | +security hooks | +none FN none |
+none FN skills_plugins |
+seed fail | +
| #70529 | +[Bug]: Desktop cannot use existing Chrome sessions: EasyClaw Google sign-in fails, and user profile attach fails with spawn npx ENOENT | +0.5000 | 1.0000 | +0.5000 | +browser_automation packaging_deployment | +browser_automation | +browser_automation packaging_deployment | +none FN packaging_deployment |
+none FN none |
++ |
| #84752 | +fix: self-heal lane wedges + restore openai-codex OAuth on embedded path | +0.0000 | 0.5000 | +0.5000 | +reliability auth_identity sessions | +none | +reliability auth_identity | +none FN none |
+none FN sessions |
+seed fail | +
| #78528 | +Security: skill SecretRef API keys still leak into exec child environments | +0.5000 | 1.0000 | +0.5000 | +security exec_tools skills_plugins | +security exec_tools | +security exec_tools skills_plugins | +none FN skills_plugins |
+none FN none |
++ |
| #74305 | +[Bug]: ACPX Codex worker fails when model/thinking overrides are configured | +0.5000 | 1.0000 | +0.5000 | +acpx codex | +acpx | +acpx codex | +none FN codex |
+none FN none |
++ |
| #79897 | +OpenAI-compatible streaming with llama.cpp saves zero usage (stream closed before final usage chunk) | +0.2857 | 1.0000 | +0.7143 | +model_serving | +telemetry_usage model_serving | +model_serving | +telemetry_usage FN none |
+none FN none |
++ |
| #80479 | +feat(memory/embeddings): add openai-compatible provider for self-hosted servers (llama.cpp, Ollama, vLLM, TGI, LocalAI) | +0.2500 | 1.0000 | +0.7500 | +self_hosted_inference memory | +local_models self_hosted_inference | +memory self_hosted_inference | +local_models FN memory |
+none FN none |
++ |
| #71216 | +Config schema: add `sandbox`, `routing.rules`, `instances`, and `gateway.nodes.denyPaths` | +0.2500 | 1.0000 | +0.7500 | +config sandboxing gateway | +sandboxing model_serving gateway | +config sandboxing gateway | +model_serving FN config |
+none FN none |
++ |
| #84789 | +Active memory crashes on Telegram forum topic sessions (dirName validation) | +0.2500 | 1.0000 | +0.7500 | +memory sessions | +memory chat_integrations | +memory sessions | +chat_integrations FN sessions |
+none FN none |
++ |
| #80783 | +Policy: add model, network, and MCP conformance checks | +0.2000 | 1.0000 | +0.8000 | +mcp_tooling config security | +mcp_tooling local_model_providers | +mcp_tooling security config | +local_model_providers FN config security |
+none FN none |
++ |
Cells in TP/FP/FN order. Rows are sorted by F1 delta, worst first.
+| Topic | Seed TP/FP/FN | Seed P | Seed R | Seed F1 | GEPA TP/FP/FN | GEPA P | GEPA R | GEPA F1 | ĪF1 | ĪFP | ĪFN |
|---|---|---|---|---|---|---|---|---|---|---|---|
| local_models | +2/1/1 | 0.6667 | 0.6667 | 0.6667 | +0/0/3 | 0.0000 | 0.0000 | 0.0000 | +-0.6667 | -1 | +2 |
| telemetry_usage | +1/1/0 | 0.5000 | 1.0000 | 0.6667 | +0/0/1 | 0.0000 | 0.0000 | 0.0000 | +-0.6667 | -1 | +1 |
| tool_calling | +2/0/0 | 1.0000 | 1.0000 | 1.0000 | +1/0/2 | 1.0000 | 0.3333 | 0.5000 | +-0.5000 | +0 | +2 |
| reliability | +9/2/6 | 0.8182 | 0.6000 | 0.6923 | +5/1/11 | 0.8333 | 0.3125 | 0.4545 | +-0.2378 | -1 | +5 |
| agent_runtime | +1/3/4 | 0.2500 | 0.2000 | 0.2222 | +0/2/5 | 0.0000 | 0.0000 | 0.0000 | +-0.2222 | -1 | +1 |
| ui_tui | +1/1/0 | 0.5000 | 1.0000 | 0.6667 | +1/2/0 | 0.3333 | 1.0000 | 0.5000 | +-0.1667 | +1 | +0 |
| gateway | +4/2/1 | 0.6667 | 0.8000 | 0.7273 | +4/4/1 | 0.5000 | 0.8000 | 0.6154 | +-0.1119 | +2 | +0 |
| sessions | +14/2/4 | 0.8750 | 0.7778 | 0.8235 | +12/1/7 | 0.9231 | 0.6316 | 0.7500 | +-0.0735 | -1 | +3 |
| acp | +13/1/0 | 0.9286 | 1.0000 | 0.9630 | +11/0/2 | 1.0000 | 0.8462 | 0.9167 | +-0.0463 | -1 | +2 |
| local_model_providers | +0/1/3 | 0.0000 | 0.0000 | 0.0000 | +0/0/3 | 0.0000 | 0.0000 | 0.0000 | ++0.0000 | -1 | +0 |
| model_releases | +0/1/0 | 0.0000 | 0.0000 | 0.0000 | +0/0/0 | 0.0000 | 0.0000 | 0.0000 | ++0.0000 | -1 | +0 |
| notifications | +0/0/1 | 0.0000 | 0.0000 | 0.0000 | +0/1/1 | 0.0000 | 0.0000 | 0.0000 | ++0.0000 | +1 | +0 |
| api_surface | +0/0/2 | 0.0000 | 0.0000 | 0.0000 | +0/0/2 | 0.0000 | 0.0000 | 0.0000 | ++0.0000 | +0 | +0 |
| approvals | +1/0/1 | 1.0000 | 0.5000 | 0.6667 | +1/0/1 | 1.0000 | 0.5000 | 0.6667 | ++0.0000 | +0 | +0 |
| browser_automation | +1/0/1 | 1.0000 | 0.5000 | 0.6667 | +1/0/1 | 1.0000 | 0.5000 | 0.6667 | ++0.0000 | +0 | +0 |
| coding_agents | +0/0/6 | 0.0000 | 0.0000 | 0.0000 | +0/0/6 | 0.0000 | 0.0000 | 0.0000 | ++0.0000 | +0 | +0 |
| docs | +2/0/0 | 1.0000 | 1.0000 | 1.0000 | +2/0/0 | 1.0000 | 1.0000 | 1.0000 | ++0.0000 | +0 | +0 |
| mcp_tooling | +3/0/0 | 1.0000 | 1.0000 | 1.0000 | +3/0/0 | 1.0000 | 1.0000 | 1.0000 | ++0.0000 | +0 | +0 |
| open_weight_models | +1/0/0 | 1.0000 | 1.0000 | 1.0000 | +1/0/0 | 1.0000 | 1.0000 | 1.0000 | ++0.0000 | +0 | +0 |
| sandboxing | +2/0/1 | 1.0000 | 0.6667 | 0.8000 | +2/0/1 | 1.0000 | 0.6667 | 0.8000 | ++0.0000 | +0 | +0 |
| self_hosted_inference | +1/0/2 | 1.0000 | 0.3333 | 0.5000 | +1/0/2 | 1.0000 | 0.3333 | 0.5000 | ++0.0000 | +0 | +0 |
| chat_integrations | +4/3/0 | 0.5714 | 1.0000 | 0.7273 | +3/0/1 | 1.0000 | 0.7500 | 0.8571 | ++0.1299 | -3 | +1 |
| memory | +3/0/2 | 1.0000 | 0.6000 | 0.7500 | +4/0/1 | 1.0000 | 0.8000 | 0.8889 | ++0.1389 | +0 | -1 |
| exec_tools | +3/0/1 | 1.0000 | 0.7500 | 0.8571 | +4/0/0 | 1.0000 | 1.0000 | 1.0000 | ++0.1429 | +0 | -1 |
| acpx | +2/0/5 | 1.0000 | 0.2857 | 0.4444 | +3/0/4 | 1.0000 | 0.4286 | 0.6000 | ++0.1556 | +0 | -1 |
| security | +3/4/2 | 0.4286 | 0.6000 | 0.5000 | +6/3/1 | 0.6667 | 0.8571 | 0.7500 | ++0.2500 | -1 | -1 |
| model_serving | +4/3/0 | 0.5714 | 1.0000 | 0.7273 | +4/0/0 | 1.0000 | 1.0000 | 1.0000 | ++0.2727 | -3 | +0 |
| codex | +3/0/4 | 1.0000 | 0.4286 | 0.6000 | +6/0/1 | 1.0000 | 0.8571 | 0.9231 | ++0.3231 | +0 | -3 |
| auth_identity | +1/0/1 | 1.0000 | 0.5000 | 0.6667 | +3/0/0 | 1.0000 | 1.0000 | 1.0000 | ++0.3333 | +0 | -1 |
| queueing | +1/0/1 | 1.0000 | 0.5000 | 0.6667 | +2/0/0 | 1.0000 | 1.0000 | 1.0000 | ++0.3333 | +0 | -1 |
| cron_automation | +1/1/3 | 0.5000 | 0.2500 | 0.3333 | +3/1/1 | 0.7500 | 0.7500 | 0.7500 | ++0.4167 | +0 | -2 |
| hooks | +0/0/1 | 0.0000 | 0.0000 | 0.0000 | +1/1/1 | 0.5000 | 0.5000 | 0.5000 | ++0.5000 | +1 | +0 |
| packaging_deployment | +0/1/2 | 0.0000 | 0.0000 | 0.0000 | +1/1/1 | 0.5000 | 0.5000 | 0.5000 | ++0.5000 | +0 | -1 |
| skills_plugins | +0/2/2 | 0.0000 | 0.0000 | 0.0000 | +2/2/1 | 0.5000 | 0.6667 | 0.5714 | ++0.5714 | +0 | -1 |
| config | +0/0/3 | 0.0000 | 0.0000 | 0.0000 | +3/4/0 | 0.4286 | 1.0000 | 0.6000 | ++0.6000 | +4 | -3 |
| Candidate | Row | Title | Error |
|---|---|---|---|
| Seed | openclaw-openclaw-63826 | security: fix HIGH/CRITICAL vulns in skill scanner, SSRF, hook priority, and token verification | classifier exit 2: prompt: /home/bob/.local/state/localpager/classifier/prompts/20260612T201105Z-3532713.md
+schema: /home/bob/.local/state/localpager/classifier/schemas/20260612T201105Z-3532713.json
+session: /home/bob/.local/state/localpager/classifier/sessions/20260612T201105Z-3532713
+localpager-agent: final_json was not called; no structured output was captured |
| Seed | openclaw-openclaw-84752 | fix: self-heal lane wedges + restore openai-codex OAuth on embedded path | classifier exit 2: prompt: /home/bob/.local/state/localpager/classifier/prompts/20260612T201357Z-3534749.md
+schema: /home/bob/.local/state/localpager/classifier/schemas/20260612T201357Z-3534749.json
+session: /home/bob/.local/state/localpager/classifier/sessions/20260612T201357Z-3534749
+localpager-agent: final_json was not called; no structured output was captured |
| Seed | openclaw-openclaw-43246 | fix(message): deny same-provider cross-context sends by default [AI-assisted] | classifier exit 2: prompt: /home/bob/.local/state/localpager/classifier/prompts/20260612T204409Z-3555993.md
+schema: /home/bob/.local/state/localpager/classifier/schemas/20260612T204409Z-3555993.json
+session: /home/bob/.local/state/localpager/classifier/sessions/20260612T204409Z-3555993
+localpager-agent: final_json was not called; no structured output was captured |
prompt-optimizer/out/gepa-12b-proper-from-best-20260613T055906Z
Live properly budgeted 12B GEPA run continuing from the best candidate found so far. Runtime settings: row_limit=18, concurrency=2, max_tokens=1536, max_metric_calls=240. This page fetches run_log.txt every 30 seconds and redraws the graph.
| GEPA iteration | Candidate | 18-row validation score | Kind |
|---|
| GEPA iteration | Old subsample sum | New subsample sum | Delta | Decision |
|---|
run_log.txt Ā· static iteration dashboard Ā· summary dashboard
+ds4.jsonl. Task model: local 12B. Concurrency: 2.The promoted v3 candidate clears the strict gates and keeps mean predicted labels at the v9.1 baseline. The main caveat is scientific, not mechanical: the manual repair used mistakes from this same 60-row set, so a fresh holdout is still needed before treating it as deployment evidence.
+| Candidate | Mean | Delta | Precision | Recall | F1 | FP | FN | Over | Struct | Exact | Mean Labels | Label Delta | Source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| v9.1 seed baseline | 0.4322 | 0.0000 | 0.7411 | 0.5804 | 0.6510 | 29 | 60 | 2 | 3 | 10 | 1.8667 | 0.0000 | source |
| GEPA six early GEPA | 0.4912 | 0.0590 | 0.7965 | 0.5960 | 0.6818 | 23 | 61 | 1 | 0 | 12 | n/a | source | |
| Previous proper best proper GEPA | 0.5355 | 0.1033 | 0.7778 | 0.6490 | 0.7076 | 28 | 53 | 5 | 2 | 21 | 2.1000 | 0.2333 | source |
| Prop20 best proper GEPA | 0.5936 | 0.1613 | 0.8707 | 0.6689 | 0.7566 | 15 | 50 | 2 | 4 | 24 | 1.9333 | 0.0667 | source |
| Hardcase repair v2 manual repair, rejected | 0.7729 | 0.3407 | 0.9343 | 0.8477 | 0.8889 | 9 | 23 | 4 | 0 | 38 | 2.2833 | 0.4167 | source |
| Cardinality repair v3 promoted | 0.6233 | 0.1911 | 0.9375 | 0.6954 | 0.7985 | 7 | 46 | 0 | 0 | 19 | 1.8667 | 0.0000 | source |
| Gate | Target | Observed | Result |
|---|---|---|---|
| mean weighted score | >= 0.5400 | 0.6233 | pass |
| score delta vs v9.1 | >= +0.1000 | 0.1911 | pass |
| micro-F1 | >= 0.7000 | 0.7985 | pass |
| precision | >= 0.8000 | 0.9375 | pass |
| recall | >= 0.6100 | 0.6954 | pass |
| false positives | <= 20 | 7 | pass |
| false negatives | <= 58 | 46 | pass |
| over-label events | 0 or 1 | 0 | pass |
| structural failures | 0 | 0 | pass |
| exact matches | >= 15 | 19 | pass |
| mean predicted-label delta | <= +0.10 | 0.0000 | pass |
Prop20 GEPA score report Prop20 candidate tree Legacy GEPA-six dashboard Prompt diffs
+Promoted routing-policy SHA-256: b2576ca027148e109a1e72029c192ea9e26be486508671ec7c11025ac80f948b.
Rows where v3 is not an exact match. These are the next useful targets for a clean holdout-aware follow-up.
+| Target | Title | Score | Gold | Predicted | FP | FN |
|---|---|---|---|---|---|---|
| #44379 | fix(pi-runner): harden context-overflow recovery with one suppress-hook retry | 0.1667 | coding_agents, memory, hooks, reliability | agent_runtime, reliability | agent_runtime | coding_agents, memory, hooks |
| #45393 | fix(errors): friendly message and last-message repair for tool_use/tool_result mismatch (#45385) | 0.2000 | tool_calling, coding_agents, reliability | tool_calling, agent_runtime | agent_runtime | coding_agents, reliability |
| #47083 | fix: respect totalTokensFresh flag to avoid showing stale token counts | 0.2000 | sessions, telemetry_usage | ui_tui | ui_tui | sessions, telemetry_usage |
| #81957 | ci: harden GitHub Actions supply-chain boundaries | 0.2500 | security | packaging_deployment | packaging_deployment | security |
| #65364 | feat(plugins): add registerProviderRuntimeAuthOverride API | 0.2500 | auth_identity, api_surface | skills_plugins, auth_identity | skills_plugins | api_surface |
| #80008 | feat(plugins): expose ACP spawn and prompt in plugin runtime | 0.2500 | acp, coding_agents | skills_plugins, acp | skills_plugins | coding_agents |
| #90146 | google-vertex: Missing gemini-3.1-flash-lite in provider catalog causes silent failure instead of error | 0.2500 | local_model_providers, reliability | local_model_providers, model_serving | model_serving | reliability |
| #73910 | BUG: OpenClaw-managed Codex ACP uses isolated CODEX_HOME without auth bridge and sends unsupported timeout config | 0.3333 | codex, acp, acpx, auth_identity | codex, acp | none | acpx, auth_identity |
| #52249 | ACP parent session stuck until refresh when yielded waiting for child completion | 0.5000 | acp, sessions, reliability | acp, sessions | none | reliability |
| #83863 | ACP/Codex child tasks can be marked succeeded with progress-only output and no final deliverable | 0.5000 | acp, codex, agent_runtime | acp, codex | none | agent_runtime |
| #48940 | ACP: add gateway-owned node-backed runtime | 0.5000 | acp, gateway, agent_runtime | acp, gateway | none | agent_runtime |
| #48580 | Bug: acpx codex sessions åå»ŗēä¼čÆē«å³éåŗ - stdin is not a terminal | 0.5000 | acpx, codex, sessions | acpx, codex | none | sessions |
| #39248 | Bug: sandbox.mode: "non-main" silently breaks sessions_spawn subagent initialization | 0.5000 | coding_agents, sandboxing, agent_runtime | sandboxing, agent_runtime | none | coding_agents |
| #71216 | Config schema: add `sandbox`, `routing.rules`, `instances`, and `gateway.nodes.denyPaths` | 0.5000 | config, sandboxing, gateway | config, gateway | none | sandboxing |
| #84477 | Discord embedded-run prep wedge before strict-agentic, recovery skips sessionId=unknown lanes | 0.5000 | sessions, agent_runtime, reliability | sessions, reliability | none | agent_runtime |
| #51667 | Feature: Native Audio Input for Omni-Modal Models (skip STT transcription) | 0.5000 | model_serving, security, config | model_serving, config | none | security |
| #43765 | Improve runtime recovery for heartbeat, Feishu, and exec sessions | 0.5000 | reliability, exec_tools, cron_automation | reliability, exec_tools | none | cron_automation |
| #80783 | Policy: add model, network, and MCP conformance checks | 0.5000 | mcp_tooling, config, security | config, mcp_tooling | none | security |
| #68187 | SSE-backed MCP sessions can stay stale after server restart and fail with 'Session not found' | 0.5000 | mcp_tooling, sessions, gateway | mcp_tooling, sessions | none | gateway |
| #78528 | Security: skill SecretRef API keys still leak into exec child environments | 0.5000 | security, exec_tools, skills_plugins | security, exec_tools | none | skills_plugins |
| #84715 | [Bug]: @openclaw/codex peer link failure reproduced on 2026.5.19 after update | 0.5000 | codex, packaging_deployment | codex | none | packaging_deployment |
| #70529 | [Bug]: Desktop cannot use existing Chrome sessions: EasyClaw Google sign-in fails, and user profile attach fails with spawn npx ENOENT | 0.5000 | browser_automation, packaging_deployment | browser_automation | none | packaging_deployment |
| #84757 | [Bug]: Telegram session can get stuck after compaction when encrypted reasoning content fails verification | 0.5000 | sessions, chat_integrations, reliability | sessions, chat_integrations | none | reliability |
| #44202 | [Bug]: local memory embeddings on Apple Silicon can crash gateway in ggml-metal / node-llama-cpp; need official Metal/GPU guidance | 0.5000 | local_models, memory, self_hosted_inference | local_models, memory | none | self_hosted_inference |
| #10467 | [Feature Request]: Multi-lane concurrency support for sub-agents via sessions_spawn | 0.5000 | queueing, sessions, coding_agents | queueing, sessions | none | coding_agents |
| #82507 | [Feature]: ACPX Codex sandbox should inherit user-installed plugins (e.g. Superpowers) | 0.5000 | acpx, codex, skills_plugins | acpx, skills_plugins | none | codex |
| #40332 | [Feature]: Per-binding and per-agent permissionMode for ACP sessions | 0.5000 | acp, approvals, acpx | approvals, acpx | none | acp |
| #84670 | [codex] fix webchat full-message reader for truncated history | 0.5000 | gateway, api_surface, ui_tui | gateway, ui_tui | none | api_surface |
| #84583 | cron announce delivery triggers EmbeddedAttemptSessionTakeoverError when user is actively chatting | 0.5000 | cron_automation, sessions, reliability | cron_automation, sessions | none | reliability |
| #68725 | feat(amazon-bedrock-mantle): add known context windows for open-weight Mantle models | 0.5000 | open_weight_models, local_model_providers | open_weight_models | none | local_model_providers |
| #56442 | feat: Add opt-in ACP parent completion notify for sessions_spawn | 0.5000 | acp, sessions, agent_runtime | acp, sessions | none | agent_runtime |
| #60979 | feature: sessions_spawn ACP delivery to channel (stream output to Zulip/Discord topic) | 0.5000 | acp, chat_integrations, sessions | acp, sessions | none | chat_integrations |
| #52747 | fix(acp): time out stuck session lane tasks | 0.5000 | acp, sessions, reliability | acp, sessions | none | reliability |
| #84763 | fix(acpx): scrub provider credential env from ACP harness spawns | 0.5000 | acpx, acp, security | acp, acpx | none | security |
| #69256 | fix(cron): prevent premature session cleanup when subagents are running | 0.5000 | cron_automation, sessions, reliability | cron_automation, sessions | none | reliability |
| #65242 | fix: CompletionDeliveryGate to prevent duplicate ACP completion delivery | 0.5000 | acp, coding_agents, reliability | acp, reliability | none | coding_agents |
| #77827 | fix: LM Studio thinking blocks invisible with Responses API | 0.5000 | model_serving, local_models | local_models | none | model_serving |
| #42027 | fix: resolve exec PATH fallback, layered browser diagnostics, and cron force-run deadlock | 0.5000 | exec_tools, browser_automation, cron_automation | exec_tools, browser_automation | none | cron_automation |
| #84752 | fix: self-heal lane wedges + restore openai-codex OAuth on embedded path | 0.5000 | reliability, auth_identity, sessions | reliability, auth_identity | none | sessions |
| #63826 | security: fix HIGH/CRITICAL vulns in skill scanner, SSRF, hook priority, and token verification | 0.5000 | security, hooks, skills_plugins | security, skills_plugins | none | hooks |
| #62428 | test(exec): land exec v2 contract follow-through | 0.5000 | exec_tools, sandboxing, approvals | exec_tools, sandboxing | none | approvals |
Loading graph…
Loading graph…
Run directory: /home/bob/repos/localpager/prompt-optimizer/out/gepa-12b-row30-prop16-continuation-20260613T214854Z
No iteration scores found yet.
No proposal deltas found yet.
Final GEPA result is not available yet.
No proposal events found yet.
Loading graph…
Run directory: /home/bob/repos/localpager/prompt-optimizer/out/gepa-12b-row30-prop16-continuation-20260613T214937Z
| Iteration | Old subsample | New subsample | Delta | Decision |
|---|---|---|---|---|
| 1 | 2.0333 | 3.5000 | 1.4667 | accepted |
| 2 | 2.3750 | 4.0000 | 1.6250 | accepted |
| 3 | 3.5000 | 3.5000 | 0.0000 | rejected |
| 4 | 1.8190 | 3.5000 | 1.6810 | accepted |
| 5 | 2.0357 | 4.0000 | 1.9643 | accepted |
| 6 | 1.5705 | 3.5000 | 1.9295 | accepted |
| 7 | 3.5000 | 3.2500 | -0.2500 | rejected |
| 8 | 2.5000 | 4.0000 | 1.5000 | accepted |
| 9 | 2.1190 | 4.0000 | 1.8810 | accepted |
| 10 | 1.7262 | 3.2857 | 1.5595 | accepted |
| 11 | 3.1538 | 2.2857 | -0.8681 | rejected |
| 12 | 1.7500 | 3.5000 | 1.7500 | accepted |
| 14 | 3.2000 | 3.2500 | 0.0500 | accepted |
| 15 | 2.0000 | 4.0000 | 2.0000 | accepted |
| 17 | 0.7679 | 3.5000 | 2.7321 | accepted |
Run directory: /home/bob/repos/localpager/prompt-optimizer/out/gepa-12b-row30-prop16-continuation-20260614T021442Z
No iteration scores found yet.
No proposal deltas found yet.
Final GEPA result is not available yet.
No proposal events found yet.
Loading graph…
Run directory: /home/bob/repos/localpager/prompt-optimizer/out/gepa-12b-row30-prop16-from-proper-20260613T172903Z
| Iteration | Old subsample | New subsample | Delta | Decision |
|---|---|---|---|---|
| 1 | 3.2000 | 1.6500 | -1.5500 | rejected |
| 2 | 3.1429 | 2.2500 | -0.8929 | rejected |
| 3 | 1.0714 | 4.0000 | 2.9286 | accepted |
| 4 | 1.3429 | 3.2500 | 1.9071 | accepted |
| 5 | 2.6190 | 4.0000 | 1.3810 | accepted |
| 6 | 2.3910 | 4.0000 | 1.6090 | accepted |
| 7 | 2.0000 | 2.5357 | 0.5357 | accepted |
| 8 | 1.0714 | 4.0000 | 2.9286 | accepted |
| 9 | 1.4500 | 3.5000 | 2.0500 | accepted |
| 10 | 1.7857 | 4.0000 | 2.2143 | accepted |
| 11 | 3.2500 | 3.5000 | 0.2500 | accepted |
| 12 | 2.2500 | 4.0000 | 1.7500 | accepted |
| 14 | 1.7000 | 3.2500 | 1.5500 | accepted |
| 15 | 1.2357 | 4.0000 | 2.7643 | accepted |
| 16 | 2.2500 | 4.0000 | 1.7500 | accepted |
Loading graph…
Run directory: prompt-optimizer/out/gepa-12b-row30-prop20-continuation-20260614T021448Z
| Iteration | Old subsample | New subsample | Delta | Decision |
|---|---|---|---|---|
| 1 | 2.2000 | 3.1538 | 0.9538 | accepted |
| 2 | 2.3750 | 4.0000 | 1.6250 | accepted |
| 3 | 2.4396 | 4.0000 | 1.5604 | accepted |
| 4 | 2.5333 | 1.0357 | -1.4976 | rejected |
| 5 | 1.0000 | 4.0000 | 3.0000 | accepted |
| 6 | 1.4038 | 4.0000 | 2.5962 | accepted |
| 7 | 2.0357 | 4.0000 | 1.9643 | accepted |
| 8 | 2.7500 | 4.0000 | 1.2500 | accepted |
| 9 | 1.8929 | 3.2500 | 1.3571 | accepted |
| 10 | 1.3214 | 4.0000 | 2.6786 | accepted |
| 11 | 3.2857 | 3.5000 | 0.2143 | accepted |
| 12 | 1.9500 | 3.2857 | 1.3357 | accepted |
| 13 | 3.5000 | 2.0333 | -1.4667 | rejected |
| 14 | 1.5929 | 4.0000 | 2.4071 | accepted |
| 15 | 2.6429 | 4.0000 | 1.3571 | accepted |
| 16 | 1.2500 | 4.0000 | 2.7500 | accepted |
| 17 | 2.6429 | 3.5000 | 0.8571 | accepted |
| 18 | 2.0000 | 4.0000 | 2.0000 | accepted |
| 19 | 2.0000 | 4.0000 | 2.0000 | accepted |
| 20 | 2.4500 | 3.1538 | 0.7038 | accepted |
Run directory: /home/bob/repos/localpager/prompt-optimizer/out/gepa-12b-row30-prop20-continuation-20260614T081715Z
No iteration scores found yet.
No proposal deltas found yet.
Final GEPA result is not available yet.
No proposal events found yet.
Loading graph…
Loading graph…
Loading graph…
You can modify this app directly by editing index.html in the Files and versions tab.
-- Also don't forget to check the - Spaces documentation. -
-ds4.jsonl. Task model: local 12B. Concurrency: 2.The promoted v3 candidate clears the strict gates and keeps mean predicted labels at the v9.1 baseline. The main caveat is scientific, not mechanical: the manual repair used mistakes from this same 60-row set, so a fresh holdout is still needed before treating it as deployment evidence.
+| Candidate | Mean | Delta | Precision | Recall | F1 | FP | FN | Over | Struct | Exact | Mean Labels | Label Delta | Source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| v9.1 seed baseline | 0.4322 | 0.0000 | 0.7411 | 0.5804 | 0.6510 | 29 | 60 | 2 | 3 | 10 | 1.8667 | 0.0000 | source |
| GEPA six early GEPA | 0.4912 | 0.0590 | 0.7965 | 0.5960 | 0.6818 | 23 | 61 | 1 | 0 | 12 | n/a | source | |
| Previous proper best proper GEPA | 0.5355 | 0.1033 | 0.7778 | 0.6490 | 0.7076 | 28 | 53 | 5 | 2 | 21 | 2.1000 | 0.2333 | source |
| Prop20 best proper GEPA | 0.5936 | 0.1613 | 0.8707 | 0.6689 | 0.7566 | 15 | 50 | 2 | 4 | 24 | 1.9333 | 0.0667 | source |
| Hardcase repair v2 manual repair, rejected | 0.7729 | 0.3407 | 0.9343 | 0.8477 | 0.8889 | 9 | 23 | 4 | 0 | 38 | 2.2833 | 0.4167 | source |
| Cardinality repair v3 promoted | 0.6233 | 0.1911 | 0.9375 | 0.6954 | 0.7985 | 7 | 46 | 0 | 0 | 19 | 1.8667 | 0.0000 | source |
| Gate | Target | Observed | Result |
|---|---|---|---|
| mean weighted score | >= 0.5400 | 0.6233 | pass |
| score delta vs v9.1 | >= +0.1000 | 0.1911 | pass |
| micro-F1 | >= 0.7000 | 0.7985 | pass |
| precision | >= 0.8000 | 0.9375 | pass |
| recall | >= 0.6100 | 0.6954 | pass |
| false positives | <= 20 | 7 | pass |
| false negatives | <= 58 | 46 | pass |
| over-label events | 0 or 1 | 0 | pass |
| structural failures | 0 | 0 | pass |
| exact matches | >= 15 | 19 | pass |
| mean predicted-label delta | <= +0.10 | 0.0000 | pass |
Prop20 GEPA score report Prop20 candidate tree Legacy GEPA-six dashboard Prompt diffs
+Promoted routing-policy SHA-256: b2576ca027148e109a1e72029c192ea9e26be486508671ec7c11025ac80f948b.
Rows where v3 is not an exact match. These are the next useful targets for a clean holdout-aware follow-up.
+| Target | Title | Score | Gold | Predicted | FP | FN |
|---|---|---|---|---|---|---|
| #44379 | fix(pi-runner): harden context-overflow recovery with one suppress-hook retry | 0.1667 | coding_agents, memory, hooks, reliability | agent_runtime, reliability | agent_runtime | coding_agents, memory, hooks |
| #45393 | fix(errors): friendly message and last-message repair for tool_use/tool_result mismatch (#45385) | 0.2000 | tool_calling, coding_agents, reliability | tool_calling, agent_runtime | agent_runtime | coding_agents, reliability |
| #47083 | fix: respect totalTokensFresh flag to avoid showing stale token counts | 0.2000 | sessions, telemetry_usage | ui_tui | ui_tui | sessions, telemetry_usage |
| #81957 | ci: harden GitHub Actions supply-chain boundaries | 0.2500 | security | packaging_deployment | packaging_deployment | security |
| #65364 | feat(plugins): add registerProviderRuntimeAuthOverride API | 0.2500 | auth_identity, api_surface | skills_plugins, auth_identity | skills_plugins | api_surface |
| #80008 | feat(plugins): expose ACP spawn and prompt in plugin runtime | 0.2500 | acp, coding_agents | skills_plugins, acp | skills_plugins | coding_agents |
| #90146 | google-vertex: Missing gemini-3.1-flash-lite in provider catalog causes silent failure instead of error | 0.2500 | local_model_providers, reliability | local_model_providers, model_serving | model_serving | reliability |
| #73910 | BUG: OpenClaw-managed Codex ACP uses isolated CODEX_HOME without auth bridge and sends unsupported timeout config | 0.3333 | codex, acp, acpx, auth_identity | codex, acp | none | acpx, auth_identity |
| #52249 | ACP parent session stuck until refresh when yielded waiting for child completion | 0.5000 | acp, sessions, reliability | acp, sessions | none | reliability |
| #83863 | ACP/Codex child tasks can be marked succeeded with progress-only output and no final deliverable | 0.5000 | acp, codex, agent_runtime | acp, codex | none | agent_runtime |
| #48940 | ACP: add gateway-owned node-backed runtime | 0.5000 | acp, gateway, agent_runtime | acp, gateway | none | agent_runtime |
| #48580 | Bug: acpx codex sessions åå»ŗēä¼čÆē«å³éåŗ - stdin is not a terminal | 0.5000 | acpx, codex, sessions | acpx, codex | none | sessions |
| #39248 | Bug: sandbox.mode: "non-main" silently breaks sessions_spawn subagent initialization | 0.5000 | coding_agents, sandboxing, agent_runtime | sandboxing, agent_runtime | none | coding_agents |
| #71216 | Config schema: add `sandbox`, `routing.rules`, `instances`, and `gateway.nodes.denyPaths` | 0.5000 | config, sandboxing, gateway | config, gateway | none | sandboxing |
| #84477 | Discord embedded-run prep wedge before strict-agentic, recovery skips sessionId=unknown lanes | 0.5000 | sessions, agent_runtime, reliability | sessions, reliability | none | agent_runtime |
| #51667 | Feature: Native Audio Input for Omni-Modal Models (skip STT transcription) | 0.5000 | model_serving, security, config | model_serving, config | none | security |
| #43765 | Improve runtime recovery for heartbeat, Feishu, and exec sessions | 0.5000 | reliability, exec_tools, cron_automation | reliability, exec_tools | none | cron_automation |
| #80783 | Policy: add model, network, and MCP conformance checks | 0.5000 | mcp_tooling, config, security | config, mcp_tooling | none | security |
| #68187 | SSE-backed MCP sessions can stay stale after server restart and fail with 'Session not found' | 0.5000 | mcp_tooling, sessions, gateway | mcp_tooling, sessions | none | gateway |
| #78528 | Security: skill SecretRef API keys still leak into exec child environments | 0.5000 | security, exec_tools, skills_plugins | security, exec_tools | none | skills_plugins |
| #84715 | [Bug]: @openclaw/codex peer link failure reproduced on 2026.5.19 after update | 0.5000 | codex, packaging_deployment | codex | none | packaging_deployment |
| #70529 | [Bug]: Desktop cannot use existing Chrome sessions: EasyClaw Google sign-in fails, and user profile attach fails with spawn npx ENOENT | 0.5000 | browser_automation, packaging_deployment | browser_automation | none | packaging_deployment |
| #84757 | [Bug]: Telegram session can get stuck after compaction when encrypted reasoning content fails verification | 0.5000 | sessions, chat_integrations, reliability | sessions, chat_integrations | none | reliability |
| #44202 | [Bug]: local memory embeddings on Apple Silicon can crash gateway in ggml-metal / node-llama-cpp; need official Metal/GPU guidance | 0.5000 | local_models, memory, self_hosted_inference | local_models, memory | none | self_hosted_inference |
| #10467 | [Feature Request]: Multi-lane concurrency support for sub-agents via sessions_spawn | 0.5000 | queueing, sessions, coding_agents | queueing, sessions | none | coding_agents |
| #82507 | [Feature]: ACPX Codex sandbox should inherit user-installed plugins (e.g. Superpowers) | 0.5000 | acpx, codex, skills_plugins | acpx, skills_plugins | none | codex |
| #40332 | [Feature]: Per-binding and per-agent permissionMode for ACP sessions | 0.5000 | acp, approvals, acpx | approvals, acpx | none | acp |
| #84670 | [codex] fix webchat full-message reader for truncated history | 0.5000 | gateway, api_surface, ui_tui | gateway, ui_tui | none | api_surface |
| #84583 | cron announce delivery triggers EmbeddedAttemptSessionTakeoverError when user is actively chatting | 0.5000 | cron_automation, sessions, reliability | cron_automation, sessions | none | reliability |
| #68725 | feat(amazon-bedrock-mantle): add known context windows for open-weight Mantle models | 0.5000 | open_weight_models, local_model_providers | open_weight_models | none | local_model_providers |
| #56442 | feat: Add opt-in ACP parent completion notify for sessions_spawn | 0.5000 | acp, sessions, agent_runtime | acp, sessions | none | agent_runtime |
| #60979 | feature: sessions_spawn ACP delivery to channel (stream output to Zulip/Discord topic) | 0.5000 | acp, chat_integrations, sessions | acp, sessions | none | chat_integrations |
| #52747 | fix(acp): time out stuck session lane tasks | 0.5000 | acp, sessions, reliability | acp, sessions | none | reliability |
| #84763 | fix(acpx): scrub provider credential env from ACP harness spawns | 0.5000 | acpx, acp, security | acp, acpx | none | security |
| #69256 | fix(cron): prevent premature session cleanup when subagents are running | 0.5000 | cron_automation, sessions, reliability | cron_automation, sessions | none | reliability |
| #65242 | fix: CompletionDeliveryGate to prevent duplicate ACP completion delivery | 0.5000 | acp, coding_agents, reliability | acp, reliability | none | coding_agents |
| #77827 | fix: LM Studio thinking blocks invisible with Responses API | 0.5000 | model_serving, local_models | local_models | none | model_serving |
| #42027 | fix: resolve exec PATH fallback, layered browser diagnostics, and cron force-run deadlock | 0.5000 | exec_tools, browser_automation, cron_automation | exec_tools, browser_automation | none | cron_automation |
| #84752 | fix: self-heal lane wedges + restore openai-codex OAuth on embedded path | 0.5000 | reliability, auth_identity, sessions | reliability, auth_identity | none | sessions |
| #63826 | security: fix HIGH/CRITICAL vulns in skill scanner, SSRF, hook priority, and token verification | 0.5000 | security, hooks, skills_plugins | security, skills_plugins | none | hooks |
| #62428 | test(exec): land exec v2 contract follow-through | 0.5000 | exec_tools, sandboxing, approvals | exec_tools, sandboxing | none | approvals |
Loading...+
The routing-policy files are usually the best thing to diff. The full prompt files include the scaffold around the editable policy.
+| Version | Note | Raw prompt | Raw routing policy | Routing policy SHA-256 |
|---|---|---|---|---|
v9.1 seed00-v9-1-seed | Baseline OpenClaw classification prompt, normalized to the optimizer placeholder format. | prompt 241 lines | routing policy 185 lines | 3d84a551239a9c26865a44758f0602548c1361482eef087c194d878a2d92ad37 |
GEPA six01-gepa-six | Early 12B GEPA winner from the first six-row run. | prompt 100 lines | routing policy 45 lines | f4b161bb9bbaf366f1d4f1841243d73544bbd3c553ca6be5eb2818e757007187 |
Previous proper best02-previous-proper-best | Earlier proper GEPA run before the prop20 continuation. | prompt 206 lines | routing policy 151 lines | 288c716272a98543b9c5d7c191ac9389173adddefd17a7fba409a2e9d6b9ff65 |
Prop20 best03-prop20-best | Best candidate from the 20-proposal GEPA continuation. | prompt 203 lines | routing policy 148 lines | 5ff17ce59da5ff6c98c7241f10a12e7ebb908b415c51a6aad631ad9d83a686e5 |
Hardcase repair v204-hardcase-repair-v2 | Manual hardcase repair that scored high but failed cardinality. | prompt 227 lines | routing policy 171 lines | 39be82b8f2e94199ed5c832e6ddd8716e886bbe95874f6afad7ff5fd89798abe |
Cardinality repair v305-cardinality-repair-v3 | Promoted cardinality-safe repair with a strict two-label cap. | prompt 233 lines | routing policy 177 lines | b2576ca027148e109a1e72029c192ea9e26be486508671ec7c11025ac80f948b |