AEP v1.0.2 — Agentic Performance Report

Generated: 2026-05-17T23:09:24.351679Z

Authority: operator 'wheres the measured performance of our agents with the new aep' 2026-05-17.

Composes with: AGENTIC-CAPABILITIES-V1.0.2-2026-05-17.md (Wave-046 substrate perf); this fills the per-AGENT gap.

TL;DR

• 10 canonical agents tracked (10 active last 7d; 10 active last 30d)
• 941 total ledger rows across all agents
• last 7d: 286 invocations
• last 30d: 941 invocations
• 1 HIGH-VETO mentions + 1 other VETO mentions across all agents
• 768 ledger rows with artifact_path set (concrete artifacts produced)

Per-agent performance (ranked by total invocations)

Cross-agent aggregate metrics

Truth-tag distribution (aggregate)

• STRONGLY PLAUSIBLE: 561 (59.6%)
• PROVEN/RELIABLE: 366 (38.9%)
• EXPERIMENTAL: 5 (0.5%)
• SPECULATIVE FRONTIER: 3 (0.3%)
• UNKNOWN: 3 (0.3%)
• STRONGLY_PLAUSIBLE: 1 (0.1%)
• PROVEN/RELIABLE on quota observation; STRONGLY PLAUSIBLE on LEGION protocol design: 1 (0.1%)
• PROVEN/RELIABLE on rule + atomic-rewrite mechanism; STRONGLY PLAUSIBLE on stderr-signature detection (EXPERIMENTAL until first cascade trip): 1 (0.1%)

Outcome distribution (aggregate)

• success: 785 (83.4%)
• recovered: 147 (15.6%)
• failed: 6 (0.6%)
• UNKNOWN: 3 (0.3%)

Activity ranking (last 7 days)

1. forge — 51 invocations
1. scribe — 37 invocations
1. pathfinder — 34 invocations
1. strategist — 31 invocations
1. judge — 28 invocations
1. curator — 27 invocations
1. scout — 22 invocations
1. adversary — 22 invocations
1. warden — 21 invocations
1. visual-judge — 13 invocations

Per-agent detail

forge

• Total invocations: 252
• First: 2026-05-06T00:00:00
• Last: 2026-05-17T23:09:24
• Activity: 7d=51 / 30d=252
• Unique sessions: 117
• Artifact rows: 245
• VETO mentions: 1 (HIGH-VETO: 0)
• Wave participation count: 4
• Top 5 waves: wave-007×1, wave-019×1, wave-016×1, wave-014×1
• Top 5 cluster tags: lodestone×84, forge×73, crucible×40, implementation×31, forge-generator-role×25
• Notes quality proxy (length stats): mean=528 / median=340 / p95=1394
• Truth-tag distribution: {'PROVEN/RELIABLE': 140, 'STRONGLY PLAUSIBLE': 110, 'EXPERIMENTAL': 1, 'SPECULATIVE FRONTIER': 1}
• Outcome distribution: {'success': 243, 'recovered': 9}

scribe

• Total invocations: 116
• First: 2026-05-05T00:00:00
• Last: 2026-05-17T00:00:00
• Activity: 7d=37 / 30d=116
• Unique sessions: 63
• Artifact rows: 111
• VETO mentions: 0 (HIGH-VETO: 0)
• Wave participation count: 1
• Top 5 waves: wave-026×1
• Top 5 cluster tags: lesson×59, scribe×29, compounding-signal-mandatory×22, compounding×11, scribe-upgrade-floor-applied×11
• Notes quality proxy (length stats): mean=366 / median=257 / p95=1116
• Truth-tag distribution: {'PROVEN/RELIABLE': 60, 'STRONGLY PLAUSIBLE': 53, 'UNKNOWN': 1, 'PROVEN/RELIABLE on quota observation; STRONGLY PLAUSIBLE on LEGION protocol design': 1, 'PROVEN/RELIABLE on rule + atomic-rewrite mechanism; STRONGLY PLAUSIBLE on stderr-signature detection (EXPERIMENTAL until first cascade trip)': 1}
• Outcome distribution: {'success': 106, 'recovered': 10}

judge

• Total invocations: 99
• First: 2026-05-06T00:00:00
• Last: 2026-05-17T23:09:24
• Activity: 7d=28 / 30d=99
• Unique sessions: 47
• Artifact rows: 56
• VETO mentions: 0 (HIGH-VETO: 0)
• Wave participation count: 0
• Top 5 cluster tags: verify×49, scope-lock×23, judge×17, two-part-judge-structure×13, reflector-role-ace×9
• Notes quality proxy (length stats): mean=262 / median=198 / p95=625
• Truth-tag distribution: {'PROVEN/RELIABLE': 60, 'STRONGLY PLAUSIBLE': 37, 'UNKNOWN': 2}
• Outcome distribution: {'success': 85, 'recovered': 11, 'failed': 2, 'UNKNOWN': 1}

warden

• Total invocations: 87
• First: 2026-05-05T00:00:00
• Last: 2026-05-17T00:00:00
• Activity: 7d=21 / 30d=87
• Unique sessions: 61
• Artifact rows: 51
• VETO mentions: 0 (HIGH-VETO: 0)
• Wave participation count: 3
• Top 5 waves: wave-006×2, wave-005×1, wave-012×1
• Top 5 cluster tags: audit×51, warden×20, pattern:WARN-N-persistence×15, pattern:defect-tag-scan×13, marathon×11
• Notes quality proxy (length stats): mean=231 / median=160 / p95=481
• Truth-tag distribution: {'PROVEN/RELIABLE': 55, 'STRONGLY PLAUSIBLE': 32}
• Outcome distribution: {'success': 57, 'recovered': 29, 'failed': 1}

pathfinder

• Total invocations: 82
• First: 2026-05-06T00:00:00
• Last: 2026-05-17T23:09:24
• Activity: 7d=34 / 30d=82
• Unique sessions: 55
• Artifact rows: 55
• VETO mentions: 1 (HIGH-VETO: 1)
• Wave participation count: 6
• Top 5 waves: wave-019×2, wave-020×2, wave-002×1, wave-007×1, wave-008×1
• Top 5 cluster tags: decomposition×39, pathfinder×22, adversary-amendment-fold-matrix×19, scope-lock×14, disconfirmer-first×14
• Notes quality proxy (length stats): mean=246 / median=171 / p95=568
• Truth-tag distribution: {'STRONGLY PLAUSIBLE': 77, 'PROVEN/RELIABLE': 3, 'EXPERIMENTAL': 1, 'STRONGLY_PLAUSIBLE': 1}
• Outcome distribution: {'success': 74, 'recovered': 6, 'UNKNOWN': 2}

strategist

• Total invocations: 77
• First: 2026-05-06T00:00:00
• Last: 2026-05-17T23:09:24
• Activity: 7d=31 / 30d=77
• Unique sessions: 57
• Artifact rows: 50
• VETO mentions: 0 (HIGH-VETO: 0)
• Wave participation count: 3
• Top 5 waves: wave-001×2, wave-003×1, wave-005×1
• Top 5 cluster tags: framing×31, strategist×14, 5-weakest-assumption-hooks×10, scope-lock×7, tier-3×6
• Notes quality proxy (length stats): mean=234 / median=182 / p95=637
• Truth-tag distribution: {'STRONGLY PLAUSIBLE': 72, 'PROVEN/RELIABLE': 3, 'EXPERIMENTAL': 2}
• Outcome distribution: {'success': 70, 'recovered': 7}

curator

• Total invocations: 76
• First: 2026-05-06T00:00:00
• Last: 2026-05-17T00:00:00
• Activity: 7d=27 / 30d=76
• Unique sessions: 48
• Artifact rows: 69
• VETO mentions: 0 (HIGH-VETO: 0)
• Wave participation count: 3
• Top 5 waves: wave-001×3, wave-005×2, wave-006×1
• Top 5 cluster tags: curator×41, honest-counting×12, anti-source-laundering×10, sibling-78-discipline×10, marathon×9
• Notes quality proxy (length stats): mean=217 / median=192 / p95=411
• Truth-tag distribution: {'STRONGLY PLAUSIBLE': 51, 'PROVEN/RELIABLE': 25}
• Outcome distribution: {'success': 75, 'recovered': 1}

adversary

• Total invocations: 66
• First: 2026-05-06T00:00:00
• Last: 2026-05-16T00:00:00
• Activity: 7d=22 / 30d=66
• Unique sessions: 49
• Artifact rows: 50
• VETO mentions: 0 (HIGH-VETO: 0)
• Wave participation count: 0
• Top 5 cluster tags: pre-mortem×32, 5-weakest-assumption-hooks×17, premortem×14, adversary×13, ace-reflector-role×10
• Notes quality proxy (length stats): mean=280 / median=189 / p95=740
• Truth-tag distribution: {'STRONGLY PLAUSIBLE': 61, 'PROVEN/RELIABLE': 5}
• Outcome distribution: {'recovered': 57, 'success': 9}

scout

• Total invocations: 61
• First: 2026-05-06T00:00:00
• Last: 2026-05-16T00:00:00
• Activity: 7d=22 / 30d=61
• Unique sessions: 36
• Artifact rows: 56
• VETO mentions: 0 (HIGH-VETO: 0)
• Wave participation count: 0
• Top 5 cluster tags: research×23, anti-source-laundering×14, sibling-78-upgrade×11, powershell×7, scout×7
• Notes quality proxy (length stats): mean=348 / median=194 / p95=959
• Truth-tag distribution: {'STRONGLY PLAUSIBLE': 49, 'PROVEN/RELIABLE': 9, 'SPECULATIVE FRONTIER': 2, 'EXPERIMENTAL': 1}
• Outcome distribution: {'success': 59, 'recovered': 2}

visual-judge

• Total invocations: 25
• First: 2026-05-07T00:00:00
• Last: 2026-05-16T00:00:00
• Activity: 7d=13 / 30d=25
• Unique sessions: 24
• Artifact rows: 25
• VETO mentions: 0 (HIGH-VETO: 0)
• Wave participation count: 0
• Top 5 cluster tags: visual-judge×20, operator-pending×16, rubric×13, lodestone×13, forge-follow-up-amendments×10
• Notes quality proxy (length stats): mean=275 / median=200 / p95=601
• Truth-tag distribution: {'STRONGLY PLAUSIBLE': 19, 'PROVEN/RELIABLE': 6}
• Outcome distribution: {'recovered': 15, 'success': 7, 'failed': 3}

Honest framing per §69.5

• These metrics are ledger-derived (proxy for agent activity). They do NOT measure:
• Per-invocation latency (would need invocation-time-stamping; not currently captured)
• Token cost per agent (would need token-usage capture per Anthropic billing API)
• Multi-agent convergence rate (would need cross-agent dispatch correlation IDs)
• Operator-judged quality per agent output (subjective; requires operator rating loop)
• The metrics ARE empirical proxies for: activity volume, recency, VETO honor pattern, artifact production, doctrine ladder usage (truth-tag distribution).

STAGED for next session (per honest disclosure)

• Per-invocation latency capture — add invoked_at + completed_at fields to ledger schema
• Token cost capture — instrument Agent dispatches to record token usage; tie to ledger via session_id
• Multi-agent convergence metric — when N agents fire on same wave, measure % agreement on TOP-1 frame
• Operator-rated quality loop — post-wave operator scores each agent's contribution 0-10, recorded in dedicated ledger