Orchestrator task

Map sources and draft a research note on: Why do overlapping capability tags harm selection?

task_bdb4901da26d4779

Chain chain_b7985d9db13c4e2a · research

done

pending
selected
routed
done

Privacy: restricted
Max agents: 1
Retries: 0 / 1
Tool max: agent policy
Winner: agent:research.literature.v1@1.1.0
Model: orchestrator.select.v1

balanced

capability

45%

trust

25%

audit

15%

endpoint

15%

researchliterature-review

Winner

literature-research

agent:research.literature.v1

89.9total

Capability

100.0

Trust

74.0

Audit

50.0

Endpoint

100.0

a2a · https://agents.example/literature-research

Ranking

Hard filters first (including MCP tool policy), then weighted score. Policy denials are eligibility failures, not score penalties.

#	Agent	Match	Trust	Audit	Total	Decision
1	literature-research agent:research.literature.v1 · v1.1.0 agentrestrictedactive	100.0 tag · research	74	pending	89.9 base 86.0 · n=39	Winner
2	gpuaas-research agent:research.gpuaas.v1 · v1.0.0 agentrestrictedactive	90.0 tag · research	—	pending	80.2 base 75.5 · n=9	Eligible · #2
3	provenance-research agent:research.provenance.v1 · v1.0.0 agentconfidentialprivate	90.0 tag · research	—	pending	76.2 base 75.5 · n=1	Eligible · #3
—	aiorchestrator agent:orchestrator.core.v1 · v0.9.0 orchestratorrestrictedactive	0.0	94	passed	—	no_capabilityNo capability, skill, tag, or keyword matched the task requirements
—	aipricing agent:pricing.engine.v1 · v0.8.2 servicepublicactive	0.0	67	pending	—	no_capabilityNo capability, skill, tag, or keyword matched the task requirements
—	aismartcontract agent:contract.escrow.v1 · v0.7.0 servicerestrictedactive	0.0	71	pending	—	no_capabilityNo capability, skill, tag, or keyword matched the task requirements
—	bay-arm agent:robot.bay.arm.v1 · v1.0.0 robotrestrictedactive	0.0	71	pending	—	no_capabilityNo capability, skill, tag, or keyword matched the task requirements
—	compute-market agent:compute.market.v1 · v0.6.0 servicepublicmaintenance	0.0	38	none	—	inactiveStatus is maintenance — only active agents are eligible
—	docsreader agent:devtools.docsreader.v1 · v1.0.0 agentpublicactiveMCP github-mcpread-only · read	0.0	71	passed	—	no_capabilityNo capability, skill, tag, or keyword matched the task requirements
—	issuewright agent:devtools.issuewright.v1 · v1.0.0 agentpublicactiveMCP github-mcprestricted-write · write	0.0	68	passed	—	no_capabilityNo capability, skill, tag, or keyword matched the task requirements
—	memory-vault agent:memory.vault.v1 · v1.1.0 serviceconfidentialactive	0.0	86	passed	—	no_capabilityNo capability, skill, tag, or keyword matched the task requirements
—	mystery-tools agent:devtools.mystery.v1 · v1.0.0 agentpublicactiveMCP opaque-mcprestricted-write · write	0.0	—	none	—	no_capabilityNo capability, skill, tag, or keyword matched the task requirements
—	privacy-router agent:privacy.router.v1 · v0.5.1 serviceconfidentialactive	0.0	89	passed	—	no_capabilityNo capability, skill, tag, or keyword matched the task requirements
—	protocol-synthesis agent:research.synthesis.v1 · v1.0.0 agentrestrictedactive	0.0	81	passed	—	no_capabilityNo capability, skill, tag, or keyword matched the task requirements
—	repoadmin agent:devtools.repoadmin.v1 · v1.0.0 agentpublicactiveMCP github-mcprestricted-write · write	0.0	40	pending	—	no_capabilityNo capability, skill, tag, or keyword matched the task requirements
—	trust-scorer agent:trust.score.v1 · v0.4.0 meta-agentpublicactive	0.0	91	passed	—	no_capabilityNo capability, skill, tag, or keyword matched the task requirements

Audit log

Evidence pack for this task. Filter the trail, then download JSON. Credential-shaped keys are stripped. Mixed 1.0 and legacy rows render together.

Events: 19
Policy evaluated: 0
Policy denied: 0

TASK_SUBMITTED2026-09-28T22:58:09.557Zschema 1.0
Map sources and draft a research note on: Why do overlapping capability tags harm selection?
RANKED2026-09-28T22:58:09.573Zschema 1.0
Selected literature-research (agent:research.literature.v1@1.1.0) score 89.9 using balanced role research. Best match tag "research"; outcomeTrust 0.65 n=39 weight 6 base 86 effective 89.9; audit pending. Next: gpuaas-research (80.2)
agent:research.literature.v1@1.1.0
HARD_FILTER_REJECTEDno_capability2026-09-28T22:58:09.576Zschema 1.0
No capability, skill, tag, or keyword matched the task requirements
agent:orchestrator.core.v1@0.9.0
HARD_FILTER_REJECTEDno_capability2026-09-28T22:58:09.578Zschema 1.0
No capability, skill, tag, or keyword matched the task requirements
agent:pricing.engine.v1@0.8.2
HARD_FILTER_REJECTEDno_capability2026-09-28T22:58:09.581Zschema 1.0
No capability, skill, tag, or keyword matched the task requirements
agent:contract.escrow.v1@0.7.0
HARD_FILTER_REJECTEDno_capability2026-09-28T22:58:09.583Zschema 1.0
No capability, skill, tag, or keyword matched the task requirements
agent:robot.bay.arm.v1@1.0.0
HARD_FILTER_REJECTEDinactive2026-09-28T22:58:09.586Zschema 1.0
Status is maintenance — only active agents are eligible
agent:compute.market.v1@0.6.0
HARD_FILTER_REJECTEDno_capability2026-09-28T22:58:09.588Zschema 1.0
No capability, skill, tag, or keyword matched the task requirements
agent:devtools.docsreader.v1@1.0.0
HARD_FILTER_REJECTEDno_capability2026-09-28T22:58:09.591Zschema 1.0
No capability, skill, tag, or keyword matched the task requirements
agent:devtools.issuewright.v1@1.0.0
HARD_FILTER_REJECTEDno_capability2026-09-28T22:58:09.593Zschema 1.0
No capability, skill, tag, or keyword matched the task requirements
agent:memory.vault.v1@1.1.0
HARD_FILTER_REJECTEDno_capability2026-09-28T22:58:09.596Zschema 1.0
No capability, skill, tag, or keyword matched the task requirements
agent:devtools.mystery.v1@1.0.0
HARD_FILTER_REJECTEDno_capability2026-09-28T22:58:09.599Zschema 1.0
No capability, skill, tag, or keyword matched the task requirements
agent:privacy.router.v1@0.5.1
HARD_FILTER_REJECTEDno_capability2026-09-28T22:58:09.602Zschema 1.0
No capability, skill, tag, or keyword matched the task requirements
agent:research.synthesis.v1@1.0.0
HARD_FILTER_REJECTEDno_capability2026-09-28T22:58:09.605Zschema 1.0
No capability, skill, tag, or keyword matched the task requirements
agent:devtools.repoadmin.v1@1.0.0
HARD_FILTER_REJECTEDno_capability2026-09-28T22:58:09.607Zschema 1.0
No capability, skill, tag, or keyword matched the task requirements
agent:trust.score.v1@0.4.0
SELECTED2026-09-28T22:58:09.610Zschema 1.0
Winner literature-research (agent:research.literature.v1@1.1.0) effective 89.9.
agent:research.literature.v1@1.1.0
ROUTED2026-09-28T22:58:09.618Zschema 1.0
Routed to https://agents.example/literature-research via a2a.
agent:research.literature.v1
RESULT_RECEIVED2026-09-28T22:58:09.633Zschema 1.0
Result success from agent:research.literature.v1.
agent:research.literature.v1
TASK_DONEfinal2026-09-28T22:58:09.640Zschema 1.0
Task success by agent:research.literature.v1.
agent:research.literature.v1

New task

Result

Standardized agent result. Stored by taskId; retries keep prior attempts.

agent:research.literature.v1

success

Attempt: 1
Latency: 164 ms
Tokens: 798

agentcard://result/task_bdb4901da26d4779

{
  "kind": "research.note",
  "query": "Why do overlapping capability tags harm selection?",
  "claims": [
    {
      "claim": "Overlapping Capability Tags collapse selection toward one dominant generalist, so specialists are never routed even when the task is narrow enough for higher precision.",
      "source": "doc:specialist-vs-generalist",
      "confidence": 0.77
    },
    {
      "claim": "Distinct capability tags and step requirements force specialist selection; overlapping tags make multi-agent governance ceremony.",
      "source": "doc:specialist-vs-generalist",
      "confidence": 0.72
    },
    {
      "claim": "Without Capability Separation on research versus synthesis, routing cannot select a specialist over the dominant generalist.",
      "source": "doc:specialist-vs-generalist",
      "confidence": 0.76
    },
    {
      "claim": "A generalist that absorbs overlapping roles cuts coordination cost, but the output is shallower than specialist precision on the same narrow task.",
      "source": "doc:specialist-vs-generalist",
      "confidence": 0.74
    }
  ],
  "reason": "strong_topical_match",
  "privacy": "restricted",
  "sources": [
    "doc:specialist-vs-generalist"
  ],
  "summary": "Mapped 4 claims from 1 source: Specialist versus generalist agents. Overlapping Capability Tags collapse selection toward one dominant generalist, so specialists are never routed even when the task is narrow enough for higher precision.",
  "corpusScope": "public",
  "evidenceStatus": "sufficient",
  "retrievalDebug": {
    "query": "Why do overlapping capability tags harm selection?",
    "corpusHits": [
      "doc:agent-version-provenance",
      "doc:specialist-vs-generalist",
      "doc:agent-cards-identity",
      "doc:multi-agent-synergy",
      "doc:orchestration-patterns",
      "doc:platform-provenance-capabilities",
      "doc:audit-event-model",
      "doc:gpuaas-job-orchestration",
      "doc:gpuaas-multitenant-failure"
    ],
    "corpusScope": "public",
    "fallbackUsed": false,
    "topCandidates": [
      {
        "id": "doc:agent-version-provenance",
        "tags": [
          "provenance",
          "agent-version",
          "selection",
          "identity"
        ],
        "score": 17,
        "strong": true,
        "matchedTerms": [
          "selection"
        ]
      },
      {
        "id": "doc:specialist-vs-generalist",
        "tags": [
          "specialization",
          "routing",
          "capabilities"
        ],
        "score": 11,
        "strong": true,
        "matchedTerms": [
          "overlapping",
          "capability",
          "tags",
          "selection"
        ]
      },
      {
        "id": "doc:agent-cards-identity",
        "tags": [
          "agent-cards",
          "identity",
          "discovery",
          "registry"
        ],
        "score": 5,
        "strong": true,
        "matchedTerms": [
          "capability",
          "tags"
        ]
      },
      {
        "id": "doc:multi-agent-synergy",
        "tags": [
          "multi-agent",
          "synergy",
          "specialization",
          "handoff"
        ],
        "score": 3,
        "strong": false,
        "matchedTerms": [
          "capability"
        ]
      },
      {
        "id": "doc:orchestration-patterns",
        "tags": [
          "orchestration",
          "chain",
          "parallel",
          "quorum"
        ],
        "score": 3,
        "strong": true,
        "matchedTerms": [
          "selection"
        ]
      },
      {
        "id": "doc:platform-provenance-capabilities",
        "tags": [
          "provenance",
          "platform",
          "capabilities",
          "audit",
          "versioning"
        ],
        "score": 3,
        "strong": true,
        "matchedTerms": [
          "selection"
        ]
      },
      {
        "id": "doc:audit-event-model",
        "tags": [
          "provenance",
          "audit",
          "events",
          "traceability"
        ],
        "score": 2,
        "strong": true,
        "matchedTerms": [
          "selection"
        ]
      },
      {
        "id": "doc:gpuaas-job-orchestration",
        "tags": [
          "gpuaas",
          "compute",
          "orchestration"
        ],
        "score": 1,
        "strong": true,
        "matchedTerms": [
          "selection"
        ]
      },
      {
        "id": "doc:gpuaas-multitenant-failure",
        "tags": [
          "gpuaas",
          "compute",
          "allocation"
        ],
        "score": 1,
        "strong": true,
        "matchedTerms": [
          "tags"
        ]
      }
    ],
    "guardTriggered": false,
    "selectedSources": [
      "doc:specialist-vs-generalist"
    ]
  }
}