From superpowers-plus
Synthesizes evidence from debug investigator branches into a ranked root cause verdict with reasoning tree. Detects contradictions and weighs evidence strength by reproducibility and specificity, not agent count.
How this skill is triggered — by the user, by Claude, or both
Slash command
/superpowers-plus:evidence-adjudicatorThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
> **Role:** Synthesize all investigator evidence into a root cause verdict. Weigh evidence strength, not investigator count.
Role: Synthesize all investigator evidence into a root cause verdict. Weigh evidence strength, not investigator count. Dispatched by:
debug-conductor— invoked after investigators complete. Evidence type:RootCauseVerdict(seeskills/_shared/evidence-schema.md)
Dispatched by debug-conductor after all investigator branches complete. Synthesizes branch evidence into a ranked root cause verdict with reasoning tree.
Receive from conductor:
For each hypothesis (branch):
Σ(supporting × confidence) - Σ(disconfirming × confidence)Root: "What caused the incident?"
├── H1: "Event ordering bug" (net: +2.4)
│ ├── [+] Deployment correlation (0.85)
│ ├── [+] Out-of-order events in logs (0.90)
│ ├── [+] Reproduction succeeded 3/3 (0.95)
│ └── [-] Network latency increase (0.30) — addressed: within normal variance
├── H2: "SIP timeout misconfiguration" (net: +0.3)
│ ├── [+] Timeout matches symptom duration (0.70)
│ └── [-] Timeout is effect, not cause (0.85) — strong disconfirmation
└── H3: "Network degradation" (net: -0.5)
├── [+] 5ms latency increase (0.20) — weak signal
└── [-] Latency within normal range (0.70) — strong disconfirmation
Where do investigators disagree?
Many real incidents have multiple contributing factors:
For the winning hypothesis:
Before accepting the verdict, apply adversarial-search thinking to the leading hypothesis:
{
"rootCause": "Call router v2.3.1 async event processing delivers events out of order under load",
"confidence": 0.88,
"supportingEvidence": [
{ "source": "deployment-history", "finding": "v2.3.1 deployed 2h before incident", "timestamp": "2026-03-29T10:15:00Z", "confidence": 0.85, "type": "supporting" },
{ "source": "call-router-logs", "finding": "Events arriving out of order", "timestamp": "2026-03-29T10:20:00Z", "confidence": 0.90, "type": "supporting" },
{ "source": "reproduction", "finding": "Reproduced 3/3 with async + load", "timestamp": "2026-03-29T11:00:00Z", "confidence": 0.95, "type": "supporting" }
],
"disconfirmingEvidence": [
{ "source": "network-metrics", "finding": "5ms latency increase — within normal variance, does not explain 2600ms increase", "timestamp": "2026-03-29T10:25:00Z", "confidence": 0.30, "type": "disconfirming" }
],
"alternativeCauses": [
{ "cause": "SIP timeout misconfiguration", "confidence": 0.15, "reason": "Timeout is effect, not cause" },
{ "cause": "Network degradation", "confidence": 0.05, "reason": "Metrics within normal range" }
],
"divergencePoints": ["Was the delay caused by network or application-layer processing?"],
"gaps": ["Have not tested under exact production load (500 concurrent vs 50 in staging)"]
}
| Score | Meaning | Requirements |
|---|---|---|
| 0.9–1.0 | Near-certain | Reproduction confirmed + all disconfirming addressed + no gaps |
| 0.7–0.9 | Strong | Reproduction or strong correlation + most disconfirming addressed |
| 0.5–0.7 | Moderate | Correlation evidence + some gaps remain |
| 0.3–0.5 | Weak | Plausible hypothesis but significant gaps |
| <0.3 | Insufficient | Speculation; investigation needs more evidence |
| Pattern | Evidence Shape |
|---|---|
| Single clear cause | One hypothesis >0.8, all others <0.3 |
| Compound cause | Two hypotheses each 0.4–0.6, together explain >0.8 |
| Wrong consensus | Multiple investigators agree but reproduction fails → all wrong |
| Minority correct | One low-confidence branch has stronger evidence than high-confidence majority |
| Insufficient evidence | All hypotheses <0.5 → need more data, not more investigation |
| Mode | Symptom | Recovery |
|---|---|---|
| Confirmation bias | Favoring first evidence seen | Weight by strength, not order |
| Missing contradiction | Overlooking conflicting evidence | Explicit contradiction detection pass |
| Over-counting agents | Same finding from N agents treated as N evidence | Deduplicate by root fact |
npx claudepluginhub bordenet/superpowers-plus --plugin superpowers-plusGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.