Release delta and runtime evidence agree; integration test coverage is incomplete.
Evidence-backed root cause and engineering investigation
An AI investigation system that finds the evidence-supported cause.
From an observed symptom, CauseSignal’s governed specialist playbooks reconstruct what happened, retrieve the evidence that matters, and challenge competing explanations — then report the cause the record supports, or state plainly when the evidence is not enough.
Your team retains authority over remediation and production change. The CauseSignal runtime is read-only — it recommends the next change or routes work to the accountable team, and never acts on production itself.
Payment authorization failures after release 7.14
First observed 13:42 UTC · production / eu-west · change window +18 min
Timing is plausible; observed failure signature conflicts with two prior events.
No supporting latency or availability evidence in admitted sources.
Establish what changed and when.
Build a common, evidence-backed account of the event before teams commit to an explanation.
Test the possible causes.
Make competing explanations, contradictions, confidence, and material unknowns visible for review.
Take the right next action.
Recommend a governed change or hand work to the platform, provider, or operating team that owns it.
Investigation intelligence
Move from scattered signals to a defensible engineering explanation.
CauseSignal combines timeline reconstruction, change intelligence, safe evidence retrieval, hypothesis challenge, and prior-case knowledge while keeping provenance and uncertainty visible.
01 / RECONSTRUCT
Build a shared account of the event.
Reconstruct the canonical timeline
Orders releases, code changes, tests, incidents, operational events, and evidence windows into a traceable chronology instead of a chat transcript.
Connect nearby changes
Finds releases, commits, configuration shifts, database changes, dependency updates, and related product changes that could explain the observed behavior.
Retrieve evidence safely
Searches code, tests, history, prior incidents, indexed knowledge, and admitted runtime context while preserving scope, identity, provenance, and stale-evidence boundaries.
02 / REASON
Challenge the explanation before accepting it.
Challenge competing hypotheses
Develops multiple causal candidates, looks for both supporting and contradictory evidence, and keeps unresolved alternatives visible until the record justifies narrowing.
Plan the investigation
Identifies the next highest-value evidence to inspect, what is still missing, and which questions can actually change the confidence of the investigation.
Use prior RCA patterns
Surfaces similar incidents and recurring causal patterns without treating a historical match as proof that the same cause applies now.
03 / DECIDE
Make the evidence useful for the next decision.
Explain the finding
Separates observation, inference, finding, contradiction, recommendation, and unknown so reviewers can see exactly what the evidence does and does not support.
Recommend the right remediation path
Distinguishes repository changes such as code, database, infrastructure-as-code, and configuration from actions that require an external owner or operational authority.
Investigation method
A governed path from symptom to defensible cause.
CauseSignal’s investigation roles do not begin with a preferred answer. They frame the event, admit evidence deliberately, reconstruct the sequence, challenge causal candidates, and state only what the available record supports.
- 01 / FRAME
Define the event
Capture expected versus observed behavior, scope, time, environment, impact, and the question the investigation must answer.
- 02 / RECONSTRUCT
Build the evidence timeline
Connect source history, releases, tests, configuration, database and runtime evidence with exact identity and provenance.
- 03 / CHALLENGE
Test causal candidates
Rank hypotheses, seek disconfirming evidence, preserve contradictions, and stop safely when evidence is insufficient.
- 04 / DECIDE
Report finding and next action
Produce a cited finding, confidence boundary, alternatives, unknowns, and the governed remediation or external-action handoff.
Governed specialist playbooks
CauseSignal is built from governed specialist playbooks, each responsible for one part of an evidence-led investigation.
There is no autonomous agent swarm. CauseSignal maintains a governed registry of named specialist roles, keeps evidence gathering separate from causal judgement, and preserves human authority over remediation and production change.
Define the event
Axiom, Lens, Scout
Gather the evidence
Trace, Intent, Recall, Mesh
Test and critique the explanation
Causa, Verity, Vigil
Report the cited finding
Scribe
Prepare a scope-bounded remediation proposal and candidate patch, and validate it in a policy-governed sandbox — never against production.
Reviews path safety, secret redaction, egress, command policy, and prompt-injection resistance across every role, at every stage.
Deploys CauseSignal itself into authorized environments. It takes no part in any investigation.
Stable role identities make responsibility, evidence, handoffs, and separation of duties visible.

Axiom
Triage Orchestrator
Plans and bounds the investigation, validates every stage handoff, and escalates or stops when the evidence is insufficient. Read-only; it orchestrates, never concludes.

Lens
Issue Parser
Normalizes the reported issue into expected-versus-observed framing and open questions, treating reported content as untrusted input rather than fact.

Scout
Search Planner
Plans bounded retrieval: query strategy, evidence-gap priority, and retrieval cost control. Plans only - it issues no findings of its own.

Trace
Code Navigator
Read-only code navigation, symbol and reference search, test-to-code localization, and config inspection - approved roots only, pinned to an exact commit.

Intent
Requirement Analyst
Resolves the effective, approved requirement and detects drift or contradiction between what was intended and what was expected.

Recall
Historical Evidence Agent
Retrieves prior incidents and release-change context with provenance, and flags stale history. Reviewed memory only; similarity is context, not proof.

Mesh
Graph Context Agent
Bounded relationship, topology, and schema-lineage expansion with provenance preserved. Bounded hops and fan-out; never crosses scope.

Causa
Root Cause Analyst
Builds competing causal hypotheses, reviews counter-evidence, and abstains where evidence is insufficient. Creates no evidence and holds no action authority.

Verity
Evidence Critic
Independently reviews evidence admission, claim traceability, contradictions, and unsupported claims. Can only downgrade or reject - never upgrade.

Scribe
Report Writer
Renders the cited finding, confidence rationale, contradictions, and explicit non-actions - restricted to claims the evidence critic has approved.

Remedy
Remediation Proposal Agent
Generates remediation options within scope, plans a bounded fix, and binds each proposal to its evidence digest.

Mender
Candidate Patch Agent
Prepares a candidate patch under minimal-diff discipline, with protected paths enforced throughout.

Warden
Sandbox Policy Evaluator
Evaluates sandbox admission policy, resource limits, and secret and egress exposure before anything is validated.

Proving
Sandbox Validation Agent
Coordinates sandbox tests and reproducible build validation, binding every result to its digest.

Vigil
Remediation Safety Reviewer
Independently reviews remediation safety, residual risk, and separation of duties before a change is recommended.

Sentinel
Safety Reviewer
Path safety, secret redaction, egress control, command policy, and prompt-injection resistance across every role. Denies or escalates; cannot override policy.

Harbor
CauseSignal Deployment Agent
Dual-provider deployment into explicitly authorized environments with immutable artifacts. Operates outside the investigation flow and never changes investigation logic.
Roster source of record: docs/agents/AGENT_ROSTER.yaml (PI-29R_49A / DEC-226). Role metadata describes contract capabilities and responsibility boundaries; it does not itself grant runtime authority, tool access, or installed packages.
Evidence domains
Cross-domain context without flattening provenance.
Every source keeps its origin, version, time, scope, and access conditions. CauseSignal can connect evidence across domains without silently turning retrieval results into facts.
Safety boundary
Analysis can recommend change without silently making it.
CauseSignal's product runtime is intentionally read-only. It can investigate, explain, and recommend. Where a remediation is representable in governed scope, it can identify the required code, database, infrastructure-as-code, or configuration change. Where the real fix belongs to an external operator, provider, network team, or platform owner, CauseSignal says so instead of fabricating a patch.
Investigation & decision outputs
Evidence that engineering teams can review, challenge, and act on.
Outputs preserve the boundary between what was observed, what was inferred, what is recommended, and what still requires evidence or another authority.
Cited causal analysis
Ranked explanation with supporting evidence, contradictions, alternatives, confidence, and material unknowns.
Canonical timeline
A durable sequence of changes, releases, tests, and operational events relevant to the investigation.
Evidence & change map
Source lineage, affected components, and claim relationships that show why a conclusion is grounded.
Investigation plan
The next evidence to collect, questions to resolve, and checks most likely to change the conclusion.
Remediation guidance
Governed recommendation for code, database, IaC, configuration, or an explicit external-remediation handoff.
Review packet
A durable investigation record for engineering, incident command, support, assurance, and follow-up work.
Emporia IT product portfolio
Connect investigation to delivery and product understanding.
CauseSignal is one part of an evidence-led product portfolio for complex software teams. Explore the products that help teams act on an investigation or establish a reliable product record.
IncrementPilot
Turn approved product work into reviewed, tested, delivery-ready change—with people retaining release authority.
Explore IncrementPilot →PRODUCT INTELLIGENCEFoundryTrace
Build a current, evidence-backed product record across requirements, releases, deployed reality, gaps, and uncertainty.
Explore FoundryTrace →Technical briefing
Evaluate CauseSignal against a real incident or defect investigation.
Discuss evidence admission, incident reconstruction, change intelligence, RCA workflows, remediation handoff, qualification, and enterprise controls.
support@emporiait.com