Known
Supported by evidence within the stated scope and limits. The report states the support basis without turning the classification into a probability scale.
Interactive field guide · Version 1.2 · August 2026
Evidence before
conclusion.
Explore the reasoning discipline that turns source material, contested claims, incomplete records, and tool-derived outputs into bounded findings that can be tested, explained, and challenged.
Evidence-to-finding pathway
Evidence does not adjudicate itself.
A finding is a human-adjudicated statement expressing what the available evidence supports, assumes, or cannot resolve within the engagement’s authority, scope, and limits.
01Source materialRecords, artifacts, outputs
02ObservationWhat was perceived or extracted
03ClaimProposed meaning
04Tested claimCompleteness, failure modes, alternatives
05Human-adjudicated findingKnown, assumed, or undetermined
06Decision supportThe decision remains with the decision-maker
Boundary: The Zemi Method produces defensible findings. It does not make organizational, legal, operational, or governance decisions.
Investigative lifecycle
Seven phases. One traceable record.
Select a phase to see its governing question, principal actions, and automation boundary.
Human adjudication gate
Three classifications of kind.
The classifications state epistemic status. They are not confidence scores or probability bands.
Assumed
Relied on for a stated and limited purpose without being directly established. The assumption remains visible and separate from what is known.
Undetermined
Cannot be resolved on the available evidence.
PendingExhausted
Artifact-to-Finding Promotion: A log, file artifact, AI output, vendor statement, detector score, prior report, or tool result may support a finding. It is not itself a finding.
AI role classification
Describe what the evidence shows.
The classification does not assign legal responsibility and does not accept labels, logs, prompts, model explanations, or vendor statements as proof without corroboration.
AI present
AI exists in the environment, artifact set, or workflow, but the evidence does not show material effect on the activity.
AI-assisted
AI supports a human through drafting, troubleshooting, summarization, coding, search, or analysis without shown independent objective selection or material execution.
AI-orchestrated
AI coordinates or triggers material workflow steps under human-configured objectives, rules, prompts, agents, tools, or integrations.
AI-autonomous claimed
A claim of independent objective selection, decision, or material execution without human control. Independent evidence is required.
Action provenanceWhat objective, tool, permission, execution, response, and state change can be reconstructed?
Control provenanceWho or what created, configured, authorized, deployed, operated, supervised, or intervened?
Representation completenessDid human-visible, machine-readable, and transformed views materially differ?
Attribution boundaryExecution, control, causation, authorization, intent, autonomy, and accountability are separate propositions.
Standing release test
The 39-item defensibility check.
Mark each applicable control as it is satisfied. Applicability must be stated, not assumed away. Progress is stored only in this browser.
A finding that fails any applicable item returns to the phase where the gap opened. Nothing reaches a client on the strength of the method’s reputation alone. It reaches them on the strength of the record behind it.