Illustrative local example. No workflow is running.
Read the evidence behind a workflow.
Was a request received? What execution was recorded? What output was reviewed? Keep the evidence for each question separate when discussing a workflow.
Three layers, three evidence requirements.
The arrows show a relationship to investigate. A record in one layer does not establish that the next layer happened.
01
Receipt
A receipt tied to the request and input version.
↓→02
Execution record
An identified attempt with its recorded events and context.
↓→03
Output review
An output reference and a recorded review of the agreed question.
Inspect a finite evidence example.
Select a prepared sample and read its records. The controls change this page only; they run no workflow and retrieve no live history.
Illustrative local example. No workflow is running.
All IDs, values and T+ offsets are illustrative. T+ describes an authored record sequence, not elapsed time. There is no live timer.
- Request ID
- EX-101
- Attempt ID
- Not recorded
- Input version
- catalog-v1
Prepared record sequence
- T+00RC-101Receipt recorded
- Receipt
- □ Record present
- Execution record
- — Execution not evidenced
- Output review
- ○ Output not evidenced; Review open
Receipt
Source: authored local example record
- T+00 · RC-101 · Receipt recorded
Output review remains open. No sample establishes reviewed business correctness or live NOGL behavior.
What does the selected example establish?
- Supported by this example
- The sample receipt was recorded for EX-101 and catalog-v1. No execution attempt is recorded.
- ○ Still unproven
- Output review remains open. No sample establishes reviewed business correctness or live NOGL behavior.
Prepare a review packet.
Bring identifiers and evidence context for the question being reviewed. No credentials, private payload, personal data or upload are needed for this guide.
- Request and attempt IDs
- Tie each record to the same request and a specific attempt. Keep an absent attempt ID as Not recorded.
- Record source and time
- Identify where the record came from and when it was recorded. This example uses authored T+ offsets.
- Input version
- Keep the version used by the recorded attempt beside its evidence.
- Execution or failure detail
- Include available events and the stated failure context without inferring a retry or resolution.
- Output reference and review scope
- Identify the output, the question to review and the responsible reviewer. An output reference alone is not a review.
Why these fields matter
A different input version may change what a record supports. An absent record must stay absent. Agree what history is available and how long it is retained; this guide promises no retention policy or cryptographic verification.
Choose the next question.
Highlight an evidence layer and a question to carry into the review packet. This selection sends no request and changes no runtime state.
Confirm execution evidence
Ask for an attempt ID, source and recorded execution events linked to the request. A receipt is not evidence that execution started.
○ Review open
Review the model output separately.
Keep the review question and its supporting evidence beside a displayed model output. These questions do not run a model or evaluate its quality.
Forecast output
Identify the observation cutoff, forecast horizon, held-out outcomes and error or range assessment.
A historical sales line or an authored future path is not a validated forecast.
Review planning assumptions →Budget scenario
Keep the chosen objective, period, constraints and model assumptions with the comparison.
A conditional allocation is not an executed campaign or a realized result.
Explore budget scenarios →Effect estimate
Identify the comparison or study design, assumptions and limits supporting a causal interpretation.
Prediction accuracy alone does not establish an intervention's effect.
Match evidence to the question →
Before explaining an unusual number
Check the metric definition, observation windows and missing records before accepting a cause.
Fictional evidence review. No detector, alert or business action is running.
Explore example →Choose a prepared case
Read the records first
All IDs, values and UTC windows are fictional. Windows use [start, end): the end is excluded.
Missing is not zero.
Reference observations
- Metric
- order-count
- Orders
- Metric version
- order-count-v1
- Unit
- count
- Scope
- example-shop/web/completed-orders
Current observation
- Metric
- order-count
- Orders
- Metric version
- order-count-v1
- Unit
- count
- Scope
- example-shop/web/completed-orders
Question focus: Check completeness
Identify every expected record and explain any absence. Ask for the collection scope and delay policy.
| Record | Definition | Window | Source | Value / Record status |
|---|---|---|---|---|
| C1-R1ReferenceAuthored record |
| Relevant to selected question
| Relevant to selected question
| Relevant to selected question
Orders |
| C1-R2ReferenceAuthored record |
| Relevant to selected question
| Relevant to selected question
| Relevant to selected question
Orders |
| C1-R3ReferenceAuthored record |
| Relevant to selected question
| Relevant to selected question
| Relevant to selected question
Orders |
| C1-R4ReferenceAuthored record |
| Relevant to selected question
| Relevant to selected question
| Relevant to selected question
Orders |
| C1-R5CurrentAuthored record |
| Relevant to selected question
| Relevant to selected question
| Relevant to selected question
Orders |
Show provenance
These references identify fields in this authored fixture. They are not external sources. Missing value and collection-time references remain absent.
C1-R1 · Authored record
- Metric
- C1-R1:metricId
- Metric version
- C1-R1:metricVersion
- Unit
- C1-R1:unit
- Scope
- C1-R1:scopeId
- Window start
- C1-R1:windowStart
- Window end
- C1-R1:windowEnd
- Source
- C1-R1:log
- Version
- C1-R1:log
- Recorded at
- C1-R1:log
- Value
- C1-R1:value
- Record status
- C1-R1:log
C1-R2 · Authored record
- Metric
- C1-R2:metricId
- Metric version
- C1-R2:metricVersion
- Unit
- C1-R2:unit
- Scope
- C1-R2:scopeId
- Window start
- C1-R2:windowStart
- Window end
- C1-R2:windowEnd
- Source
- C1-R2:log
- Version
- C1-R2:log
- Recorded at
- C1-R2:log
- Value
- C1-R2:value
- Record status
- C1-R2:log
C1-R3 · Authored record
- Metric
- C1-R3:metricId
- Metric version
- C1-R3:metricVersion
- Unit
- C1-R3:unit
- Scope
- C1-R3:scopeId
- Window start
- C1-R3:windowStart
- Window end
- C1-R3:windowEnd
- Source
- C1-R3:log
- Version
- C1-R3:log
- Recorded at
- C1-R3:log
- Value
- C1-R3:value
- Record status
- C1-R3:log
C1-R4 · Authored record
- Metric
- C1-R4:metricId
- Metric version
- C1-R4:metricVersion
- Unit
- C1-R4:unit
- Scope
- C1-R4:scopeId
- Window start
- C1-R4:windowStart
- Window end
- C1-R4:windowEnd
- Source
- C1-R4:log
- Version
- C1-R4:log
- Recorded at
- C1-R4:log
- Value
- C1-R4:value
- Record status
- C1-R4:log
C1-R5 · Authored record
- Metric
- C1-R5:metricId
- Metric version
- C1-R5:metricVersion
- Unit
- C1-R5:unit
- Scope
- C1-R5:scopeId
- Window start
- C1-R5:windowStart
- Window end
- C1-R5:windowEnd
- Source
- C1-R5:log
- Version
- C1-R5:log
- Recorded at
- C1-R5:log
- Value
- C1-R5:value
- Record status
- C1-R5:log
What the example supports
Arithmetic difference only.
The current count is 60. The four recorded reference counts average 100, a difference of −40 orders. These selected windows do not establish a normal range, a statistically unusual event or a cause.
- Reference mean
- 100 Orders
- Current minus reference mean
- −40 Orders
Selected windows only; representativeness unknown.
Choose the next investigation question
Check completeness
Identify every expected record and explain any absence. Ask for the collection scope and delay policy.
Carry the evidence into human review
Include the source and version, all reference and current observation windows, missing records and the exact question.
- Prepared case
- C1 · Recorded difference
- Investigation question
- Check completeness
- Missing records
- None in this prepared case
C1-R1 · Reference · Authored record
fictional-metric-register · fixture-1 · count
[2026-08-03T10:00:00Z, 2026-08-03T11:00:00Z) UTC
Record status: Present
C1-R2 · Reference · Authored record
fictional-metric-register · fixture-1 · count
[2026-08-10T10:00:00Z, 2026-08-10T11:00:00Z) UTC
Record status: Present
C1-R3 · Reference · Authored record
fictional-metric-register · fixture-1 · count
[2026-08-17T10:00:00Z, 2026-08-17T11:00:00Z) UTC
Record status: Present
C1-R4 · Reference · Authored record
fictional-metric-register · fixture-1 · count
[2026-08-24T10:00:00Z, 2026-08-24T11:00:00Z) UTC
Record status: Present
C1-R5 · Current · Authored record
fictional-metric-register · fixture-1 · count
[2026-08-31T10:00:00Z, 2026-08-31T11:00:00Z) UTC
Record status: Present
Reviewer unassigned. No approval, notification or resolution is recorded.
Keep the next decision separate
Use these checks across the seven-stage cycle. Evidence review does not authorize a pricing, inventory or customer action.
Does this detect anomalies?
No. These fixed records demonstrate evidence checks. No statistical model or live monitoring runs here.
Does acknowledgment prove a cause or fix?
No. Reading or acknowledging a record does not establish a cause, approval or successful correction.
Discuss the evidence requirements before planning a workflow.
Use the existing guides to prepare connections, access and the evidence a reviewer needs. The example packet is not attached or submitted automatically.
- Read the connection guide
Agree input contracts and the boundary between connection and recorded behavior.
- Plan integration inputs
Prepare access, ownership and the intended connection scope.
- Read the trust guidance
Discuss data handling and the requirements for a workspace.