CLUBEVIA / IMPACTTRACE
LOCAL-FIRST LABV4.1
INCIDENT & RELIABILITY INTELLIGENCE / RESEARCH STAGE

Turn AI uncertainty into traceable evidence.

Investigate recorded event dependencies, surface explainable warning signals, and compare observed task outcomes across models and access plans. No hidden inference. No automated remediation.

CSV / JSON / JSONLEvent lineageTier comparisonAudit exports
01 / INGESTION

Data workspace

Begin with the built-in synthetic sample or import structured, non-sensitive test logs.

01—05
↑

Import event data

Native file picker for iPhone, iPad and desktop

Choose a local event fileCSV, JSON, JSONL · up to 5 MB / 10,000 records
Ready for sample data.
Example CSV ↓Example JSON ↓
File picker not responding? Paste data

Optional structured adapters: common OpenTelemetry span exports and LangSmith run metadata. Sensitive prompts and outputs are intentionally not imported from those formats.

✓

Data quality & provenance

What we know, what remains unknown

EVENTS—
VALID LINKS—
EXPLICIT TASK RESULTS—
UNKNOWN TIERS—
No records loaded.
A missing dependency or unknown subscription tier is an evidence gap — never an automatic error verdict.
02 / DETECTION

Explainable signal scanner

Rule-triggered review leads, not verified AI errors. Findings include the triggering evidence.

Detection limitation: text/status heuristics suggest where to look. Silence is not proof of safe behavior; high severity is not a probability of harm.
—Primary signals
—Linked signals
—Human-reviewed
03 / INVESTIGATION

Downstream impact tracing

Trace only recorded parent–child dependencies and identify possible exposure.

Choose an event to investigate.

—Downstream events
—Customer IDs
—Connected systems
—Maximum path depth
EVIDENCE TRAIL

Linked event chain

NO INCIDENT

A dependency path alone is not causal proof.

Choose an event to see its recorded path.
REVIEW PRIORITIES

Systems & next checks

Do not alter records, contact affected customers or claim financial losses without human verification against source systems.
DOWNSTREAM REVIEW QUEUE

Potentially affected events

EVENTPATHSYSTEMCUSTOMERACTIONREVIEW

No downstream data selected.

04 / OBSERVATIONAL METRICS

AI reliability intelligence

Compare explicitly recorded task outcomes across providers, models, subscriptions and workflows.

No free-versus-paid verdict: different models, prompts, tools, task difficulty and reporting practices can all affect observed results. These are descriptive rates, not controlled experiments or product rankings. The synthetic sample deliberately uses different demo models in each tier.
—Valid task results
—Unknown outcomes
—Duplicate task records excluded
GROUPTASKSFAILURESOBSERVED RATE95% INTERVAL*EVIDENCE

Load a dataset with explicit task_result and outcome fields to calculate rates.

*Wilson binomial interval: sampling uncertainty only, not a correction for selection bias. Fewer than 30 results per group are marked low sample. Outcomes are not inferred from wording.

05 / EXPORTS & GOVERNANCE

Human-owned evidence packet

Consolidate source quality, signal explanations, incident lineage, and cautious reliability metrics.

Case details

Optional fields are included in your local export. They are never transmitted.

Masking IDs is not full anonymization. Reports may still contain sensitive action descriptions: review before sharing.

Exportable outputs

Files are generated locally in your browser. No backend, user accounts, or event collection.

Data retention: this page does not persist imported data. Reloading clears analysis state and any case notes. Export your report before leaving.