LAB / VERIFY

Failure Trace.

Trace a distributed-system failure from request to recovery.

GUIDED MODEL / NOT LIVE

Choose a context, follow the investigation questions in order, and open the specialist instrument that can answer each one.

INVESTIGATION CONTEXT

Latency creep

How does the selected bounded model or named record change under delay?

Changes the investigation context only. Destination tools keep their own state and evidence.

Context boundary
  • A real latency regression occurred.
  • All instruments share one latency cause.
  • Current production performance is represented.
Choose an investigation context

INVESTIGATION PATH

Follow the path. Jump when you need to.

The rail shows the proper investigation order, what Failure Trace can map for this context, and where a question must stay open.

01 / REQUEST

What happened to this request?

START HERE

CONTEXTUAL / REQUEST

Request X-Ray

Simulated

How bounded request timing changes affect the synthetic trace.

Proof boundary

Request X-Ray offers timing controls inside its synthetic request model.

Evidence detail

No live request timing is measured by the shared model.

Open Request X-Ray

02 / DEPENDENCIES

How did the problem propagate through dependencies?

DIRECT / DEPENDENCIES

CASCADE

Simulated

How added delay changes propagation and containment inside the synthetic CASCADE graph.

Proof boundary

CASCADE directly models bounded added latency in its synthetic graph.

Open CASCADE

03 / COORDINATION

How did replicated state, routing, isolation, and protection behave?

CONTEXTUAL / COORDINATION

Consensus

Simulated

How a delayed replica affects the fixed teaching model's convergence.

Proof boundary

Consensus provides a bounded delay teaching model, not a production protocol reading.

Open Consensus

04 / IMPACT

Which public components or contracts could be affected?

OPEN / NO GENERIC MAPPING

The investigation question still matters.

CURRENT MAPPING

Failure Trace has no scenario-safe impact instrument mapping for Latency creep. The stage stays visible instead of inventing an answer.

HOW IT WOULD WORK

Use bounded Atlas Twin relationship context to identify components that could be affected, then require named evidence before any stronger impact claim.

05 / INCIDENT EVIDENCE

What incident evidence was actually recorded or proven?

CONTEXTUAL / INCIDENT EVIDENCE

Blackbox

Recorded replay

A named recorded incident can show the latency frames that its recorder captured.

Proof boundary

Only a named recorded Blackbox record can support this reading.

Open Blackbox

06 / RECOVERY

What can be truthfully said about recovery?

COMPOSITE ASSESSMENT

COMPOSITE ASSESSMENT / NOT AN OBSERVATION

Recovery has no single source of truth.

Different instruments can show reset, convergence, recorded aftermath, or a named lifecycle fact. Failure Trace keeps those signals separate until they support the same bounded conclusion.

BOUNDARY

Recovery-like output, aftermath, or a named lifecycle fact does not prove universal current service recovery or root-cause resolution.

OPTIONAL INTERPRETATION

Off the main rail.

These layers can interpret a supported context without becoming investigation stages or incident authority.

Spectral ForgeSynthetic telemetry interpretationSystem SymphonySource-aware telemetry sonification

REFERENCE

Technical detail, when you need it.

Model-approved reading paths

From dependency failure to recovery

  1. Request / Normal operation / Request X-Ray
  2. Dependencies / Dependency failure / CASCADE
  3. Dependencies / Cascading failure / CASCADE
  4. Incident evidence / Cascading failure / Blackbox
  5. Recovery / Recovery / CASCADE

Coordination and Impact remain omitted rather than inferred.

From bounded isolation to recovery

  1. Request / Normal operation / Request X-Ray
  2. Coordination / Network partition / Consensus
  3. Coordination / Network partition / Neon Relay
  4. Recovery / Recovery / Consensus

Dependencies, Impact, and Incident evidence remain omitted where the model has no bounded relationship.

Evidence modes
Measured
Current source-owned measurement with source and observation time owned by the destination.
Stale measured
A previous measurement whose freshness threshold has elapsed.
Recorded replay
A named historical record. It is not current live state.
Simulated
Bounded synthetic behaviour created by an instrument.
Unavailable
Expected evidence cannot currently be retrieved or served.
Unknown
The available record does not establish the meaning or state.
Not applicable / unscored
The relationship is intentionally unsupported. No score is inferred.
Participating instruments
Unsupported relationships for selected context

Context-specific unsupported relationships load from the shared model. Open stages remain visible above without JavaScript.

Proof and interpretation rules
  • A selected context is a guide only. It does not mean a real incident occurred.
  • Simulated behaviour is not production failure.
  • A topology relationship does not prove propagation.
  • Atlas Twin says only what could be affected, never what was affected.
  • Recorded replay is historical, not current live state.
  • Sonification is interpretation, not incident authority.
  • Source or merge state does not prove deployment, runtime, or live state.
  • Recovery-like output does not create one universal recovery authority.