LAB / VERIFY / FAILURE INVESTIGATION

Failure Laboratory.

A guided investigation corridor for distributed-system failure questions. Choose a bounded scenario, follow the model-approved sequence, and enter the existing instrument that can answer that stage's question.

READ THIS FIRST

One corridor. Separate proof boundaries.

A scenario is context for choosing a bounded question. It is not a report that Atlas experienced that failure, and it does not make every instrument part of one executed event. Each room keeps its own evidence mode, source, freshness, and proof boundary.

SEQUENCE
Six investigation stages
STATE
Model relationship, not live incident
HANDOFF
Existing canonical route

01 / SCENARIO CONTEXT

Which bounded scenario are you tracing?

Changing this selection changes the explanatory relationships below. The selected label never changes a destination's mechanics or turns simulated behaviour into production evidence.

Failure Laboratory scenarios

02 / INVESTIGATION SEQUENCE

Walk the corridor in order.

Every stage has a question, an explicit relationship, and a handoff to an existing canonical instrument. A missing relationship stays visible as missing.

03 / THE CORRIDOR

Selected model path: Normal operation

Relationships are read from the shared Failure Laboratory model.

  1. REQUEST

    What happened to this request?

    START HERE

    DIRECT / REQUEST

    Request X-Ray

    Simulated

    What happened to this request?

    NATIVE SCENARIO
    Healthy baseline
    IT DEMONSTRATES
    A deterministic baseline request moving through the seven-layer browser model.
    Proof boundary

    This is a bounded synthetic request baseline.

    What it does not prove
    • A real request is healthy.
    • The current estate has no request failures.
    Open Request X-Ray
    Next: Dependencies
  2. DEPENDENCIES

    How did the problem propagate through dependencies?

    PROPAGATION

    DIRECT / DEPENDENCIES

    CASCADE

    Simulated

    How did a bounded fault propagate through dependencies and how did containment alter the result?

    NATIVE SCENARIO
    Baseline
    IT DEMONSTRATES
    The declared CASCADE graph before a synthetic fault is introduced.
    Proof boundary

    This is a baseline state in a deterministic synthetic graph.

    What it does not prove
    • Production dependencies are healthy.
    • A real incident is absent.
    Open CASCADE
    Next: Coordination
  3. COORDINATION

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

    AGREEMENT

    DIRECT / COORDINATION

    Consensus

    Simulated

    How did the bounded replicated-state and quorum teaching model behave?

    NATIVE SCENARIO
    CLEAN
    IT DEMONSTRATES
    A clean starting point for the fixed-leader three-replica teaching model.
    Proof boundary

    This is a clean state in a fixed teaching model.

    What it does not prove
    • A production cluster is healthy or converged.
    Open Consensus

    DIRECT / COORDINATION

    Neon Relay

    Simulated

    How did bounded routing, isolation, protection, and repair behave?

    NATIVE SCENARIO
    Prime board
    IT DEMONSTRATES
    A powered starting state for the relay's synthetic circuit puzzles.
    Proof boundary

    This is a synthetic circuit baseline.

    What it does not prove
    • Production routes or infrastructure are healthy.
    Open Neon Relay
    Next: Impact
  4. IMPACT

    Which public components or contracts could be affected?

    CONTEXT
    Not applicable / unscored

    No normal-operation impact target is mapped.

    Atlas Twin context is available only through the public Evidence Console and remains separate from this selected baseline. A relationship that could be affected is never rewritten as a component that was affected.

    The current model does not infer impact from a normal-operation label.

    Next: Incident evidence
  5. INCIDENT EVIDENCE

    What incident evidence was actually recorded or proven?

    RECORD

    CONTEXTUAL / INCIDENT EVIDENCE

    Blackbox

    Recorded replay

    What incident evidence was actually recorded?

    NATIVE SCENARIO
    Record-specific before-failure frames
    IT DEMONSTRATES
    A named recorder record may provide historical baseline frames before its recorded incident.
    Proof boundary

    Only a named recorded Blackbox record can support this contextual reading.

    What it does not prove
    • The current estate is normal.
    • The replay is current live evidence.
    Open Blackbox
    Next: Recovery
  6. RECOVERY

    What can be truthfully said about recovery?

    GAP KEPT OPEN
    NO SINGLE AUTHORITY

    There is no single current Phase 3 Recovery Evidence instrument.

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

    Allowed readings stay distinct: simulated, recorded replay, measured, stale measured, unavailable, and unknown.

    The corridor ends at the evidence boundary. No universal recovery claim is added.

04 / OPTIONAL INTERPRETATION

Interpretation layers stay off the main rail.

Spectral Forge and System SYMPHONY can add interpretation where the shared model permits it. They are not mandatory linear stages, and their underlying evidence mode stays visible.

CROSS-CUTTING / OPTIONAL

Spectral Forge

Simulated

How synthetic baseline telemetry can be mapped to sound and visual state.

A deterministic mapping from selected synthetic telemetry to sound and visual state. Its audio is not incident evidence.

Open Spectral Forge

CROSS-CUTTING / OPTIONAL

System SYMPHONY

SimulatedMeasuredStale measuredRecorded replayUnknown

What does bounded changing telemetry sound like while uncertainty remains visible?

A read-only sonification of selected source state and uncertainty. Musical state is not incident authority.

Open System SYMPHONY

05 / MODEL BOUNDARY

Unsupported relationships stay visible.

The selected scenario does not get a vaguely similar instrument. “Not applicable / unscored” means this relationship is intentionally not mapped; it is not a hidden failure and it is not a zero score.

Atlas Twin context

Not applicable / unscored

Twin describes bounded relationships that could be affected; it does not represent a normal-operation state.

Evidence Console

Not applicable / unscored

The Console reads named claims and context; it is not a normal-operation simulator.

Model-approved reading paths

From dependency failure to recovery

A beginner-readable bounded path from a synthetic request baseline through dependency failure and propagation to an explicitly limited recovery reading.

  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

Omitted stages: Coordination and Impact. The path does not invent evidence merely to fill the sequence.

From bounded isolation to recovery

A distributed-systems path using the closest bounded coordination and routing models, with its missing stages kept explicit.

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

Omitted stages: Dependencies, Impact, and Incident evidence. No current instrument supplies those stages for this bounded partition-oriented narrative.

These are model-approved explanatory paths, not synchronized execution traces. Each handoff still opens a separate instrument with its own state and evidence contract.

Canonical rooms and proof boundaries

These links keep every participating destination reachable without JavaScript. They are a handoff index, not a second product directory and not a shared runtime.

06 / EVIDENCE LANGUAGE

Read the label before the result.

The mode says what kind of evidence a destination can present. It does not make this route a live feed. For measured or recorded readings, source and freshness remain owned by the destination record.

Measured
A current source-owned measurement, with source and observation time shown by the destination.
Stale measured
A previous measurement whose freshness threshold has elapsed.
Recorded replay
A named historical record played back; it is not current live state.
Simulated
Bounded synthetic behaviour created by an instrument; it is not production failure.
Unavailable
Expected evidence cannot currently be retrieved or served.
Unknown
The meaning or state is not established by the available record.
Not applicable / unscored
This scenario/instrument relationship is intentionally unsupported; no score is inferred.