CZ / FORWARD
EN中文

05 / PROOF & CONTROL

A successful demo is not business acceptance.

A capability earns exposure to more realistic work only when representative cases, real users, explicit permissions, and failure paths produce acceptable evidence. Autonomy follows evidence and human authority—not presentation quality.

Explore progressive validation ↓

PROGRESSIVE EXPOSURE

Increase exposure only when evidence and human authority support it.

The stages describe increasing exposure to real work, not an automatic path toward maximum autonomy. A capability may remain at any bounded stage, move back to a safer stage, be revised, stop, or defer.

Exposure staircase · not a maturity model, value ranking, mandatory sequence, release plan, or promise of eventual autonomy.

Exposure staircase · not a maturity model, value ranking, mandatory sequence, release plan, or promise of eventual autonomy.
E1Analysis & RecommendationExposure 1 of 6
EXPOSURE BOUNDARY
Produce bounded analysis or recommendations for human judgment without changing business work.
HUMAN CONTROL
Human decides whether and how to use every material conclusion.
REQUIRED EVIDENCE
Which normal, exception, and failure cases support judgment.
FALLBACK / STOP
What happens when evidence, systems, users, or outputs fail; and what evidence or event prevents continuation or expansion.
E2Offline EvaluationExposure 2 of 6
EXPOSURE BOUNDARY
Test against accepted historical or synthetic-safe cases outside the live process.
HUMAN CONTROL
Cases, rubric, thresholds, and acceptance authority are human-approved.
REQUIRED EVIDENCE
Which normal, exception, and failure cases support judgment.
FALLBACK / STOP
What happens when evidence, systems, users, or outputs fail; and what evidence or event prevents continuation or expansion.
E3Shadow ModeExposure 3 of 6
EXPOSURE BOUNDARY
Observe or produce parallel outputs without changing the live business process.
HUMAN CONTROL
Output is non-operative; comparison, logging, privacy, and stop rules remain explicit.
REQUIRED EVIDENCE
Which normal, exception, and failure cases support judgment.
FALLBACK / STOP
What happens when evidence, systems, users, or outputs fail; and what evidence or event prevents continuation or expansion.
E4Copilot ModeExposure 4 of 6
EXPOSURE BOUNDARY
Support named users while they review every material recommendation or action.
HUMAN CONTROL
A named user retains approval, correction, override, and abandonment authority.
REQUIRED EVIDENCE
Which normal, exception, and failure cases support judgment.
FALLBACK / STOP
What happens when evidence, systems, users, or outputs fail; and what evidence or event prevents continuation or expansion.
E5Bounded AutomationExposure 5 of 6
EXPOSURE BOUNDARY
Execute only explicitly authorized actions within narrow scope, permissions, thresholds, and recovery.
HUMAN CONTROL
Human owners define exceptions, escalation, rollback, monitoring, and stop authority.
REQUIRED EVIDENCE
Which normal, exception, and failure cases support judgment.
FALLBACK / STOP
What happens when evidence, systems, users, or outputs fail; and what evidence or event prevents continuation or expansion.
E6Limited Expansion After EvidenceExposure 6 of 6
EXPOSURE BOUNDARY
Consider additional users, cases, or actions only through a fresh evidence and authority decision.
HUMAN CONTROL
Expansion is a new decision; previous evidence does not automatically transfer.
REQUIRED EVIDENCE
Which normal, exception, and failure cases support judgment.
FALLBACK / STOP
What happens when evidence, systems, users, or outputs fail; and what evidence or event prevents continuation or expansion.

CROSS-STAGE CONTROL RAIL

  1. 01
    HUMAN AUTHORITY

    who may approve, override, pause, stop, and accept.

  2. 02
    REPRESENTATIVE CASES

    which normal, exception, and failure cases support judgment.

  3. 03
    FAILURE & FALLBACK

    what happens when evidence, systems, users, or outputs fail.

  4. 04
    LOGS & VERSION

    what is traceable, reproducible, and attributable to a version.

  5. 05
    STOP CONDITIONS

    what evidence or event prevents continuation or expansion.

More exposure requires new evidence and a new human decision. It is never inherited automatically.

VISIBLE SEMANTIC EQUIVALENT

  1. E1
    Analysis & Recommendation

    Produce bounded analysis or recommendations for human judgment without changing business work.

    Human decides whether and how to use every material conclusion.
  2. E2
    Offline Evaluation

    Test against accepted historical or synthetic-safe cases outside the live process.

    Cases, rubric, thresholds, and acceptance authority are human-approved.
  3. E3
    Shadow Mode

    Observe or produce parallel outputs without changing the live business process.

    Output is non-operative; comparison, logging, privacy, and stop rules remain explicit.
  4. E4
    Copilot Mode

    Support named users while they review every material recommendation or action.

    A named user retains approval, correction, override, and abandonment authority.
  5. E5
    Bounded Automation

    Execute only explicitly authorized actions within narrow scope, permissions, thresholds, and recovery.

    Human owners define exceptions, escalation, rollback, monitoring, and stop authority.
  6. E6
    Limited Expansion After Evidence

    Consider additional users, cases, or actions only through a fresh evidence and authority decision.

    Expansion is a new decision; previous evidence does not automatically transfer.

THREE EVIDENCE PLANES

Measure business, system, and adoption evidence separately.

A technically functioning system can still lack business usefulness or adoption readiness. The three layers share traceable evidence but retain separate measures, baselines, owners, and verdicts.

SHARED TRACEABILITY · SEPARATE MEASURES, BASELINES, OWNERS, AND VERDICTS

B

Business Evidence

Determine whether the capability helps the intended work and outcome compared with the accepted baseline or alternative.

POSSIBLE MEASURES
  • cycle time
  • cost or effort
  • quality, service, availability
  • loss, rework, error
  • revenue or conversion only where causal attribution is defensible
S

System Evidence

Determine whether expected behavior is reliable, bounded, recoverable, and stable across accepted cases and versions.

POSSIBLE MEASURES
  • task completion
  • rubric or accuracy result
  • failure and escalation
  • latency and run cost
  • recovery success
  • regression stability
A

Adoption Evidence

Determine whether intended users can understand, use, correct, trust appropriately, and continue using the capability.

POSSIBLE MEASURES
  • eligible and active users
  • repeat use
  • correction and override
  • abandonment
  • confidence and usefulness
  • time-to-proficiency

NO AGGREGATE READINESS SCORE

This website shows no fabricated target, threshold, score, or result. Baselines and targets must be defined from real project evidence and client decisions.

SEPARATE VERDICTS

Passing one dimension does not make the capability ready.

Every verdict answers a different question and remains owned by the appropriate human authority. Verdicts must not be averaged, collapsed into one readiness score, or silently overridden by technical performance.

  1. 01BUSINESS USEFULNESS

    Does it improve the intended work or decision relative to the accepted baseline or alternative?

    HUMAN VERDICT
  2. 02EVIDENCE SUFFICIENCY

    Are representative normal, exception, and failure cases sufficient for the decision being considered?

    HUMAN VERDICT
  3. 03TECHNICAL FUNCTION

    Does the implementation perform the bounded task consistently within accepted conditions?

    HUMAN VERDICT
  4. 04SECURITY & PERMISSIONS

    Are data, access, authority, logging, and human approval boundaries acceptable?

    HUMAN VERDICT
  5. 05OPERATIONAL RECOVERABILITY

    Can owners detect failure, fall back, recover, roll back, and stop?

    HUMAN VERDICT
  6. 06ADOPTION READINESS

    Can intended users understand, use, correct, and support the capability in real work?

    HUMAN VERDICT
  7. 07ECONOMIC ATTRACTIVENESS

    Is expected value defensible relative to cost, effort, alternatives, and residual risk?

    HUMAN VERDICT

Seven verdicts · zero pre-filled status · no automatic release.

GATE 05 / VALIDATION DISPOSITION

Validation exists to support a decision—not to prove that AI must launch.

DECISION INPUTS

  1. 01ACCEPTED BUSINESS, SYSTEM, AND ADOPTION EVIDENCE
  2. 02SEVEN SEPARATE HUMAN VERDICTS
  3. 03KNOWN LIMITATIONS AND RESIDUAL RISKS
  4. 04PERMISSIONS, FALLBACK, RECOVERY, AND STOP CONDITIONS
  5. 05NAMED DECISION AUTHORITY AND ACCEPTED SCOPE

VALIDATION DISPOSITION?

  • CONTINUE

    continue only within the explicitly approved next boundary.

  • REVISE

    change the capability, evidence plan, controls, or scope and re-evaluate.

  • HOLD BOUNDED

    keep the capability at its current approved exposure without expansion.

  • STOP

    cease the proposed use or operation and execute the accepted stop/fallback path.

  • DEFER

    postpone the decision until named evidence, authority, readiness, or conditions exist.

Accepted evidence + separate verdicts + known limits + human authority

eligible for a VALIDATION DISPOSITION

A validation disposition is a human decision record. It does not itself authorize production release, broader access, autonomy, integration, or expansion.

This public page defines a validation method. It does not evaluate a client, run a test, accept a risk, set a target, issue a release decision, or produce an operational authorization.

Evidence category and claim type are separate dimensions.

EVIDENCE RECORD TYPES

TEST EVIDENCE
traceable results from accepted cases, criteria, versions, and conditions.
BUSINESS EVIDENCE
accepted baseline, comparator, workflow outcome, and business-owner observation.
USER EVIDENCE
corrections, overrides, abandonment, confidence, usefulness, and proficiency from intended users.
CONTROL EVIDENCE
permissions, logs, approvals, fallback, recovery, rollback, and stop-condition evidence.

These are generic evidence categories, not proof that any evidence currently exists.

CLAIM / EVIDENCE TYPE

FACT
accepted generic method definition or observed implementation state with source.
HYPOTHESIS
client-specific usefulness, reliability, safety, adoption, economics, exposure, or expansion assumption requiring validation.
UNKNOWN
missing baseline, case, data, permission, user, failure, recovery, target, owner, acceptance, cost, or value evidence.
RECOMMENDATION
bounded next validation or disposition decision, never a fabricated result, verdict, approval, release, or promise.

Evidence category says what record supports review. Claim type says what kind of statement is being made. Neither is a verdict or release status.

POSSIBLE METHOD OUTPUTS

Validation makes the evidence and decision record inspectable.

Possible outputs depend on approved scope and accepted evidence. This page does not claim that any output exists or has been accepted.

  1. 01Evaluation and Validation Report
  2. 02Business, System, and Adoption Scorecard
  3. 03Failure Mode Register
  4. 04Business Acceptance Record
  5. 05Known Limitations and Residual Risks
  6. 06Go / Revise / Hold / Stop / Defer Recommendation
Continue to Operating Partnership ↗
Back to Capability Package ↗Return to FDE Delta overview ↗

OPEN A CONVERSATION

Build something worth testing.

For AI-native products, global GTM, or independent projects, choose a channel below.

WECHAT

Scan to connect on WeChat

Scan to connect on WeChat