Evidence Collection

World-Class Evidence Preflight

This operator view shows which external and human evidence is still blocked, which commands prepare editable submission drafts, and why preflight never counts as accepted evidence.

Decisioncollection-preflight-blocked
Pending4
Ready1
Blocked3
Source6/13

Submission Kit

Generate drafts only after real provider, human-review, native-permission, or native-client work exists. Drafts remain non-evidence until valid aggregate artifact refs and SHA-256 digests are supplied.

  • submissions directory: evidence/world_class/submissions
  • drafts count as evidence: false
  • preflight accepts evidence: false

Evidence Queue

external · blocked

Provider Holdout

Evidence
provider-holdout
Ledger
pending
Intake
awaiting-submission
Review
awaiting-submission
Draft
evidence/world_class/submissions/provider-holdout.json

Next Action

Set OPENAI_API_KEY in the operator shell; never commit or print the value.

python3 scripts/yao.py world-class-submission-kit . --evidence-key provider-holdout --output-dir evidence/world_class/submissions

Prechecks

file Output eval cases
Current
present
Status
pass
Action
Keep output holdout cases available before provider execution.
file Provider runner
Current
present
Status
pass
Action
Use the provider runner instead of the local command runner for model-backed evidence.
env Provider credential
Current
not-set
Status
missing
Action
Set OPENAI_API_KEY in the operator shell; never commit or print the value.
env Provider model
Current
not-set
Status
optional
Action
Optionally set YAO_OUTPUT_EVAL_MODEL; the runbook defaults to gpt-4.1-mini.

Source Checks

model_executed_count Provider model run
Current
0
Expected
>0
Status
blocked
Action
Run provider-backed output-exec with real credentials.
timing_observed_count Timing observed
Current
10
Expected
>0
Status
pass
Action
Provider execution should record timing metadata.
token_observed_count Token usage observed
Current
0
Expected
>0
Status
blocked
Action
Provider execution should return non-estimated token usage.

Runbook

  • YAO_OUTPUT_EVAL_MODEL=gpt-4.1-mini OPENAI_API_KEY=<redacted> python3 scripts/yao.py output-exec --provider-runner openai --timeout-seconds 60
  • python3 scripts/yao.py skill-os2-audit . --generated-at <YYYY-MM-DD>
  • Copy evidence/world_class/templates/provider-holdout.intake.json to evidence/world_class/submissions/provider-holdout.json and fill only real evidence fields.
  • python3 scripts/yao.py world-class-intake . --submissions-dir evidence/world_class/submissions
human · ready-for-human-review

Human Adjudication

Evidence
human-adjudication
Ledger
pending
Intake
awaiting-submission
Review
awaiting-submission
Draft
evidence/world_class/submissions/human-adjudication.json

Next Action

Assign a real reviewer identity before claiming human adjudication.

python3 scripts/yao.py world-class-submission-kit . --evidence-key human-adjudication --output-dir evidence/world_class/submissions

Prechecks

file Blind review kit
Current
present
Status
pass
Action
Open the blind review kit and record real reviewer choices.
file Decision template
Current
present
Status
pass
Action
Fill winner_variant values with real A/B decisions.
human Human reviewer
Current
external-human-action
Status
human-required
Action
Assign a real reviewer identity before claiming human adjudication.

Source Checks

pair_count Review pairs exist
Current
5
Expected
>0
Status
pass
Action
Generate the blind A/B review pack.
pending_count No pending decisions
Current
5
Expected
==0
Status
blocked
Action
Record a reviewer choice for every pair.
judgment_count Judgments complete
Current
0
Expected
==pair_count
Status
blocked
Action
Every pair needs one valid human judgment.
invalid_decision_count No invalid decisions
Current
0
Expected
==0
Status
pass
Action
Fix malformed winner/confidence entries.

Runbook

  • python3 scripts/yao.py output-review-kit --write-template
  • Open reports/output_review_kit.md and choose A or B for each pair without opening the answer key.
  • python3 scripts/adjudicate_output_review.py --write-template
  • Edit reports/output_review_decisions.json with winner_variant values and reviewer metadata.
  • python3 scripts/yao.py output-review
  • python3 scripts/yao.py skill-os2-audit . --generated-at <YYYY-MM-DD>
  • Copy evidence/world_class/templates/human-adjudication.intake.json to evidence/world_class/submissions/human-adjudication.json and fill only real evidence fields.
  • python3 scripts/yao.py world-class-intake . --submissions-dir evidence/world_class/submissions
external · blocked

Native Permission Enforcement

Evidence
native-permission-enforcement
Ledger
pending
Intake
awaiting-submission
Review
awaiting-submission
Draft
evidence/world_class/submissions/native-permission-enforcement.json

Next Action

Attach a real target-client or external installer runtime guard; metadata fallback is not enough.

python3 scripts/yao.py world-class-submission-kit . --evidence-key native-permission-enforcement --output-dir evidence/world_class/submissions

Prechecks

file Permission policy
Current
present
Status
pass
Action
Keep approved high-permission capabilities explicit.
file Runtime probes
Current
present
Status
pass
Action
Refresh runtime permission probes after packaging changes.
external Native guard
Current
external-integration-required
Status
external-required
Action
Attach a real target-client or external installer runtime guard; metadata fallback is not enough.

Source Checks

native_enforcement_count Native enforcement
Current
0
Expected
>0
Status
blocked
Action
Collect real target-client or external runtime guard proof.
failure_count Probe failures
Current
0
Expected
==0
Status
pass
Action
Runtime permission probes must stay clean.
installer_enforcement_ready Installer support
Current
True
Expected
true
Status
pass
Action
Installer enforcement is supporting evidence, not native proof.

Runbook

  • Implement or connect a real target client or external installer runtime guard that blocks undeclared network, file_write, or subprocess capabilities.
  • Update the generated target adapter only when the guard is actually enforced by that target.
  • python3 scripts/yao.py package . --platform openai --platform claude --platform generic --platform vscode --output-dir dist --zip
  • python3 scripts/yao.py install-simulate . --package-dir dist --install-root dist/install-simulation
  • python3 scripts/yao.py runtime-permissions . --package-dir dist
  • python3 scripts/yao.py skill-os2-audit . --generated-at <YYYY-MM-DD>
  • Copy evidence/world_class/templates/native-permission-enforcement.intake.json to evidence/world_class/submissions/native-permission-enforcement.json and fill only real evidence fields.
  • python3 scripts/yao.py world-class-intake . --submissions-dir evidence/world_class/submissions
external · blocked

Native Client Telemetry

Evidence
native-client-telemetry
Ledger
pending
Intake
awaiting-submission
Review
awaiting-submission
Draft
evidence/world_class/submissions/native-client-telemetry.json

Next Action

Install a real Browser, Chrome, IDE, or provider client that emits metadata-only events.

python3 scripts/yao.py world-class-submission-kit . --evidence-key native-client-telemetry --output-dir evidence/world_class/submissions

Prechecks

file Native telemetry host
Current
present
Status
pass
Action
Use the native host to receive metadata-only client events.
file Hook recipes
Current
present
Status
pass
Action
Refresh telemetry hook recipes before external client installation.
external External client
Current
external-integration-required
Status
external-required
Action
Install a real Browser, Chrome, IDE, or provider client that emits metadata-only events.

Source Checks

external_source_events External events
Current
0
Expected
>0
Status
blocked
Action
Import at least one metadata-only event from a real client.
adoption_sample_count Adoption sample
Current
0
Expected
>0
Status
blocked
Action
Telemetry must include adoption outcome evidence.
raw_content_allowed Raw content blocked
Current
False
Expected
false
Status
pass
Action
Telemetry must stay metadata-only.

Runbook

  • python3 scripts/telemetry_native_host.py . --write-launcher /tmp/yao-telemetry-host.sh --write-manifest /tmp/yao-telemetry-host.json --allowed-origin chrome-extension://<extension-id>/
  • Install the generated native messaging manifest for the real client and send at least one accepted skill_activation or skill_output event.
  • python3 scripts/yao.py telemetry-import . --input-jsonl .yao/telemetry_spool/external_events.jsonl
  • python3 scripts/yao.py skill-atlas --workspace-root .
  • python3 scripts/yao.py skill-os2-audit . --generated-at <YYYY-MM-DD>
  • Copy evidence/world_class/templates/native-client-telemetry.intake.json to evidence/world_class/submissions/native-client-telemetry.json and fill only real evidence fields.
  • python3 scripts/yao.py world-class-intake . --submissions-dir evidence/world_class/submissions

Safety Boundary

  • Environment variables are displayed only as set or not-set; secret values are never printed.
  • Human-required and external-required states are operator work, not accepted evidence.
  • The world-class ledger remains the only source of truth for ready_to_claim_world_class.