Result verification

Skill Chenghao-Wu/superscientist/skills/result-verification

Use before marking any workflow stage as completed — verify outputs exist, are valid, and meet success criteria defined in workflow-state.jsonFrom its SKILL.md

Install

npx -y skills add Chenghao-Wu/superscientist --skill result-verification

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.2 KB, as published. Nobody here has run it

Result Verification

Overview

Claiming a stage is complete without verification is dishonesty, not efficiency.

Core principle: Evidence before claims, always.

Violating the letter of this rule is violating the spirit of this rule.

The Iron Law

NO STAGE MARKED COMPLETED WITHOUT PASSING VERIFICATION

If verification fails, the stage is marked failed and superscientist:systematic-debugging is invoked.

The Verification Checklist

You MUST complete every check. A single failure = stage is failed.

1. Output Files Exist and Are Non-Empty

# For each expected output in workflow-state.json stage.outputs:
test -s "$output_file" || echo "FAIL: $output_file missing or empty"

2. Output Files Are Parseable

Try to read/parse the output files
Check for truncation (incomplete last line, missing footer)
Check for corruption (binary garbage in text files)

3. Success Criteria Met

Read the success_criteria from workflow-state.json for this stage. Check each criterion against actual output:

Criterion type	How to verify
Convergence metric	Parse output, check final value meets threshold from success_criteria
Numerical result	Extract value, compare against target or reference
Output file produced	Check file exists, non-empty, and in expected format
Visualization	Check image file valid, non-zero size, expected content present
Script/process completion	Check exit code is 0 and no fatal errors in logs

4. Sanity Checks

Domain-specific reasonableness checks (derive from experiment design):

Output values within physically/mathematically reasonable ranges?
No obviously unphysical values (NaN, infinite, negative where impossible)?
Results consistent with prior stages or known references?
For plots: axes labeled, data present, no obvious artifacts?

5. No Silent Failures

Check log/output files for:

Warning messages that indicate problems
Error messages that were caught but indicate incomplete results
"WARN", "ERROR", "WARNING", "FAILED" in log files

Reporting

After verification, report:

If passed:

Verification PASSED for stage-N ([name]):
- All N output files present and non-empty
- Success criteria met: [specific evidence]
- Sanity checks passed: [brief summary]
→ Marking stage as completed.

If failed:

Verification FAILED for stage-N ([name]):
- Failed check: [which check failed]
- Evidence: [what was found]
- Expected: [what was expected]
→ Marking stage as failed. Invoking systematic-debugging.

Red Flags

Thought	Reality
"Output file exists, good enough"	Check it's non-empty and parseable too.
"Calculation converged, must be right"	Check the answer is physically reasonable.
"No errors in the log"	Check for warnings too.
"I'll verify the next stage instead"	Every stage gets verified. No shortcuts.
"Results look reasonable"	Compare against success criteria, not intuition.