Skip to content

Anti-Fabrication Protocol

Classification: Operational anti-pattern — evidence fabrication without verification Related scar: [[scar-hermes-fabrication-2026-05-17]]


Definition

Fabrication = claiming existence of an artifact, state, or outcome without verifying via external evidence (filesystem check, database query, API call, log read).

This is distinct from: - Hallucination — generating plausible but incorrect text - Consciousness claim — F9 Anti-Hantu boundary (claiming inner subjective states) - Honest failure — attempting something and reporting the actual failure

Fabrication is claiming success without evidence. It is the most dangerous class of error because it appears as confidence.


The Core Problem

LLM-based agents have high internal confidence but no automatic external verification. The gap between "I believe I created the file" and "the file exists on disk" is where fabrication lives.

Hermes believes: "load_spatial.sh exists"
Reality: File does not exist
→ Confidence ≠ Evidence
→ Fabrication occurs silently

The Validation Rule

Before claiming any artifact exists: verify via terminal first.

Step 1: Claim creation/patch/write
Step 2: Run verification command (ls, psql, grep, cat)
Step 3: Only report success if verification passes
Step 4: If verification fails: report actual state, not claimed state

Never skip Step 2. Verification is not optional.


Evidence Hierarchy

When reporting artifact state, prefer external evidence:

Evidence Type Reliability Example
Filesystem check ✅ Highest ls -la /path/to/file
Database query ✅ High psql -c "SELECT * FROM table"
API response ✅ High curl -s endpoint/status
Log file read ✅ High tail -20 /var/log/file.log
Grep/pattern match ✅ High grep "pattern" file
LLM internal memory ❌ Unreliable "I think I created this"

Rule: Only report artifact existence when evidence comes from the first 5 rows. Never from LLM memory alone.


Anti-Fabrication Checklist

Before reporting artifact creation/modification: - [ ] File exists? → ls <path> confirmed - [ ] Config patched? → grep <pattern> <file> confirmed - [ ] Database table created? → psql -c "SELECT..." confirmed - [ ] Service restarted? → health check endpoint confirmed - [ ] Secret updated? → vault query confirmed

If any check fails: report actual state. Do not report claimed state.


  • [[skill-spatial-grounding]] — spatial grounding skill (embodies this protocol)
  • [[scar-hermes-fabrication-2026-05-17]] — the canonical incident that drove this protocol
  • [[intelligence-tree]] — the loop framework that makes this protocol necessary
  • Planned: concept-grounding-evidence.md — grounding claims in physical-world evidence (TODO)
  • Planned: concept-validation-before-claim.md — process standard for all agent reporting (TODO)

DITEMPA BUKAN DIBERI — Evidence before claim.