R0021/2026-03-25/Q003 — Self-Audit¶
ROBIS 4-Domain Audit¶
Domain 1: Eligibility Criteria¶
Rating: Low risk
| Criterion | Assessment |
|---|---|
| Criteria defined before searching | Yes — "quantifiable" defined as including numbers/thresholds/testable criteria |
| Criteria stable during research | Yes |
| Criteria appropriate to query | Yes |
Notes: The definition of "quantifiable" was established before analysis began.
Domain 2: Search Comprehensiveness¶
Rating: Some concerns
| Criterion | Assessment |
|---|---|
| Multiple search strategies used | Yes — separate searches per vendor |
| Searches designed to test each hypothesis | Partially — searched for recommendations, not for absence |
| All results dispositioned | Yes |
| Source diversity achieved | Yes — all four named vendors covered |
Notes: Limited to primary documentation pages. Vendor cookbooks, blog posts, and tutorials may contain additional quantifiable guidance not captured here.
Domain 3: Evaluation Consistency¶
Rating: Low risk
| Criterion | Assessment |
|---|---|
| All sources scored using same framework | Yes |
| Evidence typed consistently | Yes — all Factual |
| ACH matrix applied | Yes |
| Diagnosticity analysis performed | Yes |
Notes: Same quantifiable/subjective classification applied consistently across vendors.
Domain 4: Synthesis Fairness¶
Rating: Some concerns
| Criterion | Assessment |
|---|---|
| All hypotheses given fair hearing | Yes — Google's quantifiable elements acknowledged |
| Contradictory evidence surfaced | Yes — Google as partial counterexample |
| Confidence calibrated to evidence | Yes |
| Gaps acknowledged | Yes |
Notes: The classification of "quantifiable" vs. "subjective" involves judgment. Structural recommendations (XML tags, prompt chaining) are actionable but were classified as non-quantifiable. This classification could be debated.
Overall Assessment¶
Overall risk of bias: Some concerns
The main risk is in the classification judgment. The researcher's hypothesis that "prompt engineering is not engineering" benefits from finding subjective guidance. The classification was applied consistently but could be critiqued.
Researcher Bias Check¶
- Confirmation bias: Strong risk. The finding aligns with the researcher's thesis. Mitigated by using a clear, pre-defined classification criterion and acknowledging borderline cases (structural recommendations).
- Framing bias: "Quantifiable" vs. "subjective" is itself a frame that emphasizes what's missing rather than what's present. Alternative framings (e.g., "actionable" vs. "vague") might produce different results.