Behavioral Diagnostic Engine · Phase 6b
Base vs Adapter Comparison
Same scorecards, same local hardware. Does fine-tuning change domain behavior — or only vibes scores?
EVAL · LOCAL OLLAMA
July 2026
Alex Behm · Research notes
Base quality9.0 / 10
Adapter quality9.5 / 10
Adapter ≥ base2 / 2
GatePASS
Models Base qwen3:8b · Adapter bde-coach · think:false · AMD Radeon 890M

Finding. Heuristic quality is close. Domain correctness is not. Base misreads PGR/PLR as “growth/loss rate,” treats turnover 6.27 as “~6.3%,” and dumps generic markdown. Adapter writes CRO-format memos with correct disposition language and is ~1.7× faster.

Headline results

CaseModelQualityWallDiagnosisIntervention
Biased
3 flags
base qwen3:8b10/1046.7sYesYes
adapter bde-coach10/1031.4sYesYes
Control
0 flags
base qwen3:8b8/1049.6sYesYes
adapter bde-coach9/1022.7sYesYes

Domain correctness — biased scorecard

PGR → “Portfolio Growth Rate (4.62%)” — wrong metric definition
Turnover 6.27 → “~6.3% turnover” — off by two orders of magnitude
disposition_flag → “not actively managed” — invents a process story
Format: generic markdown report with metric dump tables
CRO memo header (TO / FROM / SUBJECT)
PGR/PLR read as disposition signals, not growth rates
Turnover 6.27 treated as high annualized churn
Intervention cites underwater share + sell-below-cost = 0%

Domain correctness — control scorecard

Verbose metric dump despite zero flags
Invented interventions (stop-loss, rebalance) on a clean pack
Turnover 2.42 again misread as percent-scale activity
Subject: “No Material Behavioral-risk Flags Present”
Explicit “no mandatory intervention”
Continues monitoring language only
~2.2× faster than base on this case

What this gate proves

ClaimEvidence
Adapter ≥ base on both gold cases2/2 quality comparison
Speed improves with specialization~48s → ~27s average wall
Fine-tuning changes behavior, not just scoresMetric interpretation + memo form
Local inference is enough for the bake-offNo cloud eval VM required

Method note   Always sample-read memos. Citation-token heuristics can score a base memo high while the definitions are wrong.