Behavioral Diagnostic Engine · Research sample
Institutional Risk Coach Memo
Deterministic scorecard in · fine-tuned coach out. Synthetic long-only book, seed 42.
CRO MEMO · CONTROL
July 2026
Alex Behm · Research notes
Subjectcontrol
SeverityNone
Horizon252 days
Coachbde-coach
Output engine Qwen3-8B QLoRA · local Ollama · GGUF LoRA adapter

Bottom line. Control trader is clean on all three v1 biases. Coach correctly returns a no-material-flags memo instead of inventing interventions.

To
Chief Risk Officer
From
Institutional Risk Coach
Date
15 July 2026
Subject
Scorecard review — no material behavioral-risk flags present

Disposition: clear Overconfidence: clear Anchoring: clear

PGR / PLR
0.027 / 0.003
No disposition flag
Turnover
2.42×
Moderate
Underwater
28.2%
Within band
Sell < cost
23.3%
Exits allowed

Diagnosis

No material behavioral-risk flags are present on this scorecard. Annualized turnover ≈ 2.42 (Moderate), Sharpe ≈ 0.24, underwater day share 28.15%, sell-below-cost rate 23.33%. The manager is not exhibiting a pattern of excess turnover, cost-basis anchoring, or sell-below-cost reluctance relative to the supplied risk baseline. Decisions appear consistent with a disciplined risk-policy framework on the current pack.

Intervention

No mandatory risk-policy intervention is required based on the supplied scorecard. The portfolio is operating within acceptable behavioral-risk boundaries. Continue monitoring the PGR/PLR gap, turnover vs Sharpe, and underwater share / sell-below-cost rate on the next review pack. No change to the risk policy toolkit is indicated at this time.

Scorecard metrics cited

MetricValueRead
PGR0.0267Below disposition floor
PLR0.0029Some losses realized
Disposition ratio9.11Not flagged under v1 rules
Turnover (ann.)2.42 · ModerateBelow excess-churn threshold
Sharpe (ann.)0.24Performance context only
Underwater day share28.15%Below anchoring threshold pair
Sell-below-cost rate23.33%Willing to exit below cost
Total return1.98%Clean control path

Design rule   Coach may only narrate fields on the scorecard. Bias existence is decided by the deterministic evaluator, not the LLM.