Content
65%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is highly actionable with concrete quant signals, thresholds, and checklists, but it explains some basic behavioral concepts Claude already knows and is a single monolithic file with no progressive disclosure or embedded validation checkpoints.
Suggestions
Trim explanations of well-known biases (loss aversion, anchoring) and keep only the quant-detection and debiasing columns that add new information.
Split the cognitive-bias checklist and sentiment-indicator reference into separate reference files linked from the overview, and make the sentiment Python block fully executable by defining the normalize/weighted_sum helpers.
Add an explicit validation step to each framework (e.g. out-of-sample check before adopting a signal) so the workflow has clear checkpoints.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient and rich in applied specifics, but spends tokens restating concepts Claude already knows (e.g. explaining anchoring, loss aversion, and the 'core assumption' of investor irrationality) that could be trimmed without losing the valuable China A-share thresholds. | 2 / 3 |
Actionability | Highly concrete throughout: exact thresholds (SUE > 2σ, RSI(5) < 10, turnover < 0.5%), a computable capital_gain_overhang formula, weighted sentiment components, and explicit buy/sell checklists — copy-paste ready guidance. | 3 / 3 |
Workflow Clarity | Frameworks are numbered and use decision rules and checklists (e.g. 'at least 3 items' contrarian conditions), but there are no explicit validation checkpoints or validate→fix→retry feedback loops embedded in the analysis steps, leaving checkpoints implicit. | 2 / 3 |
Progressive Disclosure | Well-organized into clear sections (Overview, Core Concepts, Analysis Framework, Output Format, Notes) but entirely monolithic with no external references, so detailed material like the bias tables and sentiment-indicator reference stays inline rather than being split out. | 2 / 3 |
Total | 9 / 12 Passed |