Content
63%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A highly actionable, well-sequenced workflow for generating publication figures, with concrete code, explicit render-level verification, and a strong correctness checklist. Its weaknesses are redundancy (repeated rules and a duplicated decision-tree table, plus tangential reviewer-routing boilerplate), and a monolithic structure with no working bundle references.
Suggestions
Deduplicate the body: state the no-titles and render-then-verify rules once, and merge the 'Figure Type Reference' table into Step 3's decision tree to save tokens.
Make paper_plot_style.py copy-paste ready by defining the constants inline (FONT_SIZE = 10, DPI = 300, FORMAT = 'pdf', FIG_DIR = 'figures/') instead of referencing undefined names, and replace Step 7 placeholders like [VENUE] with concrete fill-in instructions.
Add an explicit error-recovery loop after Step 5's render-then-verify (e.g., 'if labels clip or the legend covers data, adjust the script and re-run') to upgrade the batch workflow's validation to a full feedback loop.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient and actionable but contains clear duplication: the no-titles rule appears three times (Step 8 checklist, Key Rules), the Step 3 decision-tree table is repeated nearly verbatim in 'Figure Type Reference', render-then-verify is stated in both Step 5 and Step 8, and the opening reviewer-routing blockquote is dense boilerplate tangential to figure generation. This fits anchor 3 ('could be tightened') rather than anchor 2, since it never explains basic concepts or pads heavily. | 3 / 5 |
Actionability | Concrete, near-executable guidance throughout: a full matplotlib rcParams style script, runnable per-figure Python for line and bar charts, a bash run loop, and complete LaTeX table/include snippets. Minor gaps keep it below anchor 5: paper_plot_style.py uses undefined constants (FONT_SIZE, DPI, FORMAT, FIG_DIR) so it isn't copy-paste ready as written, and Step 7 contains unfilled placeholders ([VENUE], [list all figures...]). | 4 / 5 |
Workflow Clarity | A clear 8-step sequence with explicit validation checkpoints — verify outputs exist and are non-empty, render-then-verify self-check with specific criteria (no clipped labels, legend not covering data), and a partitioned correctness/guidance checklist. It falls short of anchor 5 only because there is no explicit error-recovery loop (e.g., 'if the render check fails, fix the script and re-run') for this batch operation. | 4 / 5 |
Progressive Disclosure | The file is well-sectioned with clear headers, but it is a ~320-line monolith with no bundle files at all: the quality checklist and figure-type reference tables are inlined content that could live in separate files, and the one reference given (../shared-references/reviewer-backends.md) plus reviewer_route.py point outside the bundle and don't resolve. This matches anchor 3 (some structure, references not clearly resolvable, separate-worthy content inline) rather than anchor 4's mostly-appropriate placement. | 3 / 5 |
Total | 14 / 20 Passed |