sci-writing

name

sci-writing

description

Read technical control and applied-math papers in depth, extract per-paper innovation points, derive IEEE-suitable SCI paper directions, expand a supporting reference pool with emphasis on IEEE Xplore papers, and co-develop a double-column LaTeX manuscript with MATLAB simulations. Use when the task involves local PDFs, TeX, MATLAB, reference figures, or citation building for literature digestion, research-gap mapping, innovation ranking, IEEE manuscript scaffolding, bibliography expansion, simulation comparisons, or iterative write-evaluate-optimize research workflows.

SCI-WRITING

Overview

Drive a research task from literature to a compilable IEEE manuscript and reproducible MATLAB simulations. Work from evidence: every claimed contribution must trace back to a paper-reading matrix, a concrete gap, a theorem or algorithm delta, a maintained citation pool, and figures that show the gain.

Workflow

1. Gather source artifacts

• Find all cited or mentioned papers, local PDFs, TeX folders, MATLAB files, and reference figures.
• Stop and ask for missing papers if the request requires reading every paper but the files are unavailable.
• Prefer local primary sources. Do not infer innovations from abstracts alone if the full paper is available.

2. Read every paper completely

• Load paper-reading.md and fill the innovation matrix before proposing ideas.
• Extract the problem class, model assumptions, main theorem delta, proof tools, trigger or observer or controller structure, comparison baselines, figure settings, and limitations.
• Capture one replication target per paper: which figure or result should be recreated or matched first.

3. Propose and rank five directions

• Load direction-selection.md.
• Produce exactly five candidate SCI directions grounded in the extracted gaps.
• Score each direction for novelty, theorem difficulty, simulation cost, reproducibility, and IEEE venue fit.
• Reject directions that are only application renaming, parameter tuning, or cosmetic extensions.

4. Pick the easiest credible direction

• Choose the direction with the best balance of novelty and execution risk, not the most ambitious one.
• State the innovation hypothesis in two to four precise sentences.
• Define the minimum evidence needed: theorem delta, algorithm delta, baseline set, and target figures or tables.

5. Expand the literature pool after direction lock

• Load reference-expansion.md.
• Use the seed papers to grow a working bibliography toward roughly 20 to 30 relevant references for a full manuscript.
• Prefer IEEE Xplore papers for direct-method competitors, recent follow-up work, and venue-aligned related work whenever suitable papers exist there.
• Keep indispensable non-IEEE sources only when they are foundational, define the dataset or benchmark, or are the original source of a theorem or method the paper relies on.
• Build a living citation bank that covers direct competitors, theory foundations, datasets or benchmarks, performance metrics, and application context.

6. Scaffold the project

• Run scripts/init_ieee_project.py <target-dir> --title "..." --direction "...".
• Use the generated notes/ templates to track literature, ranking, and iteration decisions.
• Keep the paper in IEEE double-column format from the start. Do not draft in an unrelated template and retrofit later.

7. Co-develop theory, MATLAB, and LaTeX

• Load ieee-outline.md for section-level requirements.
• Draft sections only when a section's claims can be backed by equations, proof structure, or simulation evidence.
• When a section needs external support for a dataset, theorem, definition, benchmark, or engineering claim, search and add a matching primary reference before finalizing the text.
• Update TeX whenever MATLAB outputs, metrics, or conclusions change.
• Keep a tight claim-to-evidence mapping: every contribution sentence must point to a theorem, proposition, algorithm block, or figure or table.

8. Run the simulation improvement loop

• Load simulation-loop.md.
• Reproduce a reference-style figure before optimizing the new method.
• Add fair baselines, ablations, and stress tests that make the improvement measurable.
• Work in cycles: write, run, inspect, diagnose, optimize, rerun, update TeX, and rebuild the PDF.

9. Build and review the manuscript

• Run scripts/build_ieee_pdf.sh <project-dir>/paper after meaningful changes.
• Treat build warnings, unreadable figures, missing references, and unsupported claims as blockers.
• Treat unsupported citations as blockers too: if a sentence makes a factual, historical, methodological, or dataset-specific claim, either cite a source or rewrite the sentence.
• If results do not support the claimed innovation, weaken the claim or change the method. Do not keep a gap between narrative and evidence.

Operating Rules

• Read every referenced paper that the user identifies as relevant before freezing the research direction.
• After selecting the direction, expand beyond the seed set. A submission-grade draft should usually cite about 20 to 30 relevant papers unless the topic is unusually narrow.
• Prefer IEEE Xplore for the majority of added references when the field naturally lives in IEEE venues.
• Do not pad the bibliography. Every reference must serve a role: closest competitor, theorem foundation, model source, dataset source, metric source, benchmark source, or application context.
• Prefer theorem-level novelty, assumption relaxation, trigger reduction, robustness expansion, or complexity reduction over superficial reformulation.
• Keep comparisons anchored to the referenced papers' figures, parameters, and metrics whenever feasible.
• Separate what is proven, what is observed in simulation, and what is only a hypothesis.
• Use concise, technical prose suitable for IEEE journals. Remove speculative or inflated contribution language.
• Record failed variants in notes/iteration_log.md so the next iteration does not repeat them.
• Track citation gaps in a living note file. When a new statement needs support, add the citation task immediately rather than postponing it to the end.

Typical Requests

• "Use $sci-writing to read these event-triggered control papers, extract each paper's innovation points, and propose five IEEE-suitable follow-up directions."
• "Use $sci-writing to pick the easiest publishable direction from these zonotopic observer papers and scaffold an IEEE double-column manuscript with MATLAB simulations."
• "Use $sci-writing to expand these three seed papers into a 25-reference IEEE-centered bibliography for the introduction and related work."
• "Use $sci-writing to align my MATLAB figures with the reference paper, improve the comparison section, and rebuild the PDF after every iteration."
• "Use $sci-writing to find supporting references for this dataset choice, stability argument, and benchmark comparison, then update the .bib and manuscript text."

Resources

• scripts/init_ieee_project.py: Create a paper workspace with an IEEE LaTeX skeleton, a MATLAB starter, and note templates.
• scripts/build_ieee_pdf.sh: Compile the IEEE paper and write the PDF to paper/build/.
• paper-reading.md: Per-paper innovation extraction checklist and matrix.
• direction-selection.md: Five-direction generation and scoring rubric.
• reference-expansion.md: IEEE-centered bibliography growth and citation-gap workflow.
• ieee-outline.md: IEEE section outline and claim-to-evidence requirements.
• simulation-loop.md: Figure reproduction, comparison design, and iteration loop.