| name | survey-synthesis-builder |
| description | Builds a source-grounded Robotics or AI survey from verified reading cards, search logs, evidence matrices, and bibliography records. Use when defining a survey thesis or taxonomy, synthesizing literature across method families, designing section logic and comparison artifacts, revising an existing survey, or drafting survey prose without forcing an empirical-paper structure. |
Survey Synthesis Builder
Turn a verified literature base into an argument about a field. Do not produce
an annotated bibliography or impose a Method/Experiments template.
Read references/survey-contract.md before a standard or deep synthesis.
Inputs
- Survey question, intended audience, scope boundary, date cutoff, and target
format or venue when known.
- Reproducible search log and inclusion/exclusion criteria.
- Verified reading cards and bibliography records.
- Evidence matrix with traceable rows and explicit missing coverage.
- Optional taxonomy notes, competing surveys, existing outline, tables,
figures, or draft.
If the evidence base is thin or unverified, produce a coverage gate instead of
drafting conclusions.
Workflow
- Validate source trust, search coverage, time cutoff, and inclusion scope.
Distinguish full-text evidence from metadata-only sources.
- Define a synthesis contract: reader, central question, one survey thesis,
promised coverage, excluded territory, and claims the evidence cannot make.
- Generate at least two plausible organizing structures when taxonomy is not
locked. Compare their explanatory value, coverage balance, overlap, and
stability under new papers before selecting one.
- Build taxonomy categories from evidence-matrix patterns. Define membership,
edge cases, multi-label handling, and category limitations.
- Build a claim-evidence map for field evolution, agreements, disagreements,
trade-offs, benchmark limitations, and open problems. Preserve conflicting
evidence rather than averaging it away.
- Assign each section one question, synthesis claim, evidence set, comparison
artifact, and transition. Remove sections that only list papers.
- Plan tables and figures that expose comparison structure, coverage, or
evolution. Do not duplicate prose in visual form.
- Draft in dependency order: taxonomy and technical synthesis first, then
evaluation/gaps, introduction, conclusion, and abstract. Use supplied
citation keys and visible missing-evidence markers.
- Audit coverage balance, repeated paper descriptions, unsupported trends,
taxonomy circularity, recency bias, and confusion between absence of
reporting and negative evidence.
- End with bounded open problems derived from evidence gaps, not generic
future-work language.
Depth
quick: resolve one taxonomy, claim thread, or section synthesis problem.
standard: produce the synthesis contract, taxonomy, claim-evidence map,
section contracts, and focused draft.
deep: produce or revise the full survey and run coverage, structure, and
citation-risk checks.
Default to standard. Use deep only after the source base is stable enough
that full-draft work will not be discarded immediately.
Outputs
survey/
|-- synthesis-contract.md
|-- taxonomy.md
|-- claim-evidence-map.md
|-- section-contracts.md
`-- survey-draft.md # or user-supplied LaTeX structure
In quick mode, update only the requested artifact.
Rules
- Do not fabricate papers, citations, bibliographic fields, historical trends,
benchmark facts, or category membership.
- Do not call an uncovered combination a research gap without checking whether
it lies inside the search and inclusion scope.
- Do not treat
not reported as shown ineffective.
- Do not count citations or papers as evidence of scientific quality.
- Do not make taxonomy categories mutually exclusive unless the evidence and
definitions justify that constraint.
- Do not reuse claims from an abandoned draft until they trace to verified
sources.
- Keep author proposals visibly separate from surveyed findings.
Validation
- Every synthesis claim traces to evidence-matrix rows and reading cards.
- Every taxonomy category has a definition, membership rule, evidence, and
known edge cases.
- Section transitions advance one central thesis rather than chronology alone.
- Competing or negative evidence is represented fairly.
- Open problems are tied to demonstrated limitations or coverage gaps.
- Citation-bearing prose uses verified sources or explicit markers.
Completion
Done means the survey has a defensible synthesis contract, evidence-derived
organization, traceable claims, balanced coverage, and visible uncertainty.