# Synthesis Reporting > Synthesis Reporting - Skill: `handsomeboy990/synthesis-reporting` (Agent Skill, multi-file: 4 files) - Install (CLI): `npx skillmds@latest add handsomeboy990/synthesis-reporting` - Raw SKILL.md: https://api.skillmd.com/api/skills/handsomeboy990/synthesis-reporting/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Research & Search - Author: Handsomeboy990 (https://skillmd.com/u/handsomeboy990) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/handsomeboy990/synthesis-reporting --- # Synthesis Reporting Research that is never written up clearly is research that was not done, as far as the reader is concerned. Synthesis turns a pile of attributed findings into an answer to the question, stated so a reader can act on it and check it. It is the output half; `source-research` and `source-verification` are the input. ## 1. Conclusion first The reader needs the answer before the evidence, not after it. ``` lead the answer to the question, in the first lines, as directly as the evidence allows qualify with the confidence and the conditions, in the same breath, so the answer is not read as more certain than it is then the evidence that supports it, for the reader who needs to check ``` A report that withholds the conclusion until the end forces every reader to do the synthesis the report was supposed to do. ## 2. Support with attributed evidence ``` attribute every load-bearing statement to a consulted source, per research-core type mark fact, inference, and opinion so the reader knows which is which trail the reader can follow any claim to its source and location weight the evidence for the conclusion is presented in proportion to how much it carries; a decisive source is not buried beside a minor one ``` ## 3. Quantify confidence ``` high multiple independent primary sources agree; verified moderate good sources, some gaps or minor conflict, reasonable inference low thin sources, unresolved conflict, or heavy inference; say so plainly state the confidence per conclusion, not one global hedge; different parts of an answer are known to different degrees ``` Confidence is a statement the reader acts on. "We are highly confident of X and can only guess at Y" is more useful than a uniform "this is our best understanding". ## 4. Surface disagreement and gaps ``` conflict where sources disagreed, present it, per research-core; do not resolve it silently in favour of the tidy answer gap what could not be established, and what would establish it implication what the gaps mean for the decision: does the missing piece change the answer, or only its confidence ``` ## 5. Recommend without overreaching ``` recommend when the question asks for a recommendation, give one, tied to the evidence and the decision's weighting bound the recommendation does not claim more than the evidence supports; a moderate-confidence finding produces a recommendation hedged to match alternatives name what a different weighting or a filled gap would change separate keep the finding (what is true) distinct from the recommendation (what to do about it); a reader may accept one and not the other ``` ## 6. Prohibitions - Never bury the conclusion at the end; lead with it, qualified. - Never present a load-bearing claim without its attribution. - Never state one global confidence for an answer known to different degrees. - Never resolve a real source conflict silently to reach a clean answer. - Never omit a gap that could change the decision. - Never let the recommendation claim more certainty than the finding supports. - Never blend fact, inference, and opinion into an undifferentiated narrative. ## 7. Protocol 1. State the conclusion to the question first, with its confidence and conditions. 2. Present the supporting evidence, each attributed and typed. 3. Give confidence per conclusion, not one blanket hedge. 4. Surface the conflicts and the gaps, with what each means for the decision. 5. Recommend, if asked, bounded to what the evidence supports, with alternatives. 6. Keep finding and recommendation separate. 7. Provide the evidence trail so any claim can be checked. ## 8. Auto-critique Score from 0 to 5: conclusion led and qualified, every load-bearing claim attributed and typed, confidence stated per conclusion, conflicts surfaced, gaps named with their effect on the decision, recommendation bounded to the evidence, finding kept distinct from recommendation. Threshold: no axis below 3, average at least 4. A conclusion presented as more certain than the evidence supports, or a load-bearing claim with no attribution, caps the score until fixed. ## 9. Interfaces - Upstream: `research-core` sets the evidence and confidence standard, `source-research` supplies the attributed findings, `source-verification` supplies the verdicts and confidence, `competitive-analysis` supplies the comparative matrix. - Downstream: `report-writing` renders the synthesis as a formal document for a decision; the career and business skills consume the synthesised answer. - Lateral: `self-critique` for the reviewer roles, `decision-records` when the synthesised answer informs a recorded decision. ## 10. Note on scope This skill writes up research findings; it does not gather them. A synthesis with no `source-research` behind it has nothing to synthesise and is not written from the researcher's prior belief. If the findings are thin, the honest synthesis says so and reports low confidence, rather than compensating with confident prose.