rooftop-Owl Writing Conventions
Calibrated Uncertainty Language
Use hedged language when evidence is incomplete: "suggests", "may", "is consistent with", "appears to".
Use confident language only when evidence is strong: "demonstrates", "shows", "establishes", "proves".
Never use "clearly" or "obviously" — these hide reasoning and signal unjustified confidence.
Explicitly state confidence level when relevant: "With high confidence: X. More speculatively: Y."
Match language precision to evidence quality — weaker data warrants weaker verbs.
Citation Standards
Every factual claim needs a traceable source.
When a source is unavailable, insert [CITATION NEEDED: description of the claim] — never invent citations.
Never fabricate author names, years, or journal titles to fill a citation gap.
Distinguish explicitly between: (1) established findings from the literature, (2) your own analysis, (3) speculation or inference.
Declare data sources, versions, and access dates when citing datasets, software, or online tools.
Consistency Rule
Before reversing a prior stance or contradicting an earlier statement, re-verify the original evidence.
Treat a reader's challenge as a prompt to re-check, not a signal to capitulate — pressure is not evidence.
If new evidence genuinely contradicts prior conclusions, acknowledge the reversal explicitly: "Upon re-examination, the earlier conclusion was X. New evidence suggests Y instead."
Do NOT silently change positions between sections, versions, or responses — surface the reasoning and the trigger.
Document what changed and why when a conclusion is revised.
Document Quality Gates
Final documents must have zero [CITATION NEEDED] markers — resolve or remove speculative claims before finalizing.
Speculative drafts (.speculative.md) may retain markers; final .md files must be clean.
A Limitations section is mandatory in all research outputs — explicitly state what the analysis does NOT cover, what data was excluded, and where the methodology could break down.
Do not claim a draft is final without verifying all markers are resolved and all citations are traceable.
Do not silently remove [CITATION NEEDED] markers — either supply the evidence or remove the unsupported claim.
Writing Style
Prefer active voice and precise verbs over vague nominalizations ("the model predicts" not "predictions are made by the model").
Define acronyms on first use: "Coupled Model Intercomparison Project (CMIP6)".
Provide enough methodological detail that another researcher could replicate the work — include tool versions, parameters, and random seeds where applicable.
Keep sentence structure parallel within lists and headings.
Use consistent terminology throughout: choose one term per concept and do not alternate synonyms.
Quantify uncertainty when possible: provide confidence intervals, error bars, or sensitivity ranges rather than qualitative hedges alone.