icaire-social-performance-analysis
Analyze ICAIRE post-level snapshots when evidence-bounded social performance patterns are needed for the Audience Growth Engine.
SKILL.mdskills/audience-growth-engine/icaire-social-performance-analysis/SKILL.md
GitHubICAIRE Social Performance Analysis
Analyze validated ICAIRE LinkedIn and X observations using the canonical Brain protocol and clearly bounded evidence.
Contract Checklist
- Read the canonical ICAIRE Brain page
protocol/icaire-social-performance-analysisbefore analysis and follow its current instructions. - Use validated post-level snapshots from the Audience Growth Engine, not aggregate summaries as the primary evidence.
- Confirm platform, date, post, metric, language, format, visual, and topic coverage before comparing cohorts.
- Preserve metric definitions and separate observed facts, calculated values, analyst labels, and hypotheses.
- Do not claim causation from observational performance data.
- Label sparse, incomplete, signed-out, or non-comparable data prominently.
- Save analysis artifacts under
/Users/hq/projects/ICAIRE/Audience Growth Engine/without altering raw data. - Do not draft, publish, message, or optimize content in this skill.
Workflow
- Load the governing protocol and inputs:
- Read the live ICAIRE Brain protocol
protocol/icaire-social-performance-analysisand any linked metric definitions. - Locate the latest validated normalized snapshots and the analysis workbook.
- Record exact run IDs, collection times, platform coverage, and analysis window.
- Stop if the protocol is unavailable or the selected data cannot be traced to immutable raw artifacts.
- Anti-patterns: relying on remembered protocol text, analyzing an unvalidated workbook, treating Reem's aggregate report as post-level evidence
- Read the live ICAIRE Brain protocol
- Audit comparability and completeness:
- Quantify posts and observations by platform, date, language, topic, format, and visual status.
- Identify duplicate snapshots and select the intended observation time without deleting history.
- Check metric availability and definitions by platform. Keep non-equivalent metrics separate.
- Decide whether cross-platform analysis is supportable; otherwise analyze each platform separately.
- Anti-patterns: pooling unlike metrics, comparing unequal windows without disclosure, treating null metrics as zero
- Produce descriptive baselines:
- Calculate transparent post-level summaries allowed by the protocol, including distribution and median as well as totals and means where suitable.
- Normalize only with defensible denominators and show the exact formula.
- Identify outliers without allowing a single post to define a general pattern.
- Anti-patterns: rank-ordering on raw totals alone, hiding sample sizes, inventing follower counts or reach denominators
- Test content dimensions:
- Examine topic, observed language, posting time, platform, post format, and visual presence when sample sizes permit.
- Use matched or stratified comparisons where practical to reduce obvious confounding.
- Record each result with cohort definition, sample size, metric, effect direction, uncertainty, and counterexamples.
- Anti-patterns: claiming that visuals caused engagement, interpreting one or two examples as a trend, retrofitting topic labels to performance
- Synthesize evidence-bounded patterns:
- Classify findings as observed pattern, weak signal, insufficient evidence, or data-quality issue.
- Distinguish cross-cutting patterns from platform-specific observations.
- State plausible alternative explanations and the next data needed to strengthen each finding.
- Do not convert patterns into next-week actions here; hand validated findings to the optimization skill.
- Anti-patterns: causal language, cherry-picking winners, hiding contradictory posts, mixing recommendations into analysis
- Validate and hand off:
- Reconcile analysis tables to source row counts and spot-check calculations against the workbook.
- Save reproducible tables, formulas or scripts, and a concise analysis section with exact source run IDs.
- Report what can and cannot be concluded, plus any blocker to cross-platform interpretation.
- Anti-patterns: reporting without source traceability, omitting calculation checks, presenting incomplete X data as definitive
Evidence Labels
- Observed pattern: repeated, traceable association with adequate cohort coverage.
- Weak signal: suggestive direction with limited sample or confounding.
- Insufficient evidence: too little or incomparable data.
- Data-quality issue: missing, stale, ambiguous, or inconsistent measurement.
Anti-Patterns
- Treating correlation as causation.
- Using aggregate summaries in place of raw post-level observations.
- Combining LinkedIn and X metrics that are named similarly but measured differently.
- Hiding nulls, missing periods, small cohorts, outliers, or failed comparisons.
- Turning the analysis into copywriting, content approval, or publishing.
Output
Produce a reproducible analysis artifact with scope, source run IDs, coverage audit, methods, descriptive baselines, evidence-labeled patterns, counterexamples, limitations, and handoff findings. Report exact paths and validation checks, and state that optimization and publishing were not performed.