Data Sources
Prioritize using:
- Platform backend screenshots and export files provided by the user.
- Read-only statistics interfaces of authenticated connectors or the user's own accounts.
- Content task cards, registries, and historical retrospectives.
- Public content links, used only for public metrics and structural observation.
Do not estimate data that the platform does not provide. Reading your own account data does not equal authorization to modify the account or publish content.
When continuous recording is needed, from metrics-ledger-template.md create a raw metrics ledger. If the project already has a database, spreadsheet, or analysis system, continue using the existing system and do not create a duplicate ledger.
Analysis Process
1. Validate data
Check platform, content, publish date, observation window, field definitions, missing values, and outliers. Distinguish impressions, reads or plays, engagement, follows, conversions, and production costs.
2. Give core conclusions
Explain how performance compares with its own baseline, what the most noteworthy signals are, and which conclusions cannot hold.
3. Do attribution
Check by strength of evidence:
- Topic selection and target audience.
- Title and cover.
- The first 3 seconds or first screen of the opening.
- Structure, evidence, and information density.
- Publish time, tags, collections, and calls to action.
- External factors such as trends, paid promotion, sponsored deals, and account size.
Correlation does not equal causation. When there is no control or insufficient sample, write "to be verified."
4. Do peer comparison
Compare only within the same platform, same content type, and similar time windows. Prioritize median, P75, new follows per thousand views, deep engagement rate, and production time. Raw play counts across platforms cannot be directly ranked.
For complete metric definitions, see metrics.md。
5. Form decisions and experiments
Classify conclusions as: double down, repackage, change homepage or series, platform re-adaptation, stop, insufficient sample.
Design only one main experimental variable at a time, and clearly write the hypothesis, change, success criteria, and observation window.
Retrospective Levels
- Single-post retrospective: use content-review-template.md。
- Weekly retrospective: use weekly-review-template.md。
- Monthly retrospective: use monthly-review-template.md。
Output
Deliverables:
- Core conclusions.
- Data quality and baseline explanation.
- Attribution and strength of evidence.
- Replicable factors and non-attributable factors.
- 3 to 5 actionable actions.
- Hypotheses to be verified and the single experiment.
Analyze sponsored deals and organic content separately. When the sample is insufficient, do not adjust the long-term content ratio.