Content analytics pipeline

Context

Writing publicly about technical work is only a feedback loop if the feedback is measured. Otherwise the record is a memory of which posts felt successful.

Constraint

The platform shows recent numbers and then effectively forgets them. Without a local time series there is no way to compare a format against another, or to see the difference between a post that arrived quickly and one that kept accumulating attention for a week.

Contribution

I built a pipeline that collects post analytics on a schedule, tracks profile metrics hourly into a local database, exports to the formats an analysis actually uses, and visualizes trends, audience composition, and growth and decay.

Approach

The collector drives an already-authenticated browser session over the debugging protocol rather than handling credentials: the login happens once, by hand, and the automation never holds them. Keeping the history locally is what makes the decay curve visible at all.

Outcome

Content decisions can be argued from the series rather than from impressions, including the uncomfortable cases where the memory and the data disagree.

Evidence

The work is held in a private repository containing the collectors, the local database, the export paths, and the analysis notebook.