Corpus over time
loading liveHue names the subject, the stroke pattern names the actor: solid for the whole corpus, dotted for teknium1, dash‑dot for GottZ. L reads on the left axis (nodes), R on the right axis (comments and reviews). One candle per UTC day, one lane per series, each lane on its own scale. Body from open to close, wick from low to high; filled when close ≥ open, hollow when close < open. stock lanes track a level through the day (open = end of the first hour, close = end of the day, wick = lowest and highest the level got): a short body means the stock barely moved, not that little happened. flow lanes track the rhythm of the 24 hourly counts (open = hour 00, close = hour 23, wick = quietest and busiest hour): the body is a comparison of two single hours, not a daily total. Lane axes fit the visible range and do not start at zero — without that, every stock candle would collapse into a hairline. Today’s candle is dashed and marked now: it is a part‑day.
Loading the timeline…
The visible series as a table
What the corpus does not show
read this before reading the curves- Issues start late The issue series begins on 2026-02-10. That is the harvest boundary, not the beginning of the repository — every earlier issue is simply absent from the corpus. The gold rule in the diagram marks the day.
- Reviews start later than pull requests Review timestamps begin on 2025-09-12, roughly 48 days after the first pull requests and comments. The review series is therefore structurally zero for those first days — that is the harvest, not a defect.
-
Closing dates were backfilled on 2026-07-30
closed_athad never been written by the harvester; 48,280 closing timestamps were fetched from the GitHub API on 2026-07-30. PRs closed and issues closed therefore carry real data from the whole period, but they were reconstructed after the fact, not observed as they happened. Since the closed total now exceeds the merged total, the additive reading takesclosed_atas the pull request outflow. Before the backfill it tookmerged_at. Only one of the two is subtracted: a merged pull request is closed on GitHub too, so counting both would count the same departure twice. - The early months are nearly empty Measured on the live corpus: from 2025-07 to 2026-01 there are 14 pull requests in total; the mass begins in 2026-02 and after. The long flat run on the left of the diagram is the corpus, not a rendering fault.
- Actor series are subsets teknium1 and GottZ series are contained in the corpus series of the same hue. In the subtractive reading they are drawn as lines over the bars and never stacked into them — stacking a subset onto its superset would double count.