feat(tracker-stats): project the current bucket to end-of-bucket totals

[potiuk]

The final bucket on every chart is cut short by "now", so regenerating the dashboard mid-month reads as a collapse in reports rather than a month that is only a quarter over. On the reference tracker today (8 September, 23 % elapsed) September shows 9 opened against August's 65.

Two kinds of series, two extrapolations

Kind	Series	Formula
rate — accumulates from zero inside the bucket	opened / rejected / reported in bucket	observed / elapsed
level — carries over from the previous bucket	cumulative opened / closed / rejected / reported, every lifecycle band, the untriaged backlog	prev + (observed - prev) / elapsed

Scaling a level series whole would multiply years of accumulated history by four, so only the movement inside the bucket is extrapolated. A falling level (a backlog being worked down) projects further down, floored at zero; a rate projection is never below what already happened.

Where it lands

Every chart that carries a projectable series: the lifecycle bands, opened-vs-untriaged, the cumulative lines, and the rejections chart. Each projection is a dotted two-point segment from the last complete bucket to the projected bucket end, so a forecast cannot be mistaken for a measurement. On the stacked lifecycle chart each band is drawn at its projected position in the stack, with the band's own projected count in the hover and legend entries suppressed. Because the bands partition every tracker, their projections sum to the projected cumulative opened — a free consistency check.

The mean-time charts are deliberately not projected. A mean over the items seen so far is already an estimate of the bucket's mean, not a partial accumulation; scaling it by elapsed time would be meaningless.

Config

projection:
  enabled: true
  min_elapsed_fraction: 0.1

min_elapsed_fraction suppresses the projection early in a bucket, where one report extrapolates to a dozen. The projection is also skipped on a single-bucket axis (no baseline for the level series) and when the bucket is already complete; stdout always says which.

Reference output

Current-bucket projection (2026-09, 23% elapsed, now -> month-end):
  opened                      9 -> 38
  reported                   12 -> 51
  cum_opened                355 -> 384
  band:open_untriaged        14 -> 7
  band:open_triaged          16 -> 39

Testing

• 25 new unit tests over the three new pure helpers (bucket_bounds, elapsed_fraction, project_bucket_total / project_bucket_level) — bounds in all three bucket modes, clamping, rounding, the never-below-observed floor, the zero-floor on falling levels.
• Full prek run green (ruff, mypy, pytest, markdownlint, doctoc); lychee offline clean.
• Rendered against a live 355-tracker cache in monthly, quarterly and weekly modes, and all four skip paths exercised (disabled, below threshold, single-bucket axis, complete bucket).
• Emitted JS executed under Node against a stubbed Plotly to confirm all 13 projection traces parse and land on the right charts.

🤖 Generated with Claude Code

https://claude.ai/code/session_01WJopDiCvbwSG6t4k15JE4H