Page-hits chart reimagine — A/B sketch · 2026-05-15

Surface: dashboard.dare.co.uk Page hits — by weekday/month chart · Window: last 90 days · Source: dashboard.html embedded JSON

TL;DR

• The current Weekday/Month bar chart aggregates 70 active days into a handful of totals — losing the bimodal shape that the underlying data has.
• 6 spike days (≥6K hits) account for 48,518 hits — 20% of the total volume.
• 64 baseline days sit at median 2,931 hits (mean 3,001) — the actual day-to-day reality.
• Hottest day on record: 2026-03-07 at 10,007 hits.

Sketch A — calendar heatmap (GitHub-contributions style)

90 daily cells. Rows = weekday (Mon top, Sun bottom). Columns = calendar weeks. Colour intensity ∝ hits (log-binned). Hover any cell for the date + value.

What it reveals: the March spike cluster jumps out as a column of dark cells; baseline days form the muted background. Weekday rhythm (if any) reads as horizontal bands.

What it loses: exact numbers are gone — you read pattern, not value. Wrong tool for someone asking ‘what was last Tuesday’s count?’; right tool for ‘is anything anomalous?’

Sketch C — cumulative-volume curve (concentration view, added late evening)

Active days sorted by hits descending. Curve shows cumulative % of total hits as you walk through the top days first. The dashed 45° line is what perfect equality would look like; the bow off the line is the concentration.

What it reveals: top 6 days (8.6% of active days) carry 20% of total hits — Pareto-like concentration. Top 10% of days carry 23%. The curve’s elbow is the natural spike-threshold; the bar in B uses a hardcoded 6K cutoff, but this curve derives the boundary from the data’s own shape.

What it loses: time. Both axes are about distribution, not when. A March-clustering vs April-evenness story is invisible here. Pair with heatmap or two-mode bar; this is a complement, not a substitute.

Sketch B — two-mode bar (spike vs baseline) — DAN’S PICK

Same months on x-axis as today’s chart, but each bar is split: red top segment = sum of spike days (≥6,000 hits), green bottom = baseline days. Hover for the spike-day dates.

What it reveals: how much of each month’s total is baseline vs event-driven. March’s 111K is ~37% spike, April’s 98K is ~7% spike. The story shifts from ‘March was big’ to ‘March had a 5-day burst on top of a normal floor.’

What it loses: the spike threshold (6K) is a categorical decision and obscures values between 4-6K. Less honest if the distribution were continuous; works because this data IS bimodal — the cumulative curve in C confirms it.

Recommendation

1. Two-mode bar = the upgrade to the existing chart (Dan’s pick). Same shape, restored bimodality. Threshold becomes config knob; tooltip ports the rich bar-tooltip pattern; click-through to per-day report is the future workstream.
2. Cumulative-volume curve = the small companion beside the metric strip. It shows how concentrated the distribution is in one curve (6 days carrying 20% of hits). The elbow of the curve naturally argues for the spike-threshold in B — they’re two views of the same boundary.
3. Heatmap = the future investigation surface — when we add the per-day deep-link, each cell becomes a portal. Less urgent than B+C; park as v2.
4. Together the three views answer different questions: when (heatmap), how much (two-mode bar), how concentrated (cumulative curve). Each earns its slot by answering something the others can’t.

Sketch built one-shot from live dashboard data; production would re-read on each cron refresh + cache-bust on data change.

Generated: python3 ~/Downloads/dare_ab_preview_pagehits_reimagine_build.py