Straw-man for discussion - built from the current dataset

Data Donut - a layered concept

One idea for resolving the timing-vs-dependencies tension: a clean cyclical overview (the "donut" — what happens across the year, at a glance) sitting on top of a timeline drill-down (the real multi-year detail, with a dependency chain you can trace). Same data, two zoom levels. React freely — this exists to be argued with.

What is on each tab

Timeline detail — every milestone on a real calendar, one lane per system. Track width, month grid, dependency trace and the draggable current-date line live here.
Cyclical overview — the same records wrapped into a circle, to show what recurs rather than when exactly.
Data table — every underlying record in one list, for checking the data behind both visuals.

The time window

The window control on the Timeline and Cyclical tabs picks how much of the calendar both views show: 1, 2 or 3 years. The two views always show the same period — changing it on one changes it on the other.
At 2 or 3 years the ring is partitioned, with a divider and a year label at each January.
A system with records outside the chosen window is labelled in its lane, so a narrow window never looks like missing data.

Before you read the dates

Positions are month-precise. Where several events share a month they are spread evenly across that month's cell to keep them apart — that spacing is ordering, not a day of the month.
Systems with no records show as an explicitly empty lane or ring rather than being dropped.
The current-date line is the only day-precise element on the page.

Placeholder — to be completed before this is shared.

Data current as of: TBC
Source of the milestone dates: TBC
Owner / who to contact about a wrong date: TBC
Update cadence: TBC
Where this is hosted: TBC

Timeline detail — the calendar view. One lane per system across the selected window.

Window sets the period, and the Cyclical tab follows it. Narrowing the window is what relieves crowding: at one year each month cell gets three times the width it has at three.
Month grid switches itself off when a month cell is too narrow to read as a guide; the hint next to it says how narrow.
Small dots with no label are crowded milestones — hover for the name, or use Show all labels.
The date handle above the lanes is draggable. Focus it and use ← → for a day, shift + ← → for a month, Home to return to today. It turns amber once moved, and a reload always restores today.

Cyclical overview — the recurring rhythm. The selected window wrapped into a circle, read clockwise from January.

Rings run outer to inner in the same order as the legend above; hover any marker for its system, name and month.
Each slot is one month of one year — 12 slots at a 1-year window, 24 at 2 years, 36 at 3. Nothing is folded, so no slot is ambiguous.
A ring with no markers says whether the system is awaiting data entirely, or simply has nothing inside this window.

Data table — every record behind both visuals. Built from the same array the timeline and ring read, so it cannot drift out of step with them.

Filter by system, by a month range, by frequency, or by searching the milestone text. Filters combine, and the count below the table reports what is showing.
Month is the milestone's real position. Where several events share a month the visuals spread them across the month's cell to keep them apart — that spacing is ordering, not a day of the month.
Frequency, Reviewed by and Reviewed date are empty columns awaiting data. Nothing is inferred into any of them; the frequency filter stays disabled until records carry values.
Use this view to check the dataset. Anything wrong here is wrong in both visuals.