# Provenance — fig1-investment-divergence

**Chart type:** two stacked **small-multiple rows** — one per funding stage, stage-ordered top→bottom (top = pre-seed/seed, the earliest funnel stage; bottom = Series A), **each row on its own zero-based per-row scale**.

**Why this, not that (redesigned 2026-07-05):** the earlier version was a single shared-£-axis slopegraph. On a shared axis, Series A (more money) sat *above* seed purely because it is a larger £ figure — which fought the funnel logic and confused the reader (why is the *later* stage on top?). The story this figure exists to tell is **direction per stage** — the seed layer rose, the Series A bridge fell — not a magnitude comparison between the two stages. Two stacked small-multiple rows put each stage in its own panel, stage-ordered (earlier stage on top), so each row shows its own trend cleanly and the two opposing directions read at a glance. A shared axis would have buried the seed row's near-doubling (it is small in absolute £) beneath the Series A row. The single ember accent is on the Series A row — the piece's worry.

**The per-row-scale trade (load-bearing honesty note):** small multiples normally share a scale. Here they deliberately do **not** — each panel is zero-based on its own range (seed 0–£200m; Series A 0–£1,200m) so each *direction* is legible and neither slope is exaggerated (both panels zero-based → lie factor ~1.0). The cost of per-row scaling is that it hides cross-stage magnitude — Series A is still far larger in absolute £ than seed. That magnitude is preserved by **labelling every plotted £ value directly** (~£100m, £193m, £1.1bn, £456m). The reader sees both the per-stage direction *and* the absolute figures; the axis choice never distorts either. The subtitle and footer state "each panel on its own scale" so the reader is told, not tricked.

**Series-label honesty (operator-requested simplification):** the on-figure lower-series label reads **"Series A (early-stage VC)"** — no "–B", no "bridge". The underlying source definition of this series is broader: per PitchBook / the UCI report, **"early-stage VC = Series A to Series B rounds raised within 5 years of foundation."** The full definition is recorded here so the shortened on-figure label stays honest. "Series B" and the "bridge" metaphor are dropped from the *series label* and appear only in the figure **title** ("The seed layer held. The Series A bridge fell.") and the essay — where a metaphor belongs, not on a data label.

**Honesty notes:**
- All values are constant 2024 prices (per Figure 27).
- Each series is plotted **only at years the source states a value for**: seed at 2019 and 2024; Series A at its 2021 peak and 2024. No per-year points are interpolated. The seed row does not plot 2021 and the Series A row does not plot a 2019 point because those values are not stated — plotting them would fabricate data.
- The seed 2019 figure (~£100m) is a report-stated **approximation** ("around £100m") — marked `approx` and shown with a `~` prefix. The +93% derivation ((193−100)/100) is honest given this is a stated approximation.
- The Series A row carries two **text annotations** in the right clear space — "more than halved" and "below pre-pandemic" — not plotted points, because the source states these relations qualitatively without plotted Series A values for those years. (The related "flat across 2023–24" detail is stated in the essay's Evidence note rather than on the figure.)
- Later-stage VC (~£700m → £2.2bn) and total VC (£1.4bn → £2.8bn) are omitted from the figure to keep the two-stage divergence legible; they remain available as essay context.

| id | series | row | period | value | source (publisher) | url | anchor | quote |
|---|---|---|---|---|---|---|---|---|
| seed-2019 | seed | 1 | 2019 | ~£100m (approx) | UCI Policy Evidence Unit, Fig 27 | https://www.repository.cam.ac.uk/items/4b112cc8-ae83-486b-a335-f24170fa0fd0 | sources.md:7 | around £100m |
| seed-2024 | seed | 1 | 2024 | £193m (primary) | UCI Policy Evidence Unit, Fig 27 | https://www.repository.cam.ac.uk/items/4b112cc8-ae83-486b-a335-f24170fa0fd0 | sources.md:7 | £193m |
| seriesA-2021 | seriesA | 2 | 2021 | £1.1bn | UCI Policy Evidence Unit, Fig 27 | https://www.repository.cam.ac.uk/items/4b112cc8-ae83-486b-a335-f24170fa0fd0 | sources.md:8 | peaked at £1.1bn in 2021 |
| seriesA-2024 | seriesA | 2 | 2024 | £456m (primary) | UCI Policy Evidence Unit, Fig 27 | https://www.repository.cam.ac.uk/items/4b112cc8-ae83-486b-a335-f24170fa0fd0 | sources.md:8 | £456m in 2024, now below 2019–2020 levels |

**Source:** UCI Policy Evidence Unit, "Powering Ideas to Innovation: The significance, structure and dynamics of the UK university spinout ecosystem" (Tomas Coates Ulrichsen & Joscelyn Miller, 10 June 2025), Figure 27 — analysis of the UCI spinout dataset (data behind HESA's Spinout Register + PitchBook). Repository: https://www.repository.cam.ac.uk/items/4b112cc8-ae83-486b-a335-f24170fa0fd0 · News page: https://www.ifm.eng.cam.ac.uk/research/uci-policy-unit/uci-news/spinout-register-analytical-report/

**Encoding decisions:** ink ground `#0A1428`; bone `#F4F1EA` for the seed (top, rising) row; exactly one ember `#E85D2F` on the Series A (bottom, declining) row + its faint 2024 range-frame rule; muted `#5A6678` for year captions, subtitle, secondary annotation, and footer. Inter for the title / row labels / prose annotations, JetBrains Mono for value labels and year captions. **Left-aligned row labels** ("Pre-seed / seed", "Series A") matching the fig2 formation-clock row-label convention. Consistent per-point label grammar: **value label above the point, year caption below the point**. Direct-labelled throughout — no legend. Both rows zero-based (no truncation); viewBox 936×604. The **x-axis is a shared time axis** across both rows (2019 · 2021 · 2024), so the Series A row's 2021 peak is placed to the right of the seed row's 2019 origin — showing that series begins later — and each row's direction annotation sits in the **right clear space**, consistent between the two rows.

**Kill-list overrides:** none. (Per-row-scaled small multiples are a keep-list pattern; the departure from the usual *shared*-scale rule is deliberate and disclosed above and on-figure.)

**Checklist (tufte/checklist.md, 12-item):** PASS. Verified by headless-Chromium render on the ink panel + read-back (squint test): the two opposing directions (seed up, Series A down) read in one glance; stage order correct; one ember accent; every £ value directly labelled; no label collisions (iterated through four render passes to clear them); zero-based non-exaggerating scales; no chartjunk; row labels left-aligned and consistent. role="img" + aria-label present and describes both panels.

**Generator:** hand-authored flat inline SVG (matches the shipped figure convention in this repo). Geometry computed to each row's value→pixel map (seed 0–200 over y 128–250; Series A 0–1200 over y 372–494). No separate generator script.