M1 — Vendor → Wasabi-bucket → use-case flow

Eye-landing point

The six Wasabi buckets in the centre column. They are the corpus's gravity. Vendors flow into them from the left ; use cases extract from them on the right. The bucket sizes (1.43 TiB Databento, 3.75 TiB S&P en cours, 249 GiB RavenPack, 187 GiB CausalityLink, 47.8 GiB GDELT, 21 KiB test) carry the proportionality.

A reader scanning the diagram in three seconds should be able to answer : which buckets does Databento feed, and at what annual GB read? — answer : maqi-databento (1.43 TiB stored), feeding UC-01 (~1 500 GB/an) and UC-03 (~3 000 GB/an).

Layout discipline

• Left column : five vendors (Databento, S&P Global Xpressfeed, RavenPack, GDELT, CausalityLink). Names pulled from docs/providers/catalog.yaml.
• Centre column : six Wasabi buckets. Sizes from docs/wasabi/state.md (snapshot 2026-04-14).
• Right column : five use cases (UC-01 \(\to\) UC-05). Pulled from docs/compute/scenario-matrix.md §1.
• Left edges carry license type (paid commercial, academic license, academic dump, open/free) plus checksum availability.
• Right edges carry annual GB read per UC (rolled-up across the cohort, midpoint of the per-session range \(\times\) sessions/an).

What was deliberately excluded

• Pricing — there is no €/GB on this diagram. Pricing belongs to M2 (compute arbitrage) ; mixing it here would dilute the gravity point (the buckets).
• Latency / throughput — those are M3 territory (PoC dates) and scenario-matrix territory (architecture choice).
• Vendor sub-products — S&P Global is shown as one node, not decomposed into Compustat / Transcripts / GICRS / ESG / Panjiva. The decomposition lives in docs/providers/spgmi-*.md ; pulling it here would multiply the left column by 5 and break the eye flow.
• Anomalies — missing 2020 RavenPack, GDELT MD5 divergences, S3NS-SecNumCloud constraints, CausalityLink snapshot freeze. Captured in docs/wasabi/anomalies.md. Adding them here would turn a flow chart into a status dashboard.
• The seventh box — there is no maqi-internal or maqi-staging bucket. Six is the count, six is what the diagram shows.

Lint check (Obsidian-Mermaid pitfalls)

• [x] No <i> / <b> HTML tags inside labels (only <br/> for line breaks, which is supported).
• [x] No numbered prefixes 1. 2. at start of any label (UC-01 etc. use a hyphen, not a period).
• [x] No nested subgraph containing a single node — visual grouping uses classDef for vendor / bucket / uc instead.

Render check

• GitHub Mermaid preview : OK.
• HTML wiki build pipeline : exercised by the sibling wiki-build pipeline.
• Banlist sweep on rendered SVG : passes (no internal-only vocabulary in node IDs, labels, or edge labels).