[sajonaro]

A follow-up to T2 (#44). Diagnosis stopped listing chains there, but three commands still listed every chain the model allows just to count them.

What changes

These sites used to list every chain:

Where	Why it listed them
model validate --write-scenarios	It listed them twice: once for the message, once for the file's scenario_count.
model validate staleness check	It expanded the whole sidecar to read one hash.
model diff	It needed the scenario counts of both revisions.

Now:

• scenarios.CountChains counts with the DP that diagnosis uses. On a cycle it falls back to listing.
• scenarios.SiblingHash reads only the sidecar's hash. It keeps the full reader's error for an unknown format.

The listings that really walk chains are unchanged: simulate --from-scenarios, the MCP scenario listings, and incident scenario matching.

Speed (BenchmarkScale, new count case)

Tiers	Chains	Listing	Counting
6	273,511	3.2 s	18 ms
8	9,853,471	does not finish	38 ms
10	354,734,551	—	73 ms

Docs

• how-it-works.md still said diagnosis expands every chain, and advised meta.scenarios: none past seven tiers. It now states the current cost.
• The benchmark's comment table is re-measured. Its chain counts predated scenario IDs name their chain; no chain twice #43's duplicate removal (7,471 at four tiers, not 10,921).

Verified

• Tests: the 400-graph property test now also checks CountChains against Expand. A cycle test pins the fallback, and a SiblingHash test covers both formats, an absent file, and an unknown format.
• go test ./... and mkdocs build --strict pass.
• Downstream harness: 18/18. The storefront still writes 14,429 scenarios, its committed scenario_count.