[alexander-schneider]

Why

Show how saved external research can document manually selected Black-Litterman
views without turning sentiment scores into expected returns or confidences.
Following the scope boundary described in #748, this adds no provider client.

What changed

• An offline documentation recipe using synthetic Adanos-shaped reports and
  bundled price data, with annual assumptions and explicit ticker alignment.
• Tests execute the actual RST and verify native model results, missing research,
  reordered confidences, and rejection of missing nullable confidences.

Scope

Only documentation and its tests change. There is no account requirement,
network request, SDK, dependency, automatic view calibration, or core change.
The fictional research dates do not assert actual provider coverage or
point-in-time availability. This is not an investable backtest.

Blast Radius

Existing model behavior stays unchanged. The example uses manually chosen
returns/confidences; changing sentiment alone cannot change the model result.
I work on Adanos, a commercial API with a limited free
plan. This contribution used AI assistance and includes no actual provider data.

Verification

• 87 focused/adjacent tests and 15 native doctests passed. Numerical outputs were
  checked independently; Ruff lint and format checks passed.
• Fresh Sphinx HTML built with the same five baseline diagnostics. The rendered
  block matches current source and executes with the expected results.
• Both independent autoreview (P0-P2) and ponytail review completed before
  submission. The nullable-confidence finding was fixed with a failing
  regression first, then both reviews were repeated on the complete change.

No full suite, live API, cross-version matrix, or upstream CI result is claimed.