The body is almost entirely dense reference tables (Source Overview, Capability→Tool, Symbol Format) and concrete routing rules, assuming Claude's competence (e.g. it never explains what OHLCV or a backtest is). The redundancy between 'Source priority' and 'Ban-Risk & Fallback Notes' is minor, so it stays at the lean/efficient anchor rather than dropping to 2.

It names exact tools (`get_market_data`, `cross_validate`), env keys (`TUSHARE_TOKEN`, `FRED_API_KEY`), symbol formats, and a concrete `pass {source: value, ...}` usage. As an instruction-only router it carries no code blocks, which the rubric's scoring note says is not penalized when guidance is this specific — so it is not capped at 2.

The Decision Tree gives a clear numbered sequence for the analysis scenario with an explicit validation checkpoint ('if [the env key] is missing, report the missing key rather than failing silently') plus feedback loops (fallback on throttle/timeout, error envelopes that 'never abort the surrounding batch'). Batch operations have explicit validation, so it is not capped at 2.

Sections are well organized and off-skill references (per-source skills, `financial_rigor`, test files) are one level deep, but the body is ~140 lines dominated by two large inline tables (20-row Source Overview, 19-row Capability table) that fit the 'content that should be separate is inline' anchor. It does not reach the 3 anchor's 'content appropriately split' because nothing is split into bundle files.

It lists multiple concrete actions — 'pick the best available source/tool', 'honour auth (env) requirements', and 'avoid ban-risk providers' — matching the 'lists multiple specific concrete actions' anchor. It is above the 2 anchor because the actions are specific and several, not merely a named domain.

It answers both what ('the single ROUTER ... to pick the best available source/tool, honour auth ... avoid ban-risk providers') and when with an explicit trigger ('Load this skill BEFORE any backtest, data-fetch, or research task'). This matches the 'clearly answers both what AND when with explicit triggers' anchor, so it is above the what-only 2 anchor.

It includes natural trigger terms a target user would say — 'backtest', 'data-fetch', 'research task', 'data need' — giving good coverage of the primary phrasings. It is not the 2 anchor because these are natural (not jargon-only) and several are present, not just one keyword.

The combination of backtest/data-fetch/research triggers with 'ban-risk providers' and 'auth (env) requirements' carves a clear data-sourcing niche unlikely to fire for unrelated skills. Although 'research task' and 'every data need' are broad, the trading/auth/ban-risk signals keep it above the overlap-prone 2 anchor.