Track record due diligence from returns alone

An allocator often gets a manager's monthly returns and nothing else. Without the code or the positions nobody can prove there was no look-ahead, and the report says that first. The series still shows the classic red flags.

monte-neo verify --demo-returns                     # honest, smoothed and too-good track records
monte-neo verify --returns manager.csv --n-strategies 4 --benchmark index.csv --out cert.json

The file needs a date column and a return column, in decimals or percent (detected). At least 24 observations are required. The MCP tool is verify_returns.

Check	Red flag
returns_only_scope	Always shown: what a return series can and cannot prove
return_smoothing	Lag-1 autocorrelation above 0.2 and significant: illiquid marks or stale prices. Reports the Sharpe after unsmoothing (Getmansky, Lo and Makarov)
zero_kink	Far fewer small losses than small gains (Bollen and Pool): losses deferred or rounded up
implausible_consistency	More positive months than the mean and volatility allow; a fail with a monthly Sharpe of 3 or more
round_numbers	Too many returns on round values for the data's precision
track_record_significance	Probabilistic and Deflated Sharpe with --n-strategies (the best of four funds is not one fund), and the minimum track record length
drawdown_consistency	A maximum drawdown too shallow for the volatility
benchmark_relation	Beta, correlation, alpha t-statistic and up and down capture against --benchmark
claim_consistency	With --claim: the claimed return, Sharpe or drawdown against the series

A certificate can be reproduced with monte-neo verify --returns manager.csv --recheck cert.json.

Many managers at once: monte-neo verify --returns-dir managers/ --n-strategies 3 --html managers.html writes one row per file, with REJECT and NEEDS_MORE_EVIDENCE first, and exits with 1 when any manager fails.

Honest scope: these are signs, not proof. Arbitrage and lending strategies can be smooth and honest. Read a warning as a question to ask the manager, not as a verdict on the manager.

Not investment advice: a research example, not a call to trade or a promise of profit. Simulated results do not predict future returns.