A bounded, canonical recipe for proving a confirmed incident still fails.
A regression embeds the exact source confirmation and scenario as data. Its
SHA-256 identity detects accidental or malicious rewriting when the expected
digest is trusted separately; it is not authenticity or proof of failure.
classify/2 requires two distinct fresh confirmation records and deliberately
cannot make a passing-after claim.
Summary
Functions
Classifies exactly two fresh confirmation records against a regression.
Types
Functions
@spec canonical_binary(term()) :: {:ok, binary()} | {:error, Insider.Protocol.Error.t()}
@spec classify(term(), term()) :: {:ok, map()} | {:error, Insider.Protocol.Error.t()}
Classifies exactly two fresh confirmation records against a regression.
The stable result is data, not a new evidence artifact. A failed result
means both fresh runs reproduced the bound failure. Every other valid run
combination is inconclusive; positive repair proof belongs to verification.
@spec decode(binary()) :: {:ok, t()} | {:error, Insider.Protocol.Error.t()}
@spec decode_map(term()) :: {:ok, t()} | {:error, Insider.Protocol.Error.t()}
@spec materialize(Insider.Sentinel.Confirmation.Record.t(), Insider.Scenario.t()) :: {:ok, t()} | {:error, Insider.Protocol.Error.t()}
@spec validate(term()) :: {:ok, t()} | {:error, Insider.Protocol.Error.t()}