Skip to content

feat: failure clustering across many failing runs (gated on a corpus) #63

Description

@Melvin0070

Deferred — gated on data that does not exist.

The Phase-3 "cluster/consensus" line is two features, and only one is dead. Multi-reference
consensus (bad vs many good) was measured and killed in #45 / notebook 066. Failure clustering is
a different product
: many failing runs in, common failure modes out — triage across runs rather
than localization within a pair.

It was never attempted because it is gated on a corpus of failures that has never existed. It is
deferred for lack of data, not lack of merit, and it would need its own positioning work: it is
closer to what the observability incumbents already sell than the two-run wedge is.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions