Evidence-backed incident reconstruction
Find the code that broke production, from the logs you already have.
At 2am you get a wall of log lines and a service map. Neither tells you which function failed, what it took down with it, or which questions the telemetry cannot answer.
IncidentLens reads a Python service's source and its log files, reconstructs the failure down to the module it was logged in and a static candidate method inside it, and renders that as a narrated replay where every claim cites a log line. Each conclusion is labelled confirmed inferred or unknown — and it refuses to name a cause it cannot evidence. No agents, no instrumentation: a log file is the whole integration surface.
Press play to replay the incident.
No analysis yet
Pick a scenario above and reconstruct it. The graph replays how the failure started, spread and reached customers — every step tied to the evidence that supports it, and every gap the telemetry could not close named outright.
Or scroll down and run it on telemetry of your own.
Pre-rendered · Backblaze B2
The narrated films live in object storage
The replay above is drawn live in your browser from the analysis — it is not a recording. The narrated 3‑D films are different: they zoom from the service map down to the failing method, and each one streams straight from a public Backblaze B2 bucket with a Genblaze manifest embedded in the MP4 that you can verify.
Open the incident libraryTry it on an incident of your own
The bundled scenarios are ours. These two inputs are yours — the engine treats them identically, and reaches its own conclusions either way.