Preflight

Catch the risky change before it ships.

A pull request scored against the graph: what it can break, who should review it, and which active decisions it contradicts.

Preflight: the pre-deploy gate

The same graph that explains incidents scores a PR before it ships: what the change reaches, which revenue path, which decisions it contradicts and who must review it. Weights come from weights.yaml; every term from a FalkorDB query.

Example PRs

Run an example PR or your own body. The verdict, its terms and the evidence appear here.

What the change reaches

Loading the tenant's failure-flow graph…

service call (solid grey)
data lineage (dashed blue)
service → data crossing (dotted violet)
change applied to (long-dash orange)
chain hop not in the failure-flow view
proof path, lit hop by hop (thick red)
rejected suspect's chain (dashed light grey)

CAUSEproven cause (red)
step non the proof path (amber)
REJECTEDsuspect rejected by the gate
cap noutgoing hop capacity below 1 (PROPAGATES_TO cap: the share of the flow that hop can carry)
SYMPTOMalerting target
node size grows with betweenness (√-scaled; algo.betweenness over CALLS, Q-CHOKE-ALL; data assets and changes have none and stay small)
Shapes: rounded box = service, square dashed box = data asset, pill = change.