bi-15 — Broader Ideas
Vertical tracing as a general method
The method — follow one operation through every layer, record where resolution is lost — applies to any stack:
- a request through your CDN, load balancer, gateway, service mesh, application, and database;
- a keystroke through your editor, language server, and compiler;
- a click through your analytics pipeline to the warehouse.
Most engineers know their own layer and one neighbour. The person who can trace an operation end to end is the person who can arbitrate between teams — which is the job at this level.
The deliverable is a cost model, not a diagram
The layer-skipping table is what you actually gain: for any change, which stages run and which are skipped. That is what makes a performance review conversation quantitative rather than a swap of folklore.
Build the equivalent for your own stack. "Which layers does this request touch, and which does a cache hit skip?" is the same table.
Recording where you ran out
Explicitly marking unresolved layers is a discipline worth exporting. A design document that says "and then the database handles it" is hiding the same gap. Teams that name their unknowns can schedule work against them; teams that paper over them discover them during an incident.
Next
bi-16 synthesises the traces into the capstone explanation and the eight recurring patterns.