Most environment variables are undocumented

[stef9github]

.env.example documents 22 variables. src/ reads 137, of which 17 are in that file — so 120 are not (the remaining 5 in the file are consumed by Auth.js and the AWS SDK rather than read directly).

Reproduce:

git grep -hoE 'process\.env\.[A-Z][A-Z0-9_]+' -- 'src/*' | sed 's/.*env\.//' | sort -u | wc -l
grep -oE '^#? *[A-Z][A-Z0-9_]+=' .env.example | tr -d '#= ' | sort -u | wc -l

The undocumented ones are mostly service URLs, Kafka topic names and ConfigMap keys that have working defaults derived from VSS_NAMESPACE — which is why nothing appears broken without them, and why the gap went unnoticed. src/lib/cluster-refs.ts resolves most of them in one place.

This matters more now the repository is public: README.md points at .env.example as the configuration reference, and someone deploying against a differently-named cluster has no list of what they can override.

A patch documenting one coherent group (Kafka topics, VST endpoints, alert-bridge, GCS/Firestore) is a good first contribution — no need to do all 120 at once.

[stef9github]

Partial: 64cba9e documents the object-store and observability families — 21 more variables, taking the undocumented count from 120 to 98. Every default stated was read out of the code rather than inferred.

Remaining groups, roughly in descending value for someone deploying against their own cluster: VST/VIOS endpoints and ConfigMap names (15), Kafka topics (7), the LLM/agent and NIM tuning set (7), GCS/Firestore config store (5), the recording auto-heal and reconcile knobs (~12), RTVI-CV (6), TTS (4), test-footage (3), alert-bridge (3), evidence (2), KV-cache (3).

The measurement is reproducible:

git grep -hoE 'process\.env\.[A-Z][A-Z0-9_]+' -- 'src/*' | sed 's/.*env\.//' | sort -u > /tmp/used
grep -oE '^#? *[A-Z][A-Z0-9_]+=' .env.example | tr -d '#= ' | sort -u > /tmp/doc
comm -23 /tmp/used /tmp/doc | wc -l