jev-stack

TypeSafe Jev five-layer decision stack — status & process notes.

What this is

Decision schemas and shadow-consult services for the Jev layer stack, starting with Layer 1 (injection detection). This page is informational only — it does not run, deploy, or influence anything live.

Layer status

Layer 1 — Injection merged, shadow-only

Deterministic evaluate(text) and consult-only consult(text) entrypoints. Schema: contains_instructions, claims_authority, confidence, recommended action pass | flag | hold. Kill switch JEV_INJECTION_ENABLED=false by default — while off, effective_action is always null.

Layers 2–5 not started

Planned, no implementation yet.

Process & locks

CoS Jev routing

How a deliverable gets assigned and closed out, so writer lanes don't collide:

This page is one such single-lane deliverable: writer_claude_box was named up front (case B), not a race entrant.

Writer lanes

Work on this stack is paced across lanes to avoid multi-writer races on the same deliverable:

This page was produced on the writer_claude_box lane as the sole writer for this deliverable. That lane sits outside the four core buckets above — it runs on a shared computer ("Grok Bot") under a personal Claude Pro account. The grimstarr Claude instance is kept separate and stays personal-only; it is not part of this writer roster.

Separate seats

Adjacent decision-making surfaces that this repo does not own or duplicate — each is tracked in its own place, not here:

More