Please deep think / ultrathink as this is a complex task.

plan-2d-backpressure-survey

Survey whether the planned work can be proven by deterministic backpressure — build failures, type errors, tests, lint, runtime/smoke checks, boot probes, architecture checks (dependency rules, ArchUnit, Roslyn analyzers, CodeQL), schema validators, data-check scripts — rather than by agent inference or human eyeballing.

This is the computational-control tier pulled forward to design time. It runs after the spec and before plan-3-v3-architect, so that missing backpressure is caught and planned for (as an optional "Phase 0") before code is written — not discovered late in plan-7 review when it is expensive.

It is the deterministic counterpart to plan-7-v2-code-review. plan-7 is the inferential / eyeball tier and stays exactly as-is; this skill surveys the computational tier. Both are legitimate (see How this differs below).

Grounding (harness-foundations):

• Rule 3 — prefer deterministic validation over agent inference: "The agent can say it is done. The harness should decide whether that claim is supported by evidence."
• Principle 16 — "Improving the harness means reducing friction and increasing deterministic back pressure."
• Principle 33 — "Verification must cover experienced failure modes" — startup, integration, rendering, hydration, side effects, architecture drift, security boundaries — not just the failures that are easy to unit-test.
• Pattern 18 — tier computational vs inferential controls; run computational controls early and often.

🟢 ADVISORY INVARIANT — read first, never violate