feat(code-review): epic — review panel cost optimization (lens tiering + deterministic offload)

[bdfinst]

Umbrella for implementing the recommendations in docs/spikes/review-cost-optimization.md (merged in #1972): reduce per-change review cost by (a) tiering lenses into always-on risk gates vs. trigger-scoped value lenses and (b) moving mechanical agent checklists into deterministic static-analysis lanes.

Baseline (from the spike)

• A code-touching diff dispatches up to 14 Scope: always lenses; 4 on opus (security-review, correctness-review, arch-review, domain-review); every agent runs effort: high.
• domain-review is the only ungated always-on opus lens.
• No complexity or duplication tool is integrated in the static-analysis pre-pass, while two always-on sonnet lenses re-derive those numbers by inference every run.
• The existing gates (doc-only short-circuit, change-shape feat: gate low-yield review lenses (performance, correctness) by change shape in /code-review #1254, change-size fast path chore: pre-commit review gate has no size/complexity awareness, forcing a full 17-agent panel on trivial commits #1339, arch-impact gate, verification mode feat(code-review): verification-mode re-dispatch — narrowed context now, declared cheaper tier as opt-in #1628, deterministic-first triage chore(code-review): scope re-verification agents to what actually changed, not a full panel re-run #1610, repo_invariants.py chore(code-review): deterministic pre-pass for rediscoverable findings #1608) are the mechanism categories every slice below extends — no new mechanism categories.

Priorities

• p1 — additive or config-only; needs no new evidence to land safely.
• p2 — extends the deterministic lanes further, or produces the measurement that later slices depend on.
• p3 — acts on the p2 measurements; each carries its own evidence gate (/harness-audit output, /agent-eval before/after).

Sub-issues are attached below in priority order. PRs addressing a slice carry Closes #<sub-issue> and Part of #<this epic>; epic-auto-close (#987) closes this epic when the last sub-issue closes.

Constraint carried from the spike

Every roster-narrowing or tier-down decision cites measured data (review-value.jsonl #1624, /harness-audit criteria, /agent-eval fixtures) — the same evidence-first discipline skills/code-review/SKILL.md already requires for widening GATED_LENSES. The p1 slices are exempt only because they are additive (pre-pass lanes never gate) or move a lens to where its own charter already says it belongs.

[github-actions]

⚠️ This issue title is not a Conventional Commit.

Specs become epics and plan slices become sub-issues, and these titles seed branch names, PR titles, and release versions — so they must follow the same commitlint ruleset as commits (for example: feat:, fix:, docs:, chore:).

�[90m⧗�[39m   --- input ---
�[1mEpic: review panel cost optimization — critical-vs-optional lens tiering and deterministic offload�[22m
�[31m✖�[39m   type must be lower-case �[90m[type-case]�[39m
�[31m✖�[39m   type must be one of [build, chore, ci, docs, feat, fix, perf, refactor, revert, style, test] �[90m[type-enum]�[39m

�[1m�[31m✖�[39m   found 2 problems, 0 warnings�[22m
ⓘ   Get help: https://github.com/conventional-changelog/commitlint/#what-is-commitlint

Edit the title to fix; this check re-runs on every edit and clears the label once it passes.