policy(house-policy): v2.0.0 proposal — thesis-book sleeve + sizing disambiguation (merge = sign-off)

[npbuilds]

PROPOSED amendment — merging this PR is the sign-off

House-policy is sole-amender governed (§2/§8): this PR is the written proposal; merge = Nirav's sign-off. Written rationale lives in the private vault (skill-lab/, 2026-08-10 proposal note); this repo carries the machinery change only.

Why

Orchestration rep 4's structural gate rejected every draft house-view overlay for the same root cause: v1.1.1 could not host a conviction position at workable size (tactical path ≈ 0.09% of book under the strict tier reading), so all drafts routed through core sleeves — which the gate correctly called out (§3: core = systematic). Secondary defects: the §4 tier table read two ways an order of magnitude apart, and the reference-probability rule zeroed any honest view without a point probability.

Changes

1. §3 — new thesis sleeve class + thesis-book sleeve: 15% share funded from core (dca-investor 55→45%, rebalancer 30→25%; both remain book-cap-bound so core behavior is unchanged); per-position ⅓ of sleeve → 5.0% effective cap; max 3 concurrent (count toward book max 8); entry requires a pre-filed thesis note (disputed market-implied expectation, ≥1 pre-registered falsifier, review date); no strategist runs it — entries are Nirav-signed only. Discretionary entries may never route through core sleeves for size.
2. §4 — sizing disambiguated: tier = fraction of the sleeve's §3 effective cap (HIGH 100% / MEDIUM 50% / LOW 0%), with a worked example (MEDIUM swing = 0.625% of book; MEDIUM thesis = 2.5%). Supersedes both ambiguous v1.1 readings.
3. §4 — reference-probability rule scoped by sleeve class: core N/A; tactical unchanged (no named reference → LOW); thesis MEDIUM available without a point probability when the note is complete — honest "no reference" now caps at 2.5% instead of zeroing.
4. §6/§7: Restricted also blocks new thesis entries and adds (falsifier exits stay mandatory); negative constraints extended.

Amendment-procedure compliance (§8)

• Direction: loosening + clarifying → MAJOR (1.1.1 → 2.0.0); changelog row added.
• Cooling-off: the sleeve takes effect 7 days after merge; the clarifications at merge.
• Preconditions verified live against the paper book: flat, 0.0% drawdown, no stop-outs — not Restricted/Halt, no stop-out within 30 days.
• Open parameter for the amender: 15% share / 3 slots as proposed vs. the conservative 10% / 2 slots variant (one-line change).

Verification

• Sync trinity: sync-registry ✓ · wire-all-domains --apply ✓ (0 modifications) · recalibrate_scores ✓ (idempotent on re-run)
• CI drift trio run locally as ci.yml defines them: recalibrate --dry-run "No score changes needed" ✓ · "Registry is in sync" ✓ · full diagnostic HIGH=0 MEDIUM=0 ✓
• Downstream consistency: archon/references/portfolio-rules.md is silent on sleeves (no contradiction); runtime rules block (15%/8/−5%) untouched; no file hardcodes the old core shares
• Leak-guard term scan on changed files: clean; all examples generic

🤖 Generated with Claude Code