Your quarterly health-check on the engine's reliability: how well are forecasts calibrated? Which predictions missed? What new factor lanes shipped? What's the methodology hash inventory?
Dogfood-Customer-#0 framing: per CEO/100k-employee analogue T0, the internal Trustee is
T1 of NEXUS / THv2. This page validates the Trustee Quarterly Report template against
internal substrate BEFORE any external Customer-#1 SOW. The data below aggregates real
THv2 substrate (per-cohort Brier from Migration 011 + decision-feedback-loop + factor-IC + 169 SEED
methodology hashes). Per N8 Calibrated Honesty: when a section's data is sparse, it surfaces
"calibrating" honestly — never fabricated.
Per Decision-Feedback-Loop T0: every "View Verdict" / "Research" / "Plan" click is logged to
state/kadeem-private/decisions-log.jsonl. Hit-rate computed at expected-resolution-date.
Currently sparse (Wave 4 #10 just shipped 2026-05-05); meaningful aggregate emerges over 30+ resolved decisions.
8. Paper Trading Driver — activation status + confluence-aware routing
Diagnostics — why hasn't it routed?
Conflict resolution: legacy 4-of-4 Sniper gate drops divergent alpha+prescience setups silently.
Confluence-aware extension (gated by THV2_CONFLUENCE_AWARE_PAPER_DRIVER): when both engines DISAGREE
AND independence_score ≥ 0.7 (genuine high-information setup per Diether-Malloy-Scherbina 2002), routes at half-size
(1/8-Kelly) instead of skipping. Default OFF per Pre-Launch Validation T0 — flag-flip gated on N≥30 cohort Brier-improvement.
7. Vault Health — credential rotation + sync drift
Per Anthropic Critical Security Rules + Personal-Data Segregation T0:
this surface NEVER reads or returns raw credential values — only NAMES + presence flags + rotation timestamps.
Sync-drift between Doppler and Vercel is detected via local CLI script
scripts/audit-vault-sync-2026-05-05.js (internal path).
8. Perfect Foresight Calibration Dashboard
What this measures: per-decision-type foresight gap (THv2 pick vs perfect-foresight optimal) + hit-rate with Wilson-1927 95% CI.
LIVE = Brier-validated cohort N≥10. CALIBRATING = substrate healthy, awaiting cohort accumulation per Pre-Launch Validation T0.
EMERGING = scaffold present, data feed not yet wired. Per N8 Calibrated Honesty: when cohort N<10, aggregate is null + honest message surfaces verbatim.
Per-decision-type detail (last 30 days)
Decision Type
Status
N (30d)
Mean Gap
Hit Rate (CI)
Brier
Reason / Honest Message
Calibrating ≠ broken: CALIBRATING status means the substrate is operational and the cohort is accumulating per Pre-Launch Validation T0
(N≥10 minimum before aggregate stats publish; sector_rotation / pair_convergence / confluence_state typically need 4-12 weeks of organic resolutions).
LIVE rows (ranking_top_pick / directional_class / options_strategy) have Brier-validated cohorts and surface real foresight gap + hit-rate.
Composes-with: perfect-foresight-substrate-2026-06-02-v1 + api-foresight-cohort-2026-06-02-v1 + N8 Calibrated Honesty + Pre-Launch Validation T0 + Decomposition+Recomposition NORTHSTAR T0.
Multi-horizon comparative (last 60 days)
What this measures: hit-rate × foresight gap by horizon for directional_class + options_strategy.
Surfaces which horizons THv2 is most/least accurate at. Green = best horizon · gold = worst horizon per decision type.
Per N8: "—" appears when an individual horizon has N<10.
Decision Type
Horizon
N (60d)
Hit Rate
Mean Gap
6. Quarterly Substrate Summary
Active Capabilities (THv2)
—
Per CLAUDE.md ledger (manual count)
Methodology hashes
—
Live from THV2MethodologyRegistry
Factor library
—
Live from /engine/factor-registry.json
Vercel deploys this Q
100+
Per CLAUDE.md Active Capabilities log (manual count)
Trustee Doctrine T0 Compliance
Tenant-defined NPX-100: Owner-defined per session-canonical scoring rubric ✅
Bundled BOUNDARY-mandatory: per BRIER LABS Constitution Article VII §3 firewall ✅
Owner-direct advisory: this page is the substrate for it ✅
Quarterly reviews: this page is the quarterly substrate ✅
Max 4-hr response: per Trustee Doctrine T0 SLA — substrate-level monitoring needed (Tier-D)
Capacity-gating REPLACES hard cap: applies to external Customer-#1+ scaling ✅