• Primary count metric: count_ns_sld, delegated SLD-owner count inferred from NS records.
• Baseline mode: full tracked-universe sweep to establish complete coverage.
• Hybrid mode: core daily recounted + rolling deterministic long-tail refresh.
• Snapshot rows keep full visibility; observed rows track what was freshly recounted today.
• Raw zone files are never published. Public surfaces expose aggregate metrics and immutable run artifacts only.

• model_version: locked model contract identifier (for example, rootfetch_model_v1).
• methodology_version: methodology publication version.
• Normalization and percentile windows are explicitly versioned with the model contract.
• Weight/policy changes require a version bump and backward-compatible historical artifacts.

Tracked TLD universe list
   -> local baseline/hybrid recount
   -> daily_counts + growth_trends
   -> signals (market + trend + coverage)
   -> digest + static RAG
   -> committed artifacts
   -> read-only web runtime + MCP

p50/p90/p99 summarize spread. Tiny and small buckets show long-tail breadth.

Top1/Top3/Top10 shares and HHI quantify concentration. Higher HHI indicates stronger concentration.

• DVI_v1 = weighted normalized combination of dispersion, concentration shift, and anomaly clustering.
• Regime_v1 uses a hysteresis state machine with minimum-duration enforcement to prevent flapping.
• All model fields are versioned in artifacts (`model_version`, `methodology_version`).

• Component A: cross-sectional dispersion of delegation deltas.
• Component B: concentration shift pressure from delta-HHI and top-share movement.
• Component C: anomaly clustering intensity from outlier density.
• Output is bounded to 0-100 and banded for operational interpretation.

• States: Stable, Elevated, Consolidating, Fragmenting, Turbulent.
• Overrides use concentration direction and top-share drift.
• Hysteresis and minimum-duration rules reduce regime flapping.
• Each regime includes a confidence score derived from signal agreement.

• Inputs: committed snapshot aggregates + model contract version.
• Script: python compute_model_v1.py snapshot.json.
• Regression tests enforce output stability and transition correctness.
• Published runs are immutable and hash-verifiable via manifest SHA-256 records.

• Determinism: same snapshot inputs produce the same snapshot hash and run-scoped model outputs.
• Consistency: each page load anchors to one artifact run pointer (`data/artifacts/latest.json`).
• Immutability: `data/artifacts/runs/<run_id>/*` is append-only and hash-audited via `manifest.json`.
• Caching: run artifacts use 1-year immutable caching, while `latest` and replay index use short SWR windows.
• Alerts: at-least-once delivery with durable dedup, audit log, and dead-letter retention.