How a verdict is produced

Five evidence levels (L1–L5)

• L1 Verified — a source row exists and can be inspected (e.g. a DLD transaction or a developer disclosure).
• L2 Derived — calculated from verified or structured inputs (e.g. an area median from DLD transactions).
• L3 Estimated — produced by a model or fallback when a specific field is missing.
• L4 Inferred — reasoned from indirect signals (transaction velocity, supply pressure).
• L5 Demo / Sample — present for product demonstration only; not used for decisioning.

Three decision signals (kept separate, never collapsed)

• Timing signal: Strong buy / Buy / Hold / Wait / Avoid — the read on the current entry window given supply and recent comps.
• Decision Label: the same categories under stricter gates — must clear evidence, stress, and reliability gates before BUY is awarded.
• Stress Grade: A–F — how the project tolerates systematic stress scenarios (price, rate, occupancy, delivery slippage).

Decision Label thresholds (published and reproducible)

• STRONG_BUY when score ≥ 85
• BUY when score ≥ 75
• HOLD when score ≥ 60
• WAIT when score ≥ 45
• AVOID when score < 45

Hard guards (no project gets past these)

• stress < 50 → forced AVOID (regardless of other indicators).
• evidence < 45 → capped at HOLD (no BUY, no STRONG_BUY).
• developer reliability < 30 → score capped at 60 (limits upside drift).

What we ingest, and what we don't

What a verdict means — and doesn't