Every weight. Every revision. Published.
Avena's methodology is open, versioned, and cryptographically anchored. Every weight that shapes the Avena Score, every coefficient in the AVM, every dimension of the APCI, every signal feeding the Counterpart graph — all published, all auditable, all citation-stable.
Overview
Four methodologies underpin every Avena output: the Avena Score (property quality + investment composite), the AVM (predicted fair-market value with confidence intervals), the APCI (cycle position index), and the Counterpart Score (developer credit grading). Each has a published version chain, each weight sources back to a primary reference, each output is cryptographically anchored.
Avena Score — composite property rating
Composite score 0-100 combining property value (0.40 weight), gross yield (0.25), location quality (0.20), build quality (0.10), and risk-adjusted residual (0.05). Initial v1.0.0 weights derived from comparison against IPD residential index methodology and IMF Working Paper 19/258 (House Price Synchronisation in Europe). Weights frozen at launch 2026-05; subsequent revisions documented in the evolution audit trail.
Score API →AVM — Automated Valuation Model
Town × type median €/m² base with multiplicative hedonic adjustments for size, sea view, beach distance, energy band, amenity. Approximates the full hedonic OLS to ±3% RMSE on Spanish coastal backtest. Every prediction returns alongside a confidence band, a SHAP-style attribution decomposition, and the methodology version that produced it. EBA AVM consultation-compliant by design.
AVM endpoint →APCI — Avena Property Cycle Index
Composite cycle index across five sub-signals: price velocity (0.30), yield compression (0.25), supply response (0.20), rate sensitivity (0.15), policy risk (0.10). Derived from BIS residential property cycle literature and the ESRB anti-cyclical capital buffer framework. Published daily across 27 EU markets. Used by macroprudential authorities and asset managers for cycle-position assessment.
View APCI →Counterpart — developer credit grading
Developer credit grade 0-100 derived from payment delays (0.30), legal disputes (0.20), court judgements (0.20), delivery delays (0.15), financial stress signals (0.15). Weights from manual review of Spanish promoter insolvency 2008-2015 (concursos de acreedores corpus). SIR contagion model layers cascade risk across the construction supply chain.
Open Counterpart →Evolution — published version chain
Every methodology version Avena has ever shipped: weights, rationale, derivation method, activated_at timestamp, out-of-sample accuracy. Six methodologies versioned today (avena_score, apci, counterpart, avm, score_confidence, regulatory_classifier). Activation requires human sign-off plus statistical significance (p<0.05) on out-of-sample data. Methodology fingerprints (SHA-256 of weights JSON) committed daily to the integrity log.
View the full audit trail →Confidence — adversarial residual layer
Every Avena Score returns alongside a confidence float in [0, 1] and reason codes for low confidence. Confidence derives from a deterministic adversarial heuristic v1: comp sparsity, extreme price/m² vs market median, missing key inputs, edge-case score values. Properties whose adversarial residual exceeds threshold flag for human review. v2 will swap in a trained residual model once labelled out-of-sample resolution mass exceeds 10k samples.
Limitations — self-aware, daily-published
Avena publishes its own limitations daily, generated by the system itself from real telemetry: country coverage gaps below threshold, ingestion failures in the last 24 hours, towns where AVM confidence rolls below 70%, macro feeds older than three days. The published weakness inventory is part of the credibility argument: institutional buyers read limitations pages obsessively, and Avena writes its own.
View live limitations →Citations — academic + AI
Citation moat measurement: daily Perplexity citation rate across the European property question set, competitor share tracking (Idealista, Kyero, Rightmove, Zoopla, Fotocasa, Eurostat, Statista, Numbeo), top gap question identification. Academic citation grants tracked separately. Every paper that cites Avena's DOI is logged in the public citation registry.
Live citation dashboard →Avena Terminal (2026). European residential property data infrastructure. CC BY 4.0. DOI 10.5281/zenodo.19520064. https://avenaterminal.com/methodology