Methodology changelog

Initial: attention_density + float/liquidity normalizations, velocity/acceleration, percentile ranks, source quality, pump risk, social pressure index, divergence, narrative velocity, social surprise, attention candles, lifecycle state, market share/rotation, half-life, beta.

Additive stratified_hit_rate receipt detail: base_rate, edge_over_base_rate, and Wilson edge bounds versus the empirical in-sample hit-rate.

Additive aggregate_realness_verdict confidence detail: cross_source_corroboration now caps single-source verdict confidence and lifts confidence when two or more material sources corroborate the same attention row.

Additive cross_source_corroboration receipt detail: material source counts now exclude cross-source syndicated copies that share a content fingerprint and publish the content_fingerprint_dedup receipt.

Additive source_stale_withholding receipt: stale source families are excluded from cross-source corroboration while fresh source families can still score.

Additive pump_risk_evidence detail: account-level identical-text coordination now publishes a hash-only receipt and feeds the spam_cluster input floor.

Weighted_social_activity now applies config/sources.yaml base_weight as the source_weight factor for each source family.

Additive source_config_filter receipt detail: sources.yaml enabled=false families are excluded from ranked WSA and cross-source corroboration with a visible source_disabled receipt.

Additive cross_source_corroboration materiality detail: a source must now clear both the material share and absolute post/WSA mass floors before it can count as independent corroboration; thin_source receipts disclose sources below the floor.

Additive cross_source_corroboration configured-source detail: receipts now publish configured_source_count/configured_sources, structurally_single_source, and no_corroboration_reason so a one-source configuration is visible rather than absent.

Additive author_graph_evidence topology detail: receipts now publish suspicious edge component density, max-core, dense_subgraph_score, and the aggregate realness verdict exposes author_graph_dense_subgraph_score as an author-graph veto input.

Additive author_graph_evidence persistence detail: receipts now publish early-vs-late ring_persistence, supported-pair counts, and pair Jaccard so repeated rings differ from one-off clusters.

Additive author_graph_evidence serial-coordination detail: receipts now publish a privacy-safe ring cluster fingerprint, serial_coordination flag, and co-targeted tickers so the same opaque author ring rotating across tickers is visible.

Additive organic_score tell: staggered_author_cadence detects low-variance drip posting across a supported multi-author set so burst evasion is measured without treating slow diurnal crowds as manufactured.

Additive organic_score tell: coord_share measures the share of posts inside hash-only identical or simhash-near-duplicate multi-author time-window clusters; the same account_coordination_bot_signal receipt can feed the spam_cluster floor.

Additive attention_tiers anti-Sybil detail: profile-aware tiering now caps thin or stale lucky credibility below proven and discounts proven mass by independent_voices.independence_ratio when a tight cluster drives the spike.

Additive organic_coordination_baseline maturity gate: under-mature same-hour baselines stay in building_own_baseline with coordination_baseline_under_mature instead of declaring abnormal own-norm reads.

Additive organic_coordination_baseline universe-norm detail: receipts now carry a hierarchical bucket/universe population comparator and abnormal_vs_universe so perennially coordinated names cannot normalize themselves away.

Additive seasonal_baseline_evidence borrowed-parent detail: phase/ticker-parent borrowed seasonal baselines now publish borrowed_from_parent, parent basis, and borrowed parent n so cold-start abnormality reads are measured but visibly borrowed.

Additive seasonal_baseline_evidence structural-break detail: significant ticker-level baseline shifts now publish pre_break_excluded, excluded/post-break observation counts, and a structural break receipt so stale seasonal history is visibly excluded from abnormality reads.

Additive served abnormality framing: ad.abnormality Signal Fields now publish a served-abnormality-framing receipt with absolute_level, seasonal_z read, and dominant_basis so high absolute attention is not marketed as abnormal when the seasonal baseline reads normal.

Additive baseline_maturity_parity receipt: multi-window baseline reads now publish per-window minimum_n/current_n and flag any window marketed mature below the shared maturity floor.

Additive raw_density metric and receipt: raw source posts now publish per $1B of market cap with missing/invalid market-cap degradation visible.

Additive accepted_post_count: rows now surface the ranked WSA evidence included_post_count as the real-post denominator distinct from raw posts.

Additive raw_post_count_delta: rows now publish the current raw source-post count minus the prior same-window raw post count for that ticker.

Additive engagement_sum metric and receipt: score rows now publish observed public like/repost/reply/quote totals with partial/missing public-engagement coverage visible.

Additive engagement_weighted_density metric and receipt: observed public engagement now publishes per $1B of market cap with missing denominator and partial public-engagement coverage visible.

Additive saves_views_available boolean and receipt: optional bookmark_count/impression_count owned-post inputs can lift WSA only when observed, with missing availability visible.

Additive author_quality_receipt: live WSA author-quality coverage from social_authors.quality_score, with neutral_unrated degradation when the author graph is missing.

Additive credible_share metric and receipt: rows now publish the WSA-weighted share led by high author_credibility or disclosed known-mover authors, with missing author-credibility coverage visible.

Additive attention_impact_level metric and impact-level-v1 receipt: rows now publish a reproducible 0-4 level from credible_share, engagement-weighted density, and attention density, with Phase 4 scorecard gating and no opinion or sentiment inputs.

Additive claim_density metric and claim-density-v1 receipt: rows now publish the ratio of accepted posts from high-credibility or known-mover authors that carry deterministic non-promotional claim evidence, with missing author credibility coverage visible and no opinion or sentiment inputs.

Additive reason_descriptor text field and reason-descriptor-v1 receipt: rows now publish an observable co-mention/catalyst-window descriptor as text, not a score, with no opinion or sentiment inputs.

Additive engagement-shape measurements: rows now publish passive_amplification_ratio, engagement_discussion_imbalance, and engagement-shape-v1 receipts from observable public engagement counts only, with no opinion or sentiment inputs.