Aether Score Changelog
Every version, written down
The Aether Score is versioned. Every methodology change is recorded here with its date, exactly what changed, and whether historical scores were recalculated. History is never silently recalculated — when a change would alter a published score, both versions are kept and it is stated plainly.
Data-source change · Removed Yahoo FinanceCurrent
2026-10-08Removed Yahoo Finance from every function, feed and page. Prices and daily price history now come from FMP; news from Finnhub (Marketaux fallback). No Aether Score formula or layer weights changed.
What changed
- ●Daily price history is fetched once per ticker per day from FMP end-of-day prices and stored in our own database. Every chart, the Technical layer, the 50/200-day moving averages, RSI, 90-day volatility, beta and the 52-week range read from that stored history — never per page view.
- ●Quotes are 15-minute delayed, from FMP, cached and shared across all users. News is Finnhub (15-minute cache) with Marketaux as the fallback. The original publisher name is kept on each article.
- ●Intraday chart ranges (1D, 1W) are hidden rather than show broken data, since the stored history is daily.
- ●All 'real-time' / 'live prices' wording about our quotes was replaced with '15-minute delayed' across the site, emails and meta tags.
Technical layer now reads from stored FMP daily closes.
Only the Technical layer's data source changed (Yahoo v8 chart → stored FMP daily history). Aether Score layer weights, sub-score thresholds and guardrails are unchanged; scores were re-derived from the FMP history.
Aether Score v1.0 · EPS and calendar input correction
2026-10-07Input revision v1.0-inputs-3 separates quarterly from TTM EPS growth and connects the symbol-specific earnings calendar. Published layer weights and threshold bands are unchanged.
What changed
- ●TTM diluted EPS growth uses the latest four quarterly EPS observations against the preceding four. No quarterly growth value is substituted for TTM. Missing comparable history, non-positive prior EPS, or absolute TTM growth above 100% excludes EPS from Growth and its acceleration bonus; the panel preserves the raw figure and explicitly names the exclusion.
- ●Alphabet's reported Q2 2026 EPS of $9.11 versus $2.31 is 294.37% quarterly growth using rounded reported EPS. The provider's 294% is not TTM growth. Alphabet reported $98.0bn in other income, primarily unrealized equity gains. The panel links the primary filing and does not claim GAAP EPS is normalized.
- ●Each exact ticker's Finnhub earnings calendar supplies the earliest future date within 180 days. Broad calendar truncation cannot discard the near event. The Macro/Catalyst layer now uses this sourced date; absent dates remain explicitly unavailable.
- ●All active daily publications are input-migrated with other stored raw inputs preserved; previous publications are retained. Monthly research membership is not re-screened. Both homepage teasers use the same profile and passed-screen-first order returned to Stock Picks.
- ●Market badges open the shared breakdown independently of card navigation. Raw inputs render percentages to two decimals, multiples to one decimal and large dollar amounts in bn/tn.
Active daily scores input-corrected; previous publications retained.
The explicit correction preserves superseded publications with their prior working and timestamps. Older dates and monthly research membership are not recalculated.
Aether Score v1.0 · picks re-screen
2026-10-07 · saved 12:01–12:03 Europe/ZurichRe-ran all three existing research profiles through the v1.0 four-layer screen. Membership, funnel status and ranking now use v1.0, and every surface reads the same daily score and timestamp.
What changed
- ●Re-screened Growth / High growth / Active trading, Balanced / Long-term, and the default Balanced profile from their complete 24-stock candidate universes, rather than substituting new numbers into the old membership list.
- ●Canonical daily values after the re-run: NVDA 69, GOOGL 67, AMZN 68, AVGO 66. AVGO's superseded pre-v1.0 score was 75; its v1.0 working is Business Quality 90 × 0.35 + Valuation 49 × 0.30 + Technical/Timing 61 × 0.20 + Macro/Catalyst 48 × 0.15 = 65.60, rounded to 66.
- ●Invalidated unversioned selection batches and browser caches. New batches carry the selection model and derivation revision, and scores are hydrated from the canonical daily publication on subsequent reads.
- ●Balanced screens: UNH, CAT and MSFT entered; JPM, FCX and LLY left. Growth / High growth: BLK, AA and CAT entered; PLTR, AVGO and HOOD left. True funnel survivors and scored backfill names are now separated explicitly, each highest-score first.
- ●Fixed a previously swallowed oversized-batch save error by keeping the full daily working only in the canonical score publication and omitting duplicated derived fields from the monthly cache. All three re-screens were then saved successfully.
- ●Unified score and information-icon clicks into the same breakdown panel. Fixed nested click controls, retained named sub-scores, exposed raw inputs, and made cap and rounding adjustments explicit.
- ●Corrected threshold sensitivities: the affected layer is now reported correctly, threshold scenarios cross actual model bands, and five quantified conditions are stored for NVDA, GOOGL and AVGO. No language model produces any score or sensitivity.
- ●The numerical scoring formula and published layer weights are unchanged. Technical beta is shown as its actual multiplicative adjustment, not as an unused pseudo sub-score.
Active scores and picks re-screened; previous publications retained.
Only the active 7 October publications were migrated. Every replaced daily publication was archived with its exact prior score, inputs, model version and timestamp, and all three prior picks batches were archived before replacement. Daily scores from dates before 7 October were not recalculated.
Aether Score v1.0
2026-10-07The Aether Score became a published four-layer model with explicit weights, a derivation panel, enforced guardrails, a deterministic "what would move this" list, and a per-score model-version stamp.
What changed
- ●Restructured the core from a three-layer fundamentals-only blend (Valuation 35% / Growth 30% / Financial Health 35%) into four published layers: Business Quality 35% (mean of Growth + Financial Health), Valuation 30%, Technical / Timing 20%, Macro / Catalyst 15%.
- ●Published the layer weights and the per-sub-score threshold tables for the first time. The same formula is documented on /methodology and shown verbatim in the new derivation panel that opens from any Aether Score.
- ●Technical / Timing, previously informational only, is now a scored layer — held to a guardrail: if it scores above 75 while Valuation scores below 50, the total is capped and the cap is shown.
- ●Added a deterministic red-flag set (negative operating cash flow with losses; net debt / EBITDA above 10×) that caps the stance at Weak regardless of score.
- ●Added a deterministic "what would move this" list — three to five quantified conditions, each computed by re-running the model at a stated threshold. Never written by a language model.
- ●Stamped every stored daily Aether Score row with model_version and the full layer/guardrail breakdown so the derivation renders the stored working, not a recompute.
- ●Stance thresholds unchanged: below 40 Weak, 40–59 Mixed, 60 and above Strong.
Historical scores not recalculated.
Daily rows from dates before 7 October were not recalculated. Active pre-v1.0 rows from 7 October were explicitly migrated for the re-screen recorded above; superseded publications from that operation are retained separately with their original model and timestamp.
pre-v1.0 (historical)
prior to 2026-10-07The original three-layer fundamentals-only Aether Score: Valuation 35% + Growth 30% + Financial Health 35%, with momentum shown as informational only.
What changed
- ●Core = 0.35 × Valuation + 0.30 × Growth + 0.35 × Financial Health.
- ●Momentum scored but excluded from the core (displayed as "Not part of the score").
- ●No published layer weights beyond the three-node blend; no derivation panel; no model-version stamp on stored rows.
Historical scores not recalculated.
Frozen. These rows are kept as the historical record for charts and the picks-removed log. They are never recomputed under v1.0.