v1.2.0

Added

• Enhanced KPI long/short conjunctive-gate screener — Phase 2a (#192). src/orchestrators/enhanced_kpi_screener_longshort.py ships 13 of 16 issue criteria — every gate that reads an existing FundamentalsSnapshot field plus criterion 14 via the new analyst_recommendation field (alias recommendationKey, already in the yfinance info payload — no new HTTP). Paired output (enhanced-kpi-screener-longs + …-shorts); a ticker lands in longs iff it passes ALL 14 long-side gates, in shorts iff it passes ALL 13 inverted short-side gates, otherwise neither. Long ∩ short empty by construction. Declarative _NUMERIC_GATES table keeps _evaluate cognitive complexity at 5. PR #228. Phase 2b owes criterion 12 (FCF margin, Ticker.cashflow fetch), criterion 15 (tech rating, blocked on #21), ADR amendments, and threshold tuning.

• Demo: 1y | 5y | 10y | all time-window chips above the long-term F&G + 5s10s charts (#206). Click filters the entries array client-side before passing to Chart.js (per the issue's option 1 — no zoom plugin, no brush slider). New URL params ?ltFgWindow= / ?ycWindow= (omitted at default). Existing lazy-render guard preserved: chip click while a pane is hidden flips the rendered flag back so the next tab activation picks up the new window. PR #227.

• Demo: score-cell tooltip surfaces mean_composite on aggregator universes (#218). On aggregated-scores-best / …-worst, hovering the Score cell shows the mean-of-7-composites — the metric the aggregator actually ranks on. Closes the user-confusion gap surfaced by #202 (PRSO in worst-25 with visible screener_score 64, mean 16.5). PR #231.

• FCF margin enrichment for the long/short screener (#192 Phase 2b). src/data_sources/fundamentals.py gains _fetch_fcf_margin mirroring _fetch_rd_to_revenue: gated on info["quoteType"] == "EQUITY", reads "Free Cash Flow" from Ticker.cashflow and "Total Revenue" from Ticker.income_stmt via the shared _find_row helper, returns None on any missing data / zero revenue / network error. New fcf_margin: float | None field on FundamentalsSnapshot attached post-validate via model_copy. src/orchestrators/enhanced_kpi_screener_longshort.py adds the gate to _NUMERIC_GATES (long > 0.10, short < 0), bringing the conjunctive-gate screener to 15 of 16 criteria. Criterion 15 (tech rating) stays deferred behind #21. ADR-0002 amended to note the new enrichment field. Closes #192.

• Radar-hexagon favicon for the demo dashboard (docs/demo/favicon.svg). A dependency-free inline SVG echoing the composite-score radar chart, theme-aware via prefers-color-scheme so it stays legible on light and dark browser tabs. Linked from index.html as type="image/svg+xml".

• Self-hosted brand typography for the demo dashboard — Inter (UI/prose)
  and JetBrains Mono (numeric table cells + <code>), per the qte77
  EyeRest design tokens (#295). Latin TTF subsets ship in ui/fonts/ (SIL OFL 1.1,
  ui/fonts/OFL.txt) with font-display: swap and the prior system stack as
  fallback — no third-party font CDN request.

• Adopted the qte77 brand mark (logo-mark.paths.dejavu.svg) as the demo
  dashboard favicon, replacing the bespoke radar SVG (#295).

• Deep history for the demo's long-term charts: CNN Fear & Greed back to
  2011 and the 5s10s yield-curve slope back to 2011 (committed to the
  data branch). New scripts/backfill_yield_curve.py forces a full
  period="max" yfinance fetch — the daily cron only deepens on a first run
  (#287).

• equity-spy.yaml daily cron (#288). Recomputes the SPY indexed-return
  series and commits changed results/series/equity_spy/YYYY.json to the data
  branch via the shared verified-commit helper. Runs 23:00 UTC, staggered 30min
  after the yield-curve cron so the data-branch writers don't race the same ref.
  Mirrors yield-curve.yaml (same pinned action SHAs).

• equity_spy data source — SPY indexed-return series (#288). New
  analyze_stock_kpi.data_sources.equity_spy fetches SPY (the SPDR S&P 500 ETF)
  via yfinance and emits a derived rebased index (ret_indexed = close / epoch_close * 100, epoch = first close ≥ 2011) — never the raw close and never
  the S&P 500 index level (ADR-0011). Same per-year results/series/<kind>/
  shape + wrap-degrade boundary as yield_curve. Backend only here; the
  data-branch backfill, the cron, and the merged-chart UI follow separately.

• Governance for the #288 equity-macro overlay (ADR-0011). Records the
  decision to source the equity line from SPY (an ETF security, not the S&P
  Dow Jones index) and commit only a derived indexed-return series — never
  raw index levels — keeping it at the same redistribution tier as the existing
  yield_curve slope. Adds a docs/data-sources.md guardrail row and a repo
  NOTICE recording the non-commercial/educational, derived-data posture, the
  upstream ToS, and attribution for the bundled third-party libs (Chart.js MIT,
  Fuse.js Apache-2.0). (#288)

• Forks that enable Pages now self-host their own data branch. The dashboard
  derives DATA_BASE_URL from the Pages origin (<owner>.github.io/<repo> →
  that owner/repo's data branch) via a new pure ui/lib/data.js; ?base= still
  overrides, and the canonical qte77 deploy is unchanged. Prior art: the sibling
  agentic-job-offer-to-application-kit dashboard.

Changed

• Aggregator derives composite-field list from CompositeScores.model_fields. src/orchestrators/aggregated_scores_best_and_worst.py previously hardcoded the 7-field tuple in two places (the model definition and the orchestrator constant); adding a new composite would have been silently omitted from cross-universe ranking. Now the tuple is tuple(CompositeScores.model_fields) — single source of truth, zero hidden coupling. Regression guard added in tests/test_aggregated_scores_best_and_worst.py.

• Demo: screener_score displays as "qte77 Score" (#203). Display-only rename across KPI_GLOSSARY / detail-panel / time-series chart / radar chart. Internal field name composite_scores.screener_score is unchanged so JSON snapshots on the data branch keep validating. Table column header stays Score for column-width parity. PR #221.

• Demo: F&G rolling chart trimmed to a strict 12-month TTM window (#207). renderFearGreedChart now applies trimToRollingWindow(entries, 365) before render so the rendered window is always 12 months regardless of when in the year the dashboard opens (was drifting between ~8 and ~24 months as the loader consumes this-year + last-year files). PR #222.

• Demo: Op M % column hidden in simple view. Marks the <th> and the cellSpec as detail-only so simple view keeps the 4 headline columns (Ticker / Name / Sector / Score). PR #229.

• CI: dependabot updates grouped per ecosystem. groups: { python-deps / github-actions: { patterns: ["*"] } } on the uv + github-actions ecosystems — collapses N per-dependency PRs into one grouped PR per weekly run per ecosystem. PR #219.

• CI: demo-snapshot.yaml fan-out reads docs/demo/universes.json instead of a hardcoded JSON literal. Adding a new universe is now a one-file change (the JSON). Empty-input dispatch re-builds every universe the dashboard knows about regardless of preset .txt state — emitting a zero-row snapshot for an unpopulated universe is the right rebuild behavior. PR #232.

• Typed contracts in src/orchestrators/_shared.py: AuditRowBase + DedupedSnapshot. Closes assessment gaps #2 (3 divergent AuditRow models) and #3 (dedup_by_ticker returned dict[str, dict[str, Any]]). AuditRowBase extracts the 5-field decision-trail prefix (ticker, source_universes, snapshot_dates, eligible, excluded_reason) shared by the aggregator + longshort AuditRow subclasses; federal_contractors stays distinct (different domain — per-recipient match audit). DedupedSnapshot(BaseModel) replaces the untyped dict return of dedup_by_ticker; consumers switch from info["snapshot"] string-key indexing to info.snapshot attribute access. Rule-of-three pre-application for Phase 2b's FCF orchestrator (#192) which becomes the 3rd consumer of both shapes. JSON shape change is safe: the aggregator + longshort audit JSONs are written but never read back (JS dashboard only consumes federal_contractors audit). Matches the AGENTS.md "every structured payload is a BaseModel" rule.

• Extracted shared snapshot-loader + paired-output writer from build scripts. scripts/build_aggregated_scores_best_and_worst.py and scripts/build_enhanced_kpi_screener_longshort.py were ~80% identical (same SOURCE_UNIVERSES, same _load_snapshots, same main-loop dict-builder, same paired-preset+audit-write pattern); CodeFactor flagged the duplicated block as its sole remaining issue on main. Moved to a new scripts/_demo_snapshot_loader.py exposing load_snapshots / load_all_snapshots / write_paired_universe_and_audit. Each build script collapses from ~100 LOC to ~50 LOC of orchestrator-import + thin main(); Phase 2b's FCF orchestrator (#192) becomes a drop-in 3rd consumer. build_federal_contractors.py left alone (different input shape — no snapshot loading, single preset). Leading-underscore module name signals scripts/-internal per ADR-0007.

• fear-greed.yaml + yield-curve.yaml now route through scripts/data-branch-commit.cjs. Both daily crons carried byte-identical 36-line inline github-script blocks (getRef → getCommit → mkBlob → createTree → createCommit → updateRef) — the same verified-commit logic that demo-snapshot.yaml + universe-builder.yaml already delegate to the CJS helper. Migrating eliminates ~52 LOC of duplicated workflow YAML, ends the convention drift between the two stale crons and the two updated ones, and adds 422-race retry the inline blocks lacked (8 attempts with jittered backoff). Future verified-commit fixes propagate to all 4 workflows from a single file. Surfaced by today's scripts+workflows duplication sweep.

• README / UserStory / roadmap / architecture sync after #248. Surface the fcf_margin enrichment field in README's sample-output blurb; mark Phase 2b shipped in docs/UserStory.md and docs/roadmap.md (criterion 15 / tech rating stays deferred behind #21); record the new gate count (15 long + 14 short) and the _fetch_fcf_margin boundary in docs/architecture.md. No behaviour change — pure doc drift cleanup so the long/short screener's documented state matches the merged code on main.

• docs/demo/app.js: extract destroyChart, scoreYAxis/themedXAxis, and toggleHistoryHint to collapse repeated chart + empty-hint boilerplate. The chart-teardown idiom (liveCharts.delete + .destroy()) is now one helper across all six chart renderers; the 0–100 score y-axis and themed x-axis configs (each repeated 3×) become builder functions; the two byte-for-byte-identical empty-hint functions (sector-donut + long-term F&G) share one toggleHistoryHint, while the rolling + yield-curve variants that intentionally overwrite text stay separate. Behaviour-preserving, in-file only (no new modules/tests); node --check + tsc clean. First slice of the app.js size/repetition reduction.

• src/data_sources/fundamentals.py: single-source duplicated structure via _safe_ratio, _extract_two_rows, _equity_ratio. CodeFactor reported "0 duplication" but missed near-identical structure where only literals differ: the None/zero-denominator ratio guard (_compute_roi + both EQUITY-gated fetchers), the two-row extract-with-empty/missing-row/NaN guards (_read_rd_revenue / _read_fcf_revenue, one-frame vs two-frame), and the EQUITY-gate + try/except + safe-divide fetchers (_fetch_rd_to_revenue / _fetch_fcf_margin). Behaviour-preserving — the full regression suite stays green; per-method cognitive complexity drops (both readers and both fetchers to 0) and _read_rd_revenue gains the direct unit coverage it previously lacked. The fetch thunk in _equity_ratio keeps the yfinance property access inside the try and short-circuits the EQUITY gate before any fetch.

• docs/demo/lib/detail_rows.js: extract the pure detail-panel data (KPI_GLOSSARY, auditDetailRows, externalLinkRows) out of app.js. The KPI-glossary text and the row-tuple builders for the federal-contracts audit block + external links move to a tested lib module; app.js imports them. Adds tests/demo/detail_rows.test.mjs (7 cases for the two builders; the glossary is static data, no test). The DOM rendering (showDetail / renderRadar / renderTimeSeriesPane) intentionally stays in app.js — extracting it would require exporting the shared mutable liveCharts Set across modules or a god-object deps argument, both worse than co-location (AHA). Behaviour-preserving.

• docs/demo/lib/format.js: extract the pure value formatters + comparator (nested, fmtNum, fmtPct, compareValues) out of app.js into a unit-tested module. Adds tests/demo/format.test.mjs covering dotted-key access (incl. missing segments / nullish root), fmtNum null/NaN/precision, fmtPct null plus its intentional no-NaN-guard behaviour, and compareValues nulls-last-regardless-of-direction + string/number ordering. td() stays in app.js (DOM glue). Behaviour-preserving; prerequisite for the upcoming table.js / detail.js concern splits.

• docs/demo/lib/window.js: extract the pure time-window helpers (WINDOW_DAYS, filterByWindow, findClosestScore) out of app.js into a unit-tested module. Adds tests/demo/window.test.mjs covering the non-trivial branches (empty input, "all" passthrough, unparseable-latest fallback, inclusive cutoff boundary, arbitrary iso field; closest-score nearest-match + equidistant tie-break). Also deletes the redundant trimToRollingWindow — it was equivalent to filterByWindow(entries, "1y", "timestamp") — and rewrites its sole call site in renderFearGreedChart. Behaviour-preserving; the window math is now testable without a DOM.

• docs/demo/table.js: split the universe-table rendering concern out of app.js. Moves ALL_COLUMNS, the DOM builders (renderUniverseTable, renderRow, td, annotateEmpty) and the pure helpers (buildRowTitle, coverageCount, meanComposite, totalCompositeScore, emptyTableMessage) into a new sibling module. Module state (active universe, sort key/dir, filter query, row-click handler) is now passed in via an opts object instead of read from app.js globals, so table.js holds no mutable app state. app.js keeps a thin renderTable() adapter, so its 7 call sites are unchanged. Adds tests/demo/table.test.mjs (18 cases for the pure helpers) and widens tsconfig include to top-level *.js. Behaviour-preserving; app.js drops ~225 lines.

• docs/demo/app.js: prune verified-dead code. Removes two obsolete historical comments (the removed mobile auto-simple guard; the removed simple-mode external-link behaviour in onRowClick), the inert radarCanvas.className = "radar-canvas" assignment (no matching CSS/HTML selector — Chart.js gets the canvas node by reference), and the never-passed text parameter of renderRollingEmptyHint (its sole call site always resolves to EMPTY_HISTORY). Each removal was cross-checked against index.html / style.css / tests to confirm zero live references. Behaviour-preserving.

• docs/demo/detail_panel.js: extract the #row-detail side-panel lifecycle out of app.js. showDetail / bindDetailDismiss (plus the private dl / closeDetail helpers) move to a new sibling DOM module that mirrors table.js: it owns the detail aside but takes the row's audit record and the two chart renderers (renderRadar, renderTimeSeriesPane) via a context object instead of reading app.js globals, so it stays free of app.js mutable state and chart slots. app.js keeps a one-line onRowClick adapter and drops its now-unused detail_rows import + fmtPct. Net −163 LOC in app.js. Behaviour-preserving (logic moved verbatim); the consumed pure builders stay covered by tests/demo/detail_rows.test.mjs.

• docs/demo/lib/fetch.js: extract the data-branch fetch helpers out of app.js. fetchJson (fetch + throw-on-non-2xx + parse) and loadYearsFromBranch (this-year + last-year concat with per-leg silent-fail and ascending sort) move to a pure, DOM-free lib module; app.js imports them and keeps the thin state-closing wrappers (loadManifest, loadSnapshot, loadFearGreedYears, …). loadYearsFromBranch now takes the base URL as a parameter (loadYearsFromBranch(baseUrl, pathPrefix, sortKey)) instead of closing over DATA_BASE_URL, which is what makes it pure and testable. Adds tests/demo/fetch.test.mjs (5 cases: 2xx parse, non-2xx throw, two-leg concat+sort, per-leg silent-fail, non-array-leg skip). Net −31 LOC in app.js. Behaviour-preserving.

• docs/demo/lib/chart_axes.js: extract the themed Chart.js axis factories out of app.js. scoreYAxis (0–100, stepSize 25) and themedXAxis (tick-capped x) move to a pure, DOM-free lib module. To keep lib/ DOM-free they take an injected cssVarFn(token, fallback) instead of closing over app.js's cssVar: scoreYAxis(cssVar) / themedXAxis(cssVar) at the four call sites (renderTimeSeriesPane, renderFearGreedChart, renderMonthlyFearGreedChart, renderYieldCurveChart). The returned objects still carry deferred color closures (Chart.js resolves them lazily on theme flip), now documented as a caller-owns-lifetime contract. Adds tests/demo/chart_axes.test.mjs (4 cases: static shape + deferred-injection for both factories). Net −19 LOC in app.js. Behaviour-preserving.

• docs/demo/app.js: dedup the chart empty-hint helpers onto a single toggleHistoryHint. renderRollingEmptyHint and renderYieldCurveEmptyHint were 14-line near-twins re-implementing the create/update/remove logic; both collapse to one-line delegations (matching the existing renderDonutEmptyHint / renderLongTermEmptyHint wrappers). toggleHistoryHint gains an optional text param and switches from skip-if-existing to update-if-existing — which is what preserves the "loading… → no history yet" overwrite of the static placeholders shipped in #fg-chart-wrap / #yc-chart-wrap. Verified safe for the donut + long-term callers: they have no static placeholder and only ever store EMPTY_HISTORY, so the update path is a no-op for them. Net −24 LOC in app.js. Behaviour-preserving.

• ui/charts.js: extract the Chart.js rendering layer out of app.js (closes #268). All chart builders (sector donut, radar, F&G rolling/monthly, yield curve, detail-panel time-series) plus the shared chart infra (liveCharts/cssVar/destroyChart/bindThemeObserver/toggleHistoryHint) and the long-term-tab + window-chip wiring move verbatim into a sibling module. The only app↔chart coupling is wired through a single injected context object — initCharts(ctx) with live getters/setters for snapshot/sectorFilter/ltFgWindow/ycWindow/manifest/… and afterSectorToggle/afterWindowChange callbacks — mirroring the table.js / detail_panel.js opts-object precedent. Net app.js −722 LOC (1278 → ~615); behaviour-preserving (147 vitest pass; headless render verified: donut click-filter, tabs, window chips, detail radar + time-series all functional, 0 console errors).

• Import package renamed src → analyze_stock_kpi (src-layout; ADR-0009, resolving the ADR-0007 refactor candidate). The package now lives at src/analyze_stock_kpi/ and is imported as analyze_stock_kpi (e.g. from analyze_stock_kpi.domain.universe import resolve_universe); the CLI entry is python -m analyze_stock_kpi. Breaking for any code importing src.* — switch to analyze_stock_kpi.*. The PyPI distribution name (analyze-stock-kpi), CLI behaviour, and public API are otherwise unchanged.

• Dashboard re-themed to the qte77 EyeRest brand (zero-blue, warm). The ui/ dashboard moves from its cool-gray Apple-system palette to EyeRest's umber/parchment tones (DESIGN.md). Blue/teal accents → the amber accent + the brand's zero-blue data arc across rating chips, KPI heatmap, the score-cell ramp, the sector donut, and the favicon (#278); the neutral surfaces → warm parchment (light) / umber (dark) (#282). Both light + dark, system-theme default kept, WCAG AA on the brand pairs. Fully token-driven (CSS custom properties), so scheme/variant flips re-resolve every value.

• Data-branch results/ layout grouped by kind. Per-year series moved to
  results/series/{cnn_fg,yield_curve}/; every per-universe audit JSON moved
  under results/audit/<universe>/ (federal-contractors, aggregated-scores,
  enhanced-kpi-screener). Demo snapshots (results/demo/) and the universe
  audit (results/audit/universes-*.json) are unchanged. Config, the snapshot
  workflows, the demo loader, and docs all route through the new paths; the
  data branch is migrated in lockstep so the live demo never 404s. Plan:
  docs/plans/restructure-results.md.

• Demo header constrained to the content-column width. header now shares
  main's max-width: 1400px; margin: 0 auto, so on wide viewports the title and
  the theme / report-issue / updated meta align with the cards instead of
  stretching edge-to-edge. No effect below 1400px.

• Merged long-term-context chart (#288). The "CNN F&G long-term" and "5s10s
  slope" tabs collapse into one "Long-term context" chart: the F&G monthly median
  and the 5s10s monthly mean (normalized to 0-100) share the left axis, and the
  SPY indexed return sits on a logarithmic right axis. All three reconcile
  onto one monthly grid via the new pure ui/lib/combined.js
  (normalizeSlope / aggregateMonthly / buildCombinedSeries) +
  logRightAxis in ui/lib/chart_axes.js. SPY is the derived equity_spy series
  on the data branch (ADR-0011, never raw index levels).

• NOTICE expanded to the full third-party-attribution convention (matches the
  paperverse sibling). Now opens with the project's Apache-2.0 license header
  and reproduces/points to the verbatim license texts of every bundled,
  redistributed component shipped in the built UI: Chart.js (MIT), Fuse.js
  (Apache-2.0), and the Inter + JetBrains Mono fonts (SIL OFL 1.1) — the
  fonts were previously omitted. Non-redistributed Python/JS dependencies are
  explicitly excluded. Also corrects a stale "MIT" reference in
  docs/data-sources.md (the repo is Apache-2.0).

• qte77 watchlist expanded to 100 symbols (+27). Added Allianz (ALV.DE),
  Deckers (DECK), Enel (ENEL.MI), Comfort Systems (FIX), Intl Container
  Terminal Services (ICTEF), InterDigital (IDCC), Louisiana-Pacific (LPX),
  Moody's (MCO), Meta (META), Monster Beverage (MNST), Monolithic Power
  (MPWR), NetEase (NTES), Qualcomm (QCOM), REA Group (REA.AX), Sezzle
  (SEZL), Sterling Infrastructure (STRL), TSMC (TSM), Clear Secure (YOU),
  SK hynix (000660.KS), Zhejiang NHU (002001.SZ), kakaku.com (2371.T),
  Realtek (2379.TW), MediaTek (2454.TW), Evergreen Marine (2603.TW), Yutong
  Bus (600066.SS), Organo (6368.T) and Advantest (6857.T). Each symbol was
  verified against the yfinance KPI surfaces the screener reads
  (.info / .income_stmt / .cashflow); inline # Name comments document the
  cryptic international codes. SK hynix is the lone partial — Yahoo omits a ROIC
  input so that one composite stays None; all other KPIs populate.

• Theme toggle is now a single cycler button instead of the 3-button
  segmented control. One click advances system → light → dark → system;
  the button shows the active mode as an <icon> <word> label so it stays
  glanceable. "System" (follow OS) remains a reachable state, and the
  ?theme= URL + localStorage persistence is unchanged. A visually
  hidden aria-live="polite" status region announces each change to
  screen readers, since focus stays on the button after a click.

• JS tooling + tests consolidated under ui/ with a Vite build (ADR-0010,
  #289). package.json, the eslint/prettier/vitest configs, and tests/demo/
  moved from the repo root into ui/ (ui/tests/), so the root is Python-only.
  ui/ is now Vite source → ui/dist/ is the deployable; gh-pages.yaml
  builds (npm run build) and uploads ui/dist instead of raw-copying ui/.
  Vendored Chart.js/Fuse.js, favicon.svg, and universes.json moved to
  ui/public/ (served verbatim under the project base path). The dashboard's
  data still loads at runtime from the data branch — never bundled. The
  Makefile + validate.yaml JS steps now run from ui/; validate also builds
  the UI to catch breakage on PRs.

• 5s10s slope chart de-noised to a weekly mean on the wide windows. The
  yield-curve tab now aggregates the daily 10y−5y slope to an ISO-week mean for
  the 5y / 10y / all windows (1y stays daily), mirroring the F&G long-term
  monthly view so multi-year context reads as trend rather than noise. New pure
  ui/lib/weekly.js (aggregateWeekly); client-side only — no change to the
  data-branch results/series/yield_curve/ files. (#308)

• Refreshed the "Why these charts?" copy for the merged long-term chart
  (#288). The pane described the F&G long-term and 5s10s charts as separate
  tabs; it now describes the single combined chart — F&G monthly median + the
  normalized 5s10s on the 0–100 axis and the SPY-derived equity index on the log
  axis.

• Dashboard adopts the qte77 EyeRest brand theming, aligned with the sibling
  paperverse + agentic-job-offer-to-application-kit dashboards. CSS tokens are
  renamed to the brand-canonical names (--surface / --text-muted / --primary,
  plus a new --primary-on) and the light/dark cascade moves to html[data-theme],
  set by a new repo-local ui/theme.js that mirrors qte77.github.io/assets/theme.js:
  one cycling ◐/○/● button, localStorage["qte77-theme"], a themechange event that
  recolours the charts, a dynamic aria-label + #theme-status live region,
  prefers-reduced-motion, and a <head> anti-FOUC guard. The data arc re-tones per
  theme, so rating-chip text uses --primary-on to keep contrast in both modes. The old
  lib/theme.js + its test are removed (the toggle logic now lives in ui/theme.js).

• Theme toggle no longer writes ?theme= to the URL on click — it reads ?theme=
  on load (deep-links still work) and persists the choice to localStorage.

• README restructured to the qte77 doc-structure canon. Hero tagline under the H1,
  then What → How → Why → References → License: the one-line pitch moves to a hero
  blockquote, a new Why section states the keyless-vs-paid-feed gap, What is
  trimmed to ≤7 reader-value bullets (field lists / composite formulas / preset details
  now link out to docs/architecture.md), and the non-canon Sample output /
  Universe sources sections fold into What/How. Orientation only; depth lives in docs/.

Fixed

• CodeFactor E241: collapse multi-space alignment in enhanced_kpi_screener_longshort + its test. Three fixture dicts and a parametrize tuple table in tests/test_enhanced_kpi_screener_longshort.py plus the _NUMERIC_GATES table in src/orchestrators/enhanced_kpi_screener_longshort.py used column-aligned spacing after : / ,. CodeFactor's pycodestyle (which ignores the repo's ruff config) flagged ~57 E241 hits, tanking the two files to F / D grades. Collapsed to single-space; no semantic change. Preferred over a new setup.cfg [pycodestyle] override because no sibling qte77 repo uses one and pyproject.toml ruff stays the single lint source of truth. Inline-comment , # ... sites (PEP 8 standard) untouched — CodeFactor isn't flagging those.

• Demo: dark-mode chart grid + axis labels visibility (#205). Chart.js scale defaults were theme-blind (rgba(0,0,0,0.1) grid lines, near-black tick labels) — vanished against the dark panel. Add scriptable () => cssVar() closures for scales[].grid.color + scales[].ticks.color across fearGreedChart, monthlyFearGreedChart, yieldCurveChart, timeSeriesChart, radarChart. sectorChart's static --panel slice-seam color converts to a scriptable closure too; radarChart + sectorChart register in liveCharts so bindThemeObserver re-evaluates their scriptable colors on theme flip. PR #220.

• CI: sam.gov added to lychee's exclude list. https://sam.gov/about/terms-of-use (referenced in docs/data-sources.md) returns 403 to lychee's UA, breaking the linkChecker job on every unrelated PR. Same pattern as the existing sec.gov / aaii.com / naaim.org exclusions. PR #223.

• CI: include long/short universes in demo-snapshot.yaml fan-out. PR #228 added enhanced-kpi-screener-{longs,shorts} to universe-builder.yaml and docs/demo/universes.json but missed demo-snapshot.yaml — picker showed them, selection painted "no history yet" forever because no snapshot ever landed at results/demo/enhanced-kpi-screener-*/…. PR #230.

• Demo: dedupe the "no history yet" empty-hint string. Extracted into one module-level EMPTY_HISTORY const; replaced 5 hardcoded copies + 1 near-duplicate. AHA-warranted (5 of one literal); context-specific strings ("first cron run pending", "loading manifest…") stay inline. PR #224.

• docs/demo/types.d.ts: declare Row.composite_scores as nullable. src/data_sources/fundamentals.py:FundamentalsSnapshot.composite_scores is CompositeScores | None, which serializes to null. The TS declaration was composite_scores?: CompositeScores; — optional but not nullable; type narrowing on a literal null value would have failed. Now composite_scores?: CompositeScores | null;, matching the convention every other nullable field already uses (?: T | null). One-line drift fix surfaced by today's contracts audit; the remaining 11 FundamentalsSnapshot fields absent from Row are intentional per the file's "subset the dashboard reads" header policy and confirmed unused in docs/demo/.

• Sector donut legend now reappears after browser zoom-out using ResizeObserver-based auto-hide (#152).

• docs/demo/tsconfig.json: bump moduleResolution node → bundler (and module ES2020 → ESNext). The legacy node (node10) resolver is deprecated on the TypeScript 7.0 track and fails make lint_js / make validate with TS5107. bundler is the correct resolver for this browser-native ES-module demo (relative ./lib/*.js imports with explicit extensions, no Node package semantics). Dev-tooling only; no runtime or behaviour change.

• F&G backfill is now a strict gap-fill — it no longer clobbers a
  CNN-direct row's subindicators / previous_* fields when a whit3rabbit row
  shares the same midnight-UTC date key. The prior _upsert(force=False) path
  replaced same-timestamp rows on any content difference; the test meant to
  guard this used an unrealistic intraday timestamp and so missed it (#287).

• Demo dashboard now shows the full long-term history. The data loader only
  fetched the last two years, so the CNN F&G long-term + 5s10s slope charts
  stopped at ~2025 even after the data branch was backfilled to 2011 (#287).
  loadYearsFromBranch now fetches every year from the series floor (2011)
  through the current year, so the 5y / 10y / all window chips paint the whole
  range.

• sp500 universe: BRK.B → BRK-B. Yahoo / yfinance uses a hyphen for
  Berkshire Hathaway's Class B share-class symbol; the dotted BRK.B resolves to
  all-null KPI rows (verified 0/3 on retry-probe vs BRK-B 3/3). The qte77
  watchlist already used BRK-B; this aligns sp500.txt. The other empty-data
  symbols from the same audit (MMC, latam .SA lines and their ADRs ERJ /
  EBR) are an upstream yfinance quoteSummary issue, not symbol errors —
  tracked in #312.

• No more flash-of-wrong-theme on load for an explicit light/dark
  override. An inline guard in <body> resolves the theme (URL >
  localStorage > system) and sets the body class before first paint,
  instead of waiting for the deferred app.js module.