The flagship. 8 cross-assets (gold · oil · 10Y · real-yield · USD · NAS100 · HK50 · CNY): normalized overlay, decoupling radar, correlation matrix, gold spotlight, multi-scale tensor, snowball projection. Flashes only what newly fired.
● live 02 · signalDeep-dive any signal over time: toggle level ↔ z-score ↔ velocity ↔ decoupling residual for every asset on one axis, with tunable window. The "how did this build?" lens.
● live 03 · moversThe default screen: every asset ranked by heat (decouple · extreme · velocity + bonuses), tagged FLIP / ACCEL / DECOUPLE / EXTREME / CLIMBING / VOL. "What's turning right now."
● live 04 · landscapeThe entire world-markets landscape: pick any markets across the globe — US · Asia (Hang Seng · Nikkei · KOSPI/KOSDAQ Korea) · Europe · FX · rates · commodities · crypto — overlay them normalized vs each other AND vs SPX/NAS100 across every timeframe (90m→max), with a rolling-correlation decouple pane and a ranked decoupling board. Triangulate where any market is decoupling.
● live 05 · gammaDealer options exposure (à la VolSignals VS3D): gamma / vanna / charm by strike, a strike × expiry grid heatmap, by-expiration, and key levels (zero-γ flip, call/put walls). From Yahoo chains + Black-Scholes.
● live 06 · biasThe ensemble: fuses every cross-asset signal + the dealer-gamma regime into one equity directional bias + P(up), over time vs the S&P, with a contributor breakdown and agreement read.
● live 07 · ratioDivide any instrument by any other — treasuries ÷ oil, gold ÷ USD, VIX ÷ VVIX — from a macro universe (rates·commodities·FX·vol·factors). The ratio plots vs its own rolling ±2σ band with percentile/z extremes and S&P/NAS100 divergence. Swap legs in a click; a preset radar ranks common pairs by how extreme they are now.
● live 08 · retailIs the crowd highly leveraged? Short-dated (≤7 DTE) call/put OI skew, OTM-lottery concentration, and volume/OI churn across SPY · QQQ · TSLA · NVDA — blended into a 0–100 retail-leverage index on real Massive OI.
● live 09 · cohortsEVERY instrument as its own normalized LINE on ONE chart (z-score default · fit-to-axis · %-change), grouped by who trades it — Retail (NVDA·TSLA·QQQ·IWM), Vol complex / MM-dealer proxy (VIX·VVIX·SKEW), Bankers (USD-FX·ZN·ZB·GC·CL) — vs SPX/NAS100 baselines (solid, never muted), with a CPI/M2 real-terms adjust applied before normalize. Paired with the stretch TAPE (heatmap, every instrument vs its own history, 6 coloring modes, date-column sort) and a slow-money liquidity backdrop. Every instrument individually selectable, retail P/C reference, full pan/zoom/dateline.
● live 10 · uoaScans SPY · TSLA · NVDA · QQQ chains (Massive real-time) for any contract trading ≥24× / ≥30× its open interest — heavy NEW positioning — and fires a Telegram alert covering both tiers. Plus a real-time options-flow chart over the last full session, with the high-flow 90-min open/close windows highlighted and a toggleable SPX/NAS100 reference.
● live 11b · twitterLikes / replies / reposts per post for @BITCOINFUNDMGR, banked every ~4h. Engagement velocity (Δlikes) as a soft proxy for retail attention / speculative volume — pairs with the options retail-leverage gauge.
● live 11 · bettingExchange-wide VOLUME + OPEN INTEREST across Kalshi + Polymarket, banked every ~15 min, 24/7 (neither venue keeps the history — so ours is the alpha). A leading indicator for the prediction-market equity basket SRAD · GENI · DKNG: rising exchange vol/OI = a flow tailwind for the data-feed + sportsbook enablers; the basket overlay shows whether the equities followed, and divergence is the actionable signal. Sports-vs-politics mix, OI reconciled on a USD-notional basis, top markets across both venues.
● live 12 · fundamentalsQQQ · TSLA · NVDA 0DTE Options Volume — 5y history + live, vs SPX / US M2 / Fed balance sheet (WALCL) / reverse repo (RRP). One continuous daily series: a 5-year deep-history seed (Polygon OSI-construct, DTE≤1) fused with automatic forward accrual from the Massive real-time chain (no venue keeps this history — so ours is the alpha). Does 0DTE speculation track liquidity, lead the S&P, or diverge? Put/call VOLUME (frenzy gauge), $-premium traded, and the per-name breakdown (QQQ · TSLA · NVDA each individually selectable — the basket is always decomposable).
● live 13 · fxv2 of /fx-macro: a thumbnail grid of every currency (DXY · majors · Asia · EM · non-USD crosses like CNY/JPY · KRW/CHF) — click tiles to OVERLAY MULTIPLE currencies on ONE normalized detail chart. z-score / %-change so different-scaled pairs (¥150 · €1.08 · ₩1350) compare on one axis. USD-strength toggle inverts USD-quote pairs so every line means "up = USD stronger". Macro overlays: SPX · NAS100 · gold · 10Y real yield.
● live 14 · divergenceThe dedicated home of the multi-scale divergence panel. Two oscillators: price↔RSI (momentum, every instrument) and price↔OBV (On-Balance-Volume — volume-confirmed; equities & futures only, FX/indices/rates read honest N/A). 9 cells per instrument = 3 timeframes (1h·4h·1d) × 3 swing lookbacks; confluence + 6 view modes + a price+oscillator detail chart with the exact swing pair drawn. A comprehensive cross-asset universe vs SPX/NAS100.
● live 15 · inflationThe cohorts "every instrument vs its own history" LINE chart, standalone, with one headline superpower: Adjust for inflation or money supply. Reprice every ticker (and the SPX/NAS100 baselines) into real terms — ÷CPI (purchasing power: did it hold its value?) or ÷M2 (did it beat money printing?), applied BEFORE the z-score / fit / %-change normalize. Every instrument individually selectable, SPX/NAS baselines solid & never muted, retail P/C reference, full pan/zoom/dateline. Lines only — a focused real-vs-real comparison view.
● live 16 · sectorsEvery major sector on ONE chart: 11 GICS SPDRs + 15 thematic baskets — Defense/Military · Quantum · Bitcoin · Software · Rare Earth · Space · Semis · AI/Robotics · Clean Energy · Uranium · Gold Miners · Biotech · Cybersecurity · Homebuilders · Transportation. The mandatory tape+line pair: a normalized LINE chart (all 26 sectors, impulse-z default · fit · %, CPI/M2 real-terms adjust, SPX/NAS100 baselines solid & never muted) paired with a HOT/COLD stretch TAPE (heatmap — red = hot/stretched, blue = cold/washed-out; the mandatory-4 coloring modes: own history · cross-sectional rank · vs SPX · vs NAS100; date-column sort). Click any sector to drill into its constituents as individually-selectable lines — a basket is never an averaged line. Upgrades v1 /apps/sectors.
● live 17 · cryptoThe v1 Coin Screener, re-skinned to apps2: top-100 coins by market cap with momentum · volume-spike · RSI · pattern-detection filters and presets, across four views — sortable table, pattern grid thumbnails, compare sparklines (5y/1y/1q/1m), and the signature divergence matrix (1h/2h/4h/D × tight·std·wide·very-wide RSI/OBV divergence, 2h/4h aggregated from hourly) with an interactive detail chart + trendlines + Pine export. Topped with the apps2 tape+line pair: a normalized LINE chart (top coins, impulse-z default · fit · %, BTC/ETH baselines) paired with a HOT/COLD stretch TAPE (own history · cross-sectional rank · vs BTC · vs ETH · date-column sort). Reads the /apps/crypto per-ticker archive in place. Upgrades v1 /apps/crypto.
● live 18 · stocksThe v1 Stock Screener, re-skinned to apps2: ~390 names across US + Korea/China/HK/Japan/Taiwan with theme + region filters, volume-spike · RSI · multi-timeframe divergence (1h·2h·4h·8h·D·W·M, RSI + OBV) · price-action pattern detection (triangles · wedges · channels · H&S · trendlines, with TradingView Pine export) across three views — sortable table (now with a trailing P/E column) · pattern grid thumbnails · 4-timeframe compare sparklines — plus curated bull/bear presets, collections, an AI-research prompt builder, and a candle detail modal. Topped with the apps2 tape+line pair: a normalized LINE chart (selectable top-N of the filtered set · impulse-z default · fit · %, SPX (SPY) / NAS100 (QQQ) baselines solid & never muted) paired with a HOT/COLD stretch TAPE (own history · cross-sectional rank · vs SPX · vs NAS100 · date-column sort). Reads the /apps/stocks per-ticker archive in place (never calls Yahoo from the browser). Upgrades v1 /apps/stocks.
● live 19b · pegrokRetired 2026-08-13 — now redirects to pe, which carries all 31 names this
charted plus its 7 footnoted loss-makers, on a 56-name roster with deeper history and a
daily current-P/E read this page never had. The merge was gated on a by-name subset check
that failed first — Home Depot, Johnson & Johnson and PepsiCo were charted here
and missing from pe; all six gaps were filled and the check re-run empty before this
was retired. Kept only so old bookmarks land somewhere correct; the page is parked as
index.superseded.parked.html and its builder as
build_pegrok.superseded.parked.py, which names the two fixes it donated to
pe and stocks.
12 mega-caps — Amazon · Oracle · Apple · Micron · AMD · Nvidia · Microsoft · Cisco · Intel · Alphabet · Meta · Tesla — back to the dot-com era for relative comparison, monthly. Deep multi-order-of-magnitude data, so it defaults to LOG-scale lines (actual P/E and actual $ on the axis, not z-scores): a Historical P/E chart (~5–1600) and a Historical market cap chart ($1B–$4.7T), each stock its own selectable line (log · linear · z-score · fit toggle), the dot-com window (1998–2002) shaded. Each line chart is paired with its HOT/COLD stretch TAPE (heatmap — red = expensive/large vs own history, blue = cheap/small; own-history · cross-sectional rank · vs-SPX/NAS100 shown but disabled with a reason — no index P/E in the set; date-column sort). Local monthly dataset (SEC EDGAR 2007+ · curated dot-com seed ramping · price to 1984); coverage & gaps stated honestly.
● live 21 · oilWhat flows into Korea and what it costs. Crude imports, refinery intake, the reserve level (crude stocks vs the ~146 Mbbl KNOC strategic reserve + days-of-cover), and LNG imports · gas demand · gas stocks — Korea imports all its gas as LNG (pipeline imports = 0). Paired with daily prices (Brent · WTI · Henry Hub). Import volumes + reserves are monthly (JODI World Database — the finest FREE granularity; daily tanker-arrival volume is paid); prices are daily (FRED). Every series its own selectable line with the full chart control set; coverage & granularity stated honestly.
● live 22 · insightThe cross-app "what's firing right now" feed. A 2×/day cron reads ONLY our own recorded stores (the prices spine · signal · bias · gamma · cohorts · divergence · 0DTE flow) and surfaces the six non-eyeballable computed reads: percentile extremes (a metric at the edge of its own history), correlation flips (a pair whose rolling corr just changed sign), decouplings (a normally-correlated pair at an extreme-low corr percentile), divergence confluence (RSI/OBV divergences stacking across scales), lead/lag flips (a measured leader whose concurrent corr flipped), and put/call spreads (volume-P/C vs premium$-P/C gap). Ranked WITHIN each type — no fused score — every read with its own-history distribution, the date range it has been active, and a deep link into the source app. Nothing forward-looking. The feed is itself record-first.
● live 23 · peTrailing P/E line chart across 49 deliberately-mixed names spanning P/E EXTREMES: legacy depth (Apple · Intel · IBM · Oracle · Cisco), mega-caps (NVIDIA · Tesla · Amazon · Meta · Broadcom · Netflix · Salesforce · Adobe), semis (TSMC · SK Hynix · AMD · Micron · Qualcomm · ASML · Arm), new/quantum/AI (Palantir · Super Micro · IonQ · Rigetti · D-Wave), fintech (Coinbase · Robinhood), cross-sector leaders & unusual P/E (Costco · Chipotle · Lilly rich · Exxon · JPMorgan · Ford cheap · Coca-Cola · Walmart · P&G · UnitedHealth · Caterpillar · Visa), sleepers / underdogs on a low P/E that arguably should re-rate (Dell · NXP · ON Semi · Baidu · Alibaba · PayPal · Uber), and private SpaceX (no P/E).
Monthly, log-scale default (actual P/E on the axis), every stock its own selectable line vs the S&P 500 P/E reference (own right axis, never muted), R-UI-29 end labels, click-to-isolate with an own-history 10–90% band. Paired with the HOT/COLD stretch TAPE (own history · cross-sectional rank · vs-S&P; date-column sort) and computed relative-value insight along THREE axes — cheap-vs-own-history (most-stretched hero + tape), cheap-vs-peers (cross-sectional rank, own-history percentile/z), and cheap-vs-own-potential (the trailing↔forward gap, the sleeper lens). A P/E-focused sibling of pemarcap; deep history reused from it, extended via SEC EDGAR (10-K/20-F/IFRS), Yahoo fallback where EDGAR tags are non-standard (Visa · Baidu). Loss-makers & SpaceX carry no line (honest gaps, never faked). ● live 24 · bleedingDIRECTION OF CAPITAL ROTATION across borders in the most innovative chip · robotics · AI names — 30 peers across China · Korea · Taiwan · Japan · Netherlands · UK · US, grouped by sector (AI-GPU · memory · foundry · image sensors · equipment · robotics · platforms · global semi). Within each hot sector it surfaces which COUNTRY's peers are being ACCUMULATED vs DISTRIBUTED — is capital rotating US chips → China chips, Korea memory → US memory? Everything is RELATIVE (different currencies/timezones → normalized, daily-by-date) with NO SPX/NAS/FRED basis (broad-market moves everything uniformly = noise for rotation): peer-vs-peer normalized lines + a hot/cold stretch TAPE (own history · % move · vs own mean · cross-sectional rank) + a ranked overbought/oversold & accumulation/distribution table (RSI · own-history stretch · OBV) + a sector×country cycle matrix + the seven rivalry ratios (Cambricon÷NVIDIA · SK Hynix÷Micron · SMIC÷TSMC · Will Semi÷Sony · UBTech÷Fanuc · Biren÷NVIDIA · Naura÷ASML) each vs its own ±2σ band, and the multi-scale price↔RSI/OBV divergence panel. Premise-first: every chart prints what it measures + its directional premise; the signal is the deviation.
● live 26 · countertrendThe full-landscape counter-trend detector. 7 instruments — Gold · Oil · USD/CHF · SOXX · S&P 500 · US 30Y (^TYX) · China A50 (2822.HK) — on a 21-pair × 8-timescale matrix (90m · 4h · 1d · 1w · 1m · 1q · 1y · 5y). Premise-first: every pair prints its baseline premise + what each direction means; the signal is the DEVIATION. Premise classes (strong/moderate/weak/regime/none) honestly arm or disarm the sign-alarm — five pairs openly have no premise to violate, and a pair auto-demotes when its empirical sign contradicts the stated one. No arbitrary-lookback sliders — three pre-registered sensitivity tiers (sensitive/standard/strict) over FIXED canonical windows. Two reads: Pearson-free direction-rarity on the short scales, rolling correlation (n≥24) on the long ones. Per-pair detail (both legs + rolling-corr chart with premise band, onset date + live elapsed timer), an all-instrument overlay with z-score / fit / %-change modes, an extremes board, and an everything-rally co-move flash. Record-first (record.php banks daily). Data via decoup/data.php (Yahoo + FRED) — no new paid feeds.
Cross-asset relationship violations — not RSI pivots. Gold · oil · USDCHF · SOXX · SPX · US 30Y · China (FXI) on pair×horizon matrix + recency flash. Slider-tunable floors. Data via decoup/data.php.
Same div universe as 1-minute OHLC candles, default last 4 hours, polled every ~20s. Equities via Alpaca SIP when available, futures/FX/index via Yahoo. Window anchored to last trade. Feed: div/rt/ohlc.php + universe.json.
Fed Z.1: nonfinancial corporate equities outstanding vs held on corporate books, open-market proxy (outstanding − corp-held), household holdings — vs 20y SPY / NAS100 price action. Quarterly structure + daily prices, shared window.
● live 27 · gasPaste any Ethereum address → all deposits / withdrawals (ETH + ERC-20 + internal), unique summaries by counterparty / token / pair, filter from–to or before / after dates. CSV export. Blockscout free (optional Etherscan key).
● live 28 · 401kEstimated 401k / defined-contribution buying pressure — the mechanical, price-insensitive equity bid from payroll deferrals — modeled from the wage base (FRED aggregate wages × contribution rate × participation × equity allocation × a documented auto-enrollment adoption glide) and overlaid on SPX. A comprehensive 29-series macro panel (payrolls · GDP · employment · income · saving · liquidity · sentiment · household equity) with the composite AND every constituent individually toggleable. Interactive assumption sliders (contribution rate · participation · equity allocation) recompute the estimate live. The analytical layer surfaces correlations: a sortable correlation table (each metric's YoY-growth vs SPX and vs the 401k bid, 1y/3y/5y/full, per-column heat) and a lead-lag heatmap (±12 months — who turns before whom; does the 401k bid lead the market?). All on year-over-year growth (levels are non-stationary and correlate spuriously). FRED key-free + Yahoo.
● live 29 · rsidivReal-time RSI(14) divergence scanner on ES (S&P 500) and NQ (Nasdaq-100) futures, 30-minute + 1-hour candles, across three lookback windows (8 / 13 / 21 bars). Flags bullish (price lower-low while RSI higher-low → momentum turning up) and bearish (price higher-high while RSI lower-high → turning down) regular divergences, drawing each as a trendline on both the price and RSI panels. Fires a Telegram alert on every newly-confirmed divergence (deduped, stale-guarded) plus a live on-page alert feed. Server-side scanner, auto-continuous futures feed, full per-chart controls.
● live 30 · xyzA rotatable 3D market chart: X = time, Y = price, Z = trading volume (depth). Four instruments on one scene — QQQ · SPX · NAS100 · Russell 2000 — each a colored polyline that weaves front↔back as its volume swells or fades, with optional per-bar volume fins to the floor. Built on the paid real-time feeds with real per-candle volume: SPX/NAS100/Russell = Massive CME futures front-month (ES/NQ/RTY, ~23h sessions; multi-year daily stitched from the quarterly contracts), QQQ = Alpaca equity bars — no ETF volume proxies. Prices are normalized (% change / z-score) so all four compare on one axis; depth is each instrument's own-max volume. Drag to orbit, wheel-zoom, shift-pan, with Front (classic price chart), Top (time×volume map) and Iso presets, on-canvas drawing tools (trendline / horizontal / ray / note), the full window set (30m→5y) and hover tooltips carrying the real price + real volume + its contract.
Where BlackRock's money is actually moving. Net flow = primary-market creations/redemptions (Δ shares × NAV) from the official fund.xls, refreshed 3×/day. A summary dashboard tops the page: combined US-equity flow over time, an 11-sector heatmap (green=buying/red=selling by z-score), and an "unusually strong buying/selling" panel that flags when a sector's 20-day flow breaks its own 2-year range. Below, a per-ETF radar of canaries — credit (HYG/LQD), rates (TLT), EM debt (EMB), small-caps (IWM), gold (IAU), IBIT — plus the broad core and all 11 US sector funds. Per-ETF page: 3-panel price/volume/flow chart, rolling cumulative overlay, PNG/CSV.
● live 31b · blackrockewyRetired 2026-08-04 — now redirects to blackrock / ewy, which is the live series
(official fund.xls + paid Massive aggregates, refreshed 3×/day). This route had been serving
its own snapshot, frozen at 2026-07-20 off the free Yahoo feed. Kept only so old bookmarks land
somewhere correct; the v1 page is parked as index.superseded.parked.html.
Live Gantt chart and countdowns for pre-market, regular, after-hours, and futures sessions across the US, Korea, Japan, China, Hong Kong, Singapore, Germany, and London. Holiday-aware Yahoo trading periods, 8–72 hour zoom, pan, pinned date lines, time override, fullscreen, and PNG export.
● live 33 · antsWhat can be measured honestly about the crowd: real OCC customer-range 0DTE volume, a configurable tiny-ticket tape proxy (contracts + premium cap), and the spread-aware profit/loss distribution of estimated buyer-initiated tickets — explicitly tickets, never invented unique-trader win rates. Adds optional WSB comments / 12h and Robinhood app downloads / day attention/acquisition banks, with every estimate visibly labeled.
● live 34 · basicsThe five markets that frame every macro conversation, drawn together on one shared time axis, each on its own scale so shape and timing compare directly. Realtime Yahoo candles (gold GC=F, BTC-USD, crude CL=F, S&P ^GSPC in hours and ES=F basis-adjusted outside them, 10Y ^TNX) with a ~25s view-preserving poll. Lines by default; tap any line to turn it into candles and move the real-value Y ruler onto it. Canonical window set 1h→5y at the finest candles the source can actually cover (impossible combinations greyed out with the reason), per-ticker and group opacity, a hairline trend-line tool that survives zoom, and a stacked stretch tape companion with six colouring modes. Built mobile-first: pinch-zoom, one-finger pan, tap-to-identify, long-press dateline.
Everything /basics does — gold, BTC, the US 10Y, the S&P 500, crude, semis, DXY and the live USD gauge on one shared time axis, each on its own scale, with the trend-line tool and the stretch tape — plus the air temperature in New York City and Seoul as two toggleable background series on their own °C scale. A persisted Overlay ↔ Pane switch breaks the temperatures out into their own flush-stacked indicator pane with a real °C axis and its own height grip, while the time axis, panning, zooming and the date line stay shared across every canvas. Highest resolution Open-Meteo offers: 15-minute points to a week, hourly to a quarter, ERA5 daily means (with the day's max/min shaded) for 1 and 5 years — observations only, no forecast bars. Market data is read from the /basics endpoint and store: one writer, never fetched twice.
● live 36 · A/DTrue daily breadth from the existing per-stock archive: each covered equity contributes +1 when its close rises, −1 when it falls, or 0 unchanged; net advances = advancers − decliners and the ADL cumulatively sums that net count. Includes 21-session and 50-session ADL averages, selectable stored US / Nasdaq-listed / NYSE-listed baskets, exact eligible and covered counts, and SPX, NAS100, NVDA, SKHY and 000660.KS price overlays. Full /basics interaction set: independent opacity, bold references, candles, Fill/Z-score, pinch/wheel zoom, pan, Y stretch, datelines, styled trend lines, resize, fullscreen, tape, and PNG/CSV/JSON export. Scope is explicit: current archived equities, not every exchange listing or historical membership.
● live 37 · fibtimePlace three collinear points on a fully interactive ES/NQ chart. Points 1–2 set the angle; point 3 sets distance along that ray. Segment lengths match the Fibonacci sequence through F108 (consecutive preferred, inverses and closer non-adjacent ratios allowed); points 4 and/or 5 project forward along the same line. After three points, drag the whole line. Copyable forecast: at this rate, the next time price hits this trend line will be DD-MM-YY… Data from the /basics store (Massive + Yahoo, single writer).
● live 37b · pentagramThree triangles: Gold / USD basket / Oil, Oil / Gold / NAS100, and Oil / Gold / SPX. Each vertex is one market; the floating point is the vol-scaled relative move over day · week · month · quarter · year (1–100 from each market’s own 30-session realised vol; equal scores sit at the centre). A fading trail shows the path of the equilibrium point. Layout: 1–3 triangles per row (stacks on mobile). Real-time from the /basics store (single writer).
● live 37c · volrsiGold · USD · oil · ES · NQ · RTY · YM DJI plus NVDA · TSLA · SKHY · SK Hynix KR · EWY (BlackRock Korea ETF). Section 1 is as comparable as price series get: log return ÷ each name’s own rolling realised σ, then cumulative from the window start (all lines origin 0 — units and vol stripped). Section 2 is Wilder RSI only (7/14/21), 30/70 guides, no vol adjust. Snapshot table ranks both. Data from the /basics store (Massive 1m tip).
● live 37d · svlThe real positioning answer, measured and stated as a percentile of its own history. CFTC leveraged funds (the actual hedge-fund / CTA category, weekly, 20y) and non-commercial net (large speculators, back to 1997) for E-mini S&P 500 · Nasdaq 100 · Russell 2000 · VIX, counted both in contracts and as a share of open interest. Plus FINRA twice-monthly open short interest per stock (split-flagged, days-to-cover). Keeps the original daily short-sale volume chart — now captioned with why it is a tape statistic and not positioning: the median day is ~53% short-marked because most of it is market-maker liquidity. SPX / NAS100 own-scale backdrops throughout.
● live 38 · pivotsA port of /apps/pivots onto the paid feed stack, with the mechanic intact: five families (Traditional · Fibonacci · Camarilla · Woodie · DeMark), five timeframes (daily → annual), PP plus R1–R4 and S1–S4, the proximity ladder and the Pine Script v6 exporter — all ported formula for formula and A/B verified against the original to four decimals. What changed is underneath: periods now come from the exchange calendar, so a cash pivot uses the 09:30–16:00 New York session and a futures pivot the CME trade date, and both are drawn at once, each against its own price line. Every period is drawn with the levels that governed it, so a level is a time series instead of one horizontal line, with the next period’s set armed dashed the moment the current one closes and an optional developing set that moves with the live feed. ES/NQ real time via Massive Futures Advanced through the /basics store (single writer, no duplicate fetch); ^GSPC/^NDX from Yahoo into this app’s own record-first store, continued after hours from the future with the basis flagged. Full /basics chart contract: corner chip cluster, measurement tool, real fullscreen, spliced axis, session classifier, stretch tape, dense tooltips, mobile-first.
● live 39 · regimeTwo regime diagnostics that need no dealer-positioning feed — both come out of S&P 500 price alone, so they are free and unambiguous. Signal 1, gamma regime from realised range: the rolling 10-session median absolute daily move read against the two ranges Imran Lakha quotes — 0.50–0.60%/day with dealers long gamma, 2–3%/day with dealers short — and the gap between them reported as an explicit transitional state rather than a fudged boundary. His confirming tell rides as a genuinely second dimension: net displacement against path travelled (a Kaufman efficiency ratio), because big bars going nowhere is short gamma while big bars going somewhere is just a trend — plotted together on a 2-D state map.
Signal 2, volatility carry with two estimators: realised vol by close-to-close AND by range (Parkinson primary, Garman-Klass beside it, never blended), on a fast 10-session and a standard 21-session window against VIX9D and VIX. The point is the filter: when range-based realised runs hot versus close-to-close, any “vol is rich” reading is suppressed on screen as an estimator artifact. No hardcoded VIX floor anywhere — every gate is a computed percentile of the metric’s own trailing five years. History back to 1970 (14,267 sessions) for the range reads, 1990 for the VIX carry, with an audited fix for the S&P index’s synthetic daily open. Conditioning variables, not entry triggers: the copy says what state price is in and how it behaves there, and names no side. Record-first store + daily recorder; VIX/VIX9D from CBOE’s own daily file, S&P OHLC from Yahoo. ● live 39b · flowEvery other options read in the suite infers intent from unsigned aggregates — put/call volume, premium $,
open interest. None of them can tell a retail call-buying frenzy apart from institutions distributing into it:
both print as “call volume up”. This lens signs the tape by aggressor. Each OPRA print is matched to the
prevailing NBBO at its nanosecond sip_timestamp and classified buy-initiated vs sell-initiated
(Lee-Ready adapted to options; at-mid prints get an explicit tick-test fallback rather than being dropped or silently
called buys). Output is buy premium $ vs sell premium $ — never a lone net — per DTE cohort
(0-1d · 2-5d · 6-20d · 21-60d · 60d+) and per call/put, plus session options CVD and an aggressor-imbalance
percentile against its own history.
One page that answers whether the data underneath every other card is alive. Built from six real outages that each
ran for about a month while passing every easy test: apps/indicators published over FTPS to a
decommissioned host and never checked the return code, so the nightly task reported success while
uploading nothing; apps2/pegrok’s builder was never deployed, so no cron could ever have run it;
apps/oracle’s recorder existed but sat in no crontab; apps/board served
HTTP 200 with valid JSON while every row was {ok:false, thin:true}; a vendor 403’d the
datacenter while answering fine from another host; and a crontab lost 70 jobs at once. So three independent
tests run per store.
The S&P 500 has an implied volatility quoted at four horizons at once — tomorrow (VIX1D), nine days (VIX9D), one month (VIX), three months (VIX3M). Every other vol read in the suite banks those as levels; the shape they make together existed nowhere as a signal. This lens reads the curve: the six slopes between those four nodes, plus five cross-asset legs (VVIX, SKEW, VXN, GVZ, OVX — each divided by VIX, so they read as “how many times S&P vol”). Curve state is pure ordering — contango, backwardation, or the honest third answer mixed for a kinked curve, never rounded to the nearest tidy label — with the exact segments that are inverted named, and a count of how many of the three are.
There is no tuned threshold and no hardcoded volatility floor anywhere in this app. Inversion is stated as the arithmetic fact that a ratio crossed 1.0, always beside that segment’s own-history percentile; the “floor” is replaced by VIX’s own percentile against 9,200 days, because the trader this framework comes from has quoted 11-12, 15-16 and 20-25 as the floor in different regimes — mutually inconsistent by his own testimony, so a constant would be false precision. Deep daily history is accretive (a day observed once is never dropped, because the upstream history tree is externally pruned) and the live recorder banks the whole curve every 15 minutes — the finest grid on which all four nodes actually publish — because one observation a day cannot see an inversion form and un-form inside a session. SKEW is carried at daily cadence and never forward-filled onto the intraday grid; the table states every leg’s true bar age. Conditioning variables only: nothing here names a side. ● live 42 · surfaceTwo shape reads on the SPX option surface, both from Imran Lakha, neither of which existed in the suite because nothing here measured skew at a constant tenor. Signal 1, the three-month skew: he singles out the 3-month line as “the most informative” and dismisses 1-week and 1-month as “quite volatile”, so three months is the headline and the other two are drawn only as confirmation.
The convention is stated, not assumed — a 25-delta risk reversal, put IV at |Δ| 0.25 minus call IV at Δ 0.25, each leg interpolated in delta across that expiry’s own ladder rather than snapped to the nearest listed strike, then each leg interpolated to exactly 91 days in total variance between the two bracketing expiries (never an interpolation of the skew itself, which would mix two time-scalings). Delta not moneyness, because a 10%-away strike is a 3σ event at one week and 0.7σ at three months — the very comparison the signal is about. The headline is the normalised form (skew ÷ ATM), since a raw vol-point skew widens mechanically when the whole surface lifts. And the signal is the divergence, not the level: skew bid for puts while spot makes highs, scored as a co-occurrence and, more strictly, as the residual of a rolling 5-year regression of skew on the index’s distance from its one-year high. Signal 2, the convexity premium: a VIX-style index is a variance-swap rate integrating every listed strike while the at-the-money uses one point of the surface, so the gap between them is the price of the wings — computed as a term structure against all four CBOE tenors (VIX1D · VIX9D · VIX · VIX3M), which says whether the wing bid is a front-end event or a whole-curve condition. His “usually 3–3.5, not normally 5” appears only as labelled reference lines; every gate is a percentile of the measure’s own record. The dealer-short-the-wings read is marked an inference, because the same open interest is equally consistent with a client put spread. Chain via the sharedI:SPX
index reader (spot from put-call parity) asking for five monthly OPEX expiries — which is what makes 91 days reachable at
all. Honest about its own age: there is no historical option surface in this stack, so the live measures are recorded
forward and say ramping until they have a distribution, while CBOE’s ^SKEW (9,139 sessions from 1990) carries
the divergence meanwhile and is labelled a proxy everywhere it appears. Conditioning variables, not entry triggers.
● live
43 · gaps
/health answers "is the pipeline running right now". This answers "what history do we not have, why, can we get it back, and what is it costing us." health's unit of truth is the latest write, so it is deliberately blind to the past — a store can be perfectly fresh at the tail and be full of holes behind it, and health will correctly call that green. This walks the whole span of every store health knows about and reports the gaps inside the range: expected days against observed days on each store's own calendar, taken from its own crontab line rather than guessed, with NYSE closures excluded so a Saturday is never a fault.
It does not re-inventory anything — the 64 stores, the cadence provenance, the crontab reader and the market calendar are all read live fromhealth/inventory.php, because a
second copy of that list is exactly the drift this suite keeps getting bitten by.
The ranking is the product: a hole in a store nothing reads is noise; a hole holding a registered metric below
a2_dist's minObs 60 — so an app is rendering null percentiles right now — is the headline, and
every store carries the calendar-day count until it crosses that line. Each gap gets a cause (never collected ·
pruned by rotation · no scheduled writer · off-host writer · page-load accrual · shipped recently) and a concrete
recovery route with its cost: OPRA flat files are already wired, Yahoo caps 1m at 7 days / 5m at 60 / 1h at 730,
OCC and FRED are free and deep, and a Databento re-request bills as a new purchase. Two things a single-store view
cannot see are computed here: a day missing from half the stores at once is named one host outage instead of thirty
separate bugs, and seed depth sitting on disk that metrics.json never declares is flagged — hist.php cannot merge
what the registry does not name. Where the filesystem genuinely cannot measure coverage (a directory of per-symbol builds
tells you only when a file was written) it says not measurable and why, rather than printing a fabricated
percentage. manual-by-design, retired and shipped-today are first-class states, not failures — a store one day old
holding one row is correct. The inventory is itself a record-first store, so whether a hole is opening or closing is
answerable next month instead of re-guessed.
● live
44 · continuation
The S&P and the VIX normally move opposite ways — measured here at a median correlation of −0.81, the leverage effect plus the fact that index puts are the hedging instrument, so a rally releases hedges. When they rise together, upside convexity is being bought and a dealer short those calls hedges into strength: the self-reinforcing melt-up. This lens measures that state, how rare it is, and — the part that makes it a lens rather than a dashboard — what states like it have actually preceded across 9,183 sessions with a complete forward month (1990→).
Four facts, never fused: today's quadrant (sign of the return × sign of ΔVIX, with an explicit FLAT fifth state), the share of the last 21 sessions in each quadrant, the 21-session correlation with its 10- and 63-session companions beside it, and that correlation's own-history percentile — because −0.8 is ordinary, not bearish, and only the percentile makes it readable. ΔVIX is in points, never percent: VIX is already a percentage, so a pct-change of it would weight a 12-handle day three times harder than a 40-handle one. The conditional study is the product. 70 states declared up front — quadrant, quadrant-share terciles, correlation quintiles, path-efficiency terciles, VIX level, term-structure inversion and the crosses those imply — each measured for forward return at +1/3/5/10/21, forward run length (consecutive same-sign closes), longest streak inside the month, forward path efficiency (did the multi-day move go anywhere or just thrash), forward drawup/drawdown and the size of the move. Every cell carries n, every conditional number is shown beside the unconditional base rate, and every effect size divides n by the forward horizon first — two adjacent 21-day windows share twenty of their twenty-one days, so 1,817 sessions is really 86 independent months, and that is the number the significance rests on. Every bucket is reported, including the boring ones: a single spot-up-vol-up day does essentially nothing, and path efficiency — the obvious continuation prior — does not predict forward persistence at all. What does separate: a decoupled quarter (63-session ρ in its least-negative 5%) precedes longer one-way runs but smaller moves, and an inverted VIX term structure is the single strongest condition for a big multi-day move in either direction. Pumps and dumps are measured separately and with equal weight, and the bearish side is deliberately not built as the mirror of the bullish one: the melt-up shape is unusual because it inverts the normal relationship, while its literal negation is simply the ordinary decline. Two bearish hypotheses go through the same study with their own falsifying controls — a decline nobody is hedging, and a disorderly liquidation confirmed by the vol structure — and a measured level-matched control pair separates what the curve contributes from what the volatility level does (they correlate +0.69, so the naive reading is a confound). The answer is a strong negative worth having: there is no clean multi-day dump tell here. Only 0.07% of months are 60%+ path-efficient and net down and no state raises it materially, against a 3.25% clean-pump rate — a 46× asymmetry. Several states do raise the odds of a deep drawdown; none raise the continuity. Downside arrives as violent chop, not as a smooth slide. Centrepiece is a 2-D spot↔vol map with a rarity grid built from the full 36-year history, so "unusual" is literal. Dealer context (GEX, flip distance, put/call, skew, and now vanna + charm) is on the page raw and explicitly unscored — 22 sessions against the 60 the distribution contract needs — and gates nothing. Record-first store, weekly seed rebuild, zero vendor calls: every input is a store this suite already banks. Conditioning variables only; the page names no side. ● live 45 · basisThe front-month future and the index it settles to are the same exposure priced twice. The gap between them is not an opinion about direction — it is the market quoting what it costs to carry that exposure with borrowed money instead of owning the shares. Sign convention stated once and never restated differently: basis = cash − future, so negative means the future trades above cash. Annualised over the days left on the contract it becomes an implied net cost of carry — financing minus the dividend yield, which is what an index future actually prices, not an interest rate — and it is shown beside the 10-year, never blended into it, because a rich basis against cheap funding is a different fact from a rich basis against dear funding.
Two pairs, never merged: S&P 500 / ES=F and Nasdaq 100 / NQ=F, 6,281 matched sessions from 2001. The roll is the thing that silently ruins this app, so it is measured rather than assumed. These are continuous front-month series and must switch contracts four times a year; the CME convention would put that switch eight days before expiry, and the data says otherwise — median day-over-day basis move stays at an ordinary 6–9 bp straight through the supposed roll week, then jumps to 40–54 bp on expiry Friday and 64–71 bp on the first session after it. The series holds the front contract to expiration. So every session carries the contract it is priced off, roll sessions are flagged from contract identity rather than a guessed window, and the discontinuity is drawn, marked and excluded from every distribution — never smoothed away. The days-to-expiry floor is measured too: the implied rate divides by days left, and its inter-quartile range runs 45.5 points at one day against 3.8 at twenty-one, so below 15 days the number is a data gap rather than a rate. Differencing two daily closes is not automatically a basis — the index closes at 16:00 New York and the future trades to 17:00 — so both are shipped: the deep 25-year series, and an exact one pairing the cash close with the futures print at the same 16:00 instant. The clock error between them is served as its own series and measures 1.5 bp at the median against a basis that ranges to a hundred, which is what licenses the deep series carrying the statistics. Half-day sessions get the actual close hour, not an assumed one. Because the future trades ~23h and the index prints 6.5h, the move is reported in parts — overnight leg, cash-session leg, total beside them — with the vendor's synthetic index opens blanked rather than counted as a flat overnight. The fair-value gap is a scatter of the cash opening gap against what the future did overnight: the fitted hand-over is 0.82 for the S&P and 0.98 for the Nasdaq, and what does not arrive at the open is named as a residual instead of being assigned to either leg. A contract-cycle scatter plots the basis against days left, so the calendar confound is visible rather than argued about. Conditional forward study over three conditioners × ten deciles × three horizons, buckets fixed up front, every bucket reported with n, the overlap-corrected standard error and the unconditional baseline pinned beside it — and every bucket is reported including the eight in ten that say nothing. What survives is narrow and it shows up on both indices, which is what makes it worth printing: only the cheapest-carry decile — the future at its widest discount to fair — separates from the base rate, and it separates downward. Nasdaq, carry-minus-10-year bottom decile (n=295 across 20 distinct years, so not one crisis): the index finished higher 46.1% of the time over the next session against a 55.4% base rate, 44.1% vs 58.1% over three and 46.1% vs 59.2% over five — 3.2, 2.8 and 2.0 standard errors apart, the only bucket in the whole study below baseline at every horizon. The S&P version of the same state is real but one session deep (48.1% vs 54.9%) and gone by the third. The rich end is weaker and roughly symmetric. Everything else, including the entire top half of the raw basis, is inside two standard errors of doing nothing. Record-first, zero vendor calls — and the basis definition is proved bit-for-bit against the oneapps2/pivots already uses over 4,295 shared timestamps, with
the build refusing to write if that ever fails.
● live
46 · omniverse
The exhaustive cross-metric explorer for the obscure commonalities and anti-correlations between two market regimes — two pumps, two dumps, a pump vs a dump, accumulation vs distribution.
Reads a purpose-built 180-column daily feature matrix (176 selectable metrics; 2026, 9 OPRA option roots + a decomposable MARKET aggregate) where every metric is banked three honest ways where it applies: all nine greeks as ATM-front, OI-weighted dealer exposure (GEX · net vanna · charm · volga · veta), and chain-total flow — kept distinct, never fused — beside put/call (volume · premium · OI), open interest, gamma walls & the local-γ regime read, 0DTE, IV term structure & 25-delta skew, the full VIX complex and its cross-asset ratios, VIX option open interest BY STRIKE (where the crash hedge is actually parked — centre of gravity of the call and put books vs spot, the biggest cluster and its distance, the OTM call share), ES/NQ futures basis and futures open interest, credit spreads, rates volatility, implied correlation, leveraged-ETF positioning, FINRA dark-pool share and the FRED rate. Pick two windows (preset regimes or free dates); every metric is compared across them by level (percentile + z), slope, acceleration, threshold-crossings and path-shape, in a vertically stacked A-over-B table with a dual encoding — font-size = how big the move is, colour = its direction — so a steep move stays big even at an average level and acceleration is visible at a glance. A mechanical finder ranks the strongest shared signatures and the strongest divergences; a correlation + A−B diff canvas heatmap surfaces which metric pairs stayed coupled vs decoupled; lead-lag tests which moved first; and SPX + NAS100 are drawn for both windows aligned to a common t=0 so two pumps'/dumps' shapes compare directly. Every metric can also be ranked through five lenses — most retail / dealer / institutional / directional-bias surfacing (four attribution questions: whose positioning does this reveal?) and most unusual since Jan (a separate extremity question) — where the score is a published, arguable weight times how far from normal the metric is right now, so a metric sitting at its own median ranks low no matter how well it reveals a crowd; metrics with no weight on a lens are excluded and counted, never ranked last. Every correlation carries its n, the signed-mirror (a metric that moves alike in a pump and a dump is volatility, not direction) is flagged inline, and the recency confound is stated on the page. Describes what happened; never predicts. ● live 47 · omniverse / analysisA real-time conditions dashboard built on the Imran Lakha (Options Insight, @options_insight) options framework — his lens; our data and implementation, with no involvement or endorsement by him. His method starts with two boxes, never one: a view on spot (up / down / sideways / no view) and a view on vol (rich / cheap / fairly priced / no view), with the structure family following from the pair.
This page resolves both boxes at three grains — 1 hour (live futures hour + the 5-minute options snapshot re-solved with our own solver + differenced chain flow, against an 888-hour baseline), 1 day (daily closes against 2026's sessions) and 1 week (the last complete ISO week, constructed with an explicit unit rule per metric — flows sum, returns compound, levels average, ratios rebuilt from aggregated components — against a 30-week baseline) — and reports whether the three agree, since each had to reach its answer against a different baseline. The key metric sits on top: cash vs futures vs the options-implied forward, the last derived by put–call parity (F = K + (C−P)·erT, an identity, no vol model), with the futures basis and the options carry compared as annualised carries and the ATM straddle shown as the market's own stated expected move. A separate magnitude lens answers the different question of how far and for how long: sustained runs are detected mechanically across 2026 with a swept percentage zigzag, each run's opening sessions fingerprinted, and today scored against them — with the false-positive counts shown raw and non-discriminating tells listed as dropped. All 78 rules carry a disposition (vote / detector / framing / not-computable-with-reason) and each vote shows its rule number; where the source rules file was wrong we say so (R020's absolute gamma buckets are replaced by his actual front-implied breakeven rule). Dealer gamma deliberately casts no direction vote — R010 says it is not directional — and a coverage harness asserts every reachable state resolves to written copy. Fully server-rendered for no-JS and AI fetchers. Describes what is priced; never predicts.
● live
48 · gex
One question, answered in plain words before any chart: is the market set up to move in one direction, or to chop? Long dealer gamma means market-makers must sell rallies and buy dips to stay hedged — pinning, failed breakouts, small ranges, the jagged broken-shard tape. Short dealer gamma means they chase — trend, follow-through, gaps, the blast tape. The regime is read from the net gamma inside ±1% of price (not the γ-flip, which is a level and much weaker than it is usually given credit for), with how fast it decays — the share of the book dying at the front expiry — as the differentiating read.
Built on the full options chain captured every 5 minutes, weighted by open interest only, never volume: a missing OI is excluded and counted, never zeroed, and per-strike netted contract counts are primary because dollar-gamma at one strike balloons as price approaches and collapses as it leaves. Shows where the walls sit now, how they migrated through the session, and the forward book with today's expiry stripped out. It also publishes the measurement of its own thesis — and that thesis did not hold: raw, negative gamma does precede bigger next-session ranges, but once the range is measured in units of what options already charged for it the relationship vanishes, path-shape shows nothing, and failed breakouts run the wrong way. Major support and resistance are labelled explicitly as a spot-anchored ladder — top 3 concentrations each side, not just the headline wall — with the γ-flip kept separate as a regime boundary and the gamma centre as a magnet, never conflated with a barrier. Those labels carry their own measurement too: reaching the call wall, price closed back below it just 37.2% of the time (n=215) against 38.0% for an arbitrary level the same distance away, so the ladder ranks by concentration and says so. Because walls come from settled open interest they legitimately hold flat after the close, so live ES / NQ / RTY futures are overlaid against the frozen levels — anchored to the cash close so the basis and index/ETF ratio cancel — answering whether price is drifting toward or away from a wall overnight, with both clocks shown side by side. Directional bias is deliberately subordinate: no lean is the standing answer, because no asymmetry beat the base rate of sessions that simply closed up. Describes positioning; never predicts. ● live 49 · driveThe question every trend follower actually has, answered against a pre-registered backtest that was frozen before anything was tested. Eight horizon columns — 1d · 2d · 3d · 4d · 5d · 2w · 3w · 1m — each carrying the base rate (how often a one-way drive has actually happened since 1970) next to what a reshuffle of the same market's own returns does, so no percentage ever appears without the number it has to beat.
Under it, a predictor × horizon grid: eighteen candidates — three Hurst estimators at three windows, variance ratios, autocorrelation, ADX, the Kaufman efficiency ratio, realised-vol regime, momentum — each cell a dot on a fixed effect-size axis with the null band drawn in grey, at three heights for all-history, fitted-to-2014 and scored-on-2015-onward, so a sign flip between the last two is visible as a line. Most cells read "no edge", and that is the finding. The headline results it publishes rather than buries: the Hurst exponent flips sign in 98 of 189 instrument × window × estimator combinations out of sample (51.9%, a coin toss) and measures −0.007 after 2000; the conventional H > 0.5 threshold is meaningless — a memoryless market reads above it 78–89% of the time, and the honest boundaries are 0.815 / 0.734 / 0.682 at 64/128/256-day windows; the S&P is less one-directional than its own reshuffle at 5, 10 and 20 days; every conventional trend gauge points negative; of 663 tests 17 survive correction, 14 are realised-volatility measures, all 17 point negative, and none is on the S&P. Dealer gamma renders as "not testable yet" with the date it becomes answerable — 113 rows, ~10 independent observations — never as a number. A positive control proves the pipeline detects effects down to |rho| 0.027, which is what makes this a strong null rather than a weak test. Every figure traces to a named results file; the page computes no statistics of its own and issues no forecasts. ● live 50 · baselineThe owner's thesis, instrumented: stocks drift up by default (401k paychecks + money printing are a standing, price-insensitive bid) — so the question is never "did stocks rise" but how far ahead of the fuel is the pump running? Eleven levers — M2 (weekly pulse + YoY) · DXY · CPI (headline + core) · Treasury flows (Fed SOMA + foreign custody) · 401k bid (modeled + ICI weekly) · rates (Fed funds + 10Y) — each z-scored at its NATIVE cadence and mixed as one BASELINE with 11 signed weight sliders (−2…+2; negative INVERTS a lever) that recompute everything live, client-side. The headline is the PUMP-EXTREMITY GAUGE: divergence D = market impulse z − baseline B, scored as percentile vs its own trailing 4y, drawn as a TradingView-style oscillator ABOVE the raw SPX/NAS100 price band (flush pair, shared x-axis), with persistent ≥p95/≤p5 episodes shaded as start→stop ranges. Plus the mandatory tape+line pair (weekly force tape, 6 colour modes + all 11 lever z-lines vs SPX/NAS baselines), rolling 13w/52w correlation, a five-window comparison table (1w/1m/3m/1y/4y), custom symbols, and 70 served state-dependent plain-words sentences. Active weights burned into every PNG. Descriptive only — never predicts.
● live 51 · gdpOfficial ratios, no invented numerators: nonfinancial corporate equities ÷ GDP (Fed Z.1 NCBEILQ027S / BEA GDP) · after-tax corporate profits ÷ GDP (CP/GDP) · household equities as % of household financial assets (Fed BOGZ1FL153064486Q, already a percent) · FINRA margin debt ÷ GDP · Magnificent 7 TTM GAAP net income ÷ GDP (SEC EDGAR 10-Q NetIncomeLoss for AAPL MSFT NVDA AMZN GOOGL META TSLA — missing if any name lacks four quarters) · gross federal debt ÷ GDP (FRED GFDEGDQ188S, already a percent). Quarterly prints are held as steps, never interpolated into fake monthly noise. Each tile is the latest print plus its own-history percentile. The page also measures its own thesis: in-sample quintiles of each ratio versus the next 5- and 10-year S&P 500 price return (dividends excluded, and said so). Describes how large, profitable, owned, levered, concentrated and fiscally stretched the market/economy pair is. Never predicts.
S&P 500 · NAS100 · Russell 2000 · WTI crude · US 30Y T-Bond · Gold — all six as continuous front-month futures (ES1! NQ1! RTY1! CL1! ZB1! GC1!), Massive realtime + Yahoo history from the /basics store (single writer). Portrait shows 3 charts per screen; landscape shows one full-screen chart with edge-arrow / edge-swipe flip; desktop is a 3×2 grid. Charts are drag-sortable (long-press to lift). Seven timeframes 1h·1d·1w·1m·1q·1y·5y at the max candle resolution each window allows (1m up to a week, 5m month, 1h year, 1d 5y). Candles are pixel-crisp — fillRect-only, 1px columns, zero gap, no antialiasing — and every control is a 30%-transparent in-canvas overlay: pan, pinch/wheel zoom, Y-scale, crosshair tooltip with Copy, drawing tools (horizontal ray · vertical time callout · % measuring box, persisted per symbol), fullscreen, opaque-background PNG, JSON dump. Live edge on the &last= 5s Massive fast lane.
● live 53 · demarkCME E-mini S&P (ES=F) and Nasdaq 100 (NQ=F) futures, 10 years of daily, every candle size. TD Combo Version I (the multi-day spike print — 13 bars, four strict conditions) and TD Sequential 9-13 (perfected 9, countdown 13 vs 8, TDST, risk level, 12-bar stale clock). Same bar-count rules on 1m and on weekly. Two charts, shared dateline and zoom, independent height and candle width. Intraday from the /basics store; daily/weekly 10y from Yahoo. A 9 is a location, not a trade. Never predicts.
● live 54 · parabolACCEL_UP only if the last 2–3 confirmed high-to-high legs and the last 2–3 low-to-low legs are both strictly steepening, all slopes positive. One-sided is NONE (the tile names which side failed). Onset dateline = first confirmation of that state — earlier than a human naming a parabola, not a forecast. No moving average, no curve-fit fallback. Never predicts.
● live 55 · obvFinds and ranks on-balance-volume divergences — price making a new swing extreme while the running volume flow refuses to confirm it — across all 21 /basics assets that report real traded volume (6 CME futures ES·NQ·RTY·YM·ZB·GC + CL + BTC + 11 US equities/ETFs + SK Hynix KR + USDT/KRW) × 5 windows (1d·1w·1m·1q·1y). Every variable of the OBV equation is a slider: ε deadband, k close-lookback, α volume exponent, V÷SMA(n) normalization, SMA/EMA smoothing — classic textbook OBV at slider-zero (parity test-enforced) — plus detector sliders (pivot width, scan window, min strength, classic/hidden). All recomputed live client-side. Sortable per-column-heat board; click a row → price candles + OBV pane with the divergence drawn on both, SPX/NAS100 baselines, P/C line, click-any-line-to-explain. Descriptive only — never predicts.
● live 56 · rsiobvRuns RSI (Wilder-14) and OBV on all 21 volume-bearing assets × 6 windows and surfaces when momentum and volume combine or contradict: same-direction divergences (confirmed), opposing divergences (conflict), and joint percentile extremes — same side (blowoff / capitulation watch) or opposite (hollow rally / conviction-less selling). Five rules frozen BEFORE backtesting, then scored over ~25y daily × 20 assets against each asset's own tops/bottoms — the on-page table shows hit vs base, lift, n and era-consistency including the rules that showed no edge (R4 joint extremes: lift 2–3×, consistent all eras, ~900 episodes each). Three-pane detail (candles + RSI + OBV, extremes shaded, both divergences drawn), full coverage strip, res selector, Telegram alerts labeled [RSI+OBV] citing each rule's own lift. Never predicts.
● live 57 · overnightFor the top 20 most popular US-listed stocks right now (measured: highest 20-session average dollar volume from the house 545-ticker store), splits every close-to-close move into its two halves: what $1 became held only overnight (buy each close, sell the next open — the "overnight effect" split) vs held only intraday (buy open, sell close), plus the best fixed clock window (in-sample, modeled — 64 candidate buy→sell time ranges on 1-hour boundaries, runner-up + era-split shown) over 1q · 1y · 2y · 3y. Strategy chart is exactly 3 lines; a separate SPX/NAS100 chart shows each index's most bullish vs most bearish hours.
Session map: a grid of the 20 names plus a pinned market-wide pooled row against the 11 parts of the session — overnight, each hour block, morning, afternoon, whole day — showing how often each part closed higher, with a tunable line (default 55%, mirrored below) and the same cells as average move per session so frequency and size can be read apart. Its banner states the arithmetic for whatever range is chosen: cell count, the up-rate that reaches the 5-in-100 level at that session count and after dividing across 220 tests, and how many cells chance alone flags — and it makes no judgement about which range belongs on screen. Runs on any custom from/to range as well as the trailing lookbacks, with presets built from the data (calendar years, SPY drawdown spans). Segment definitions and numbers match the house research script exactly, asserted against a frozen parity fixture. Judgement is driven by criteria you give it: a criteria panel takes a round-trip cost, a minimum edge, what you are optimising, a confidence stance and a free-text note of what you are actually measuring — and only then does every cell carry a plain-words verdict naming the criterion that drove it. The default stance is the owner's own: anomalous versus random, both tails, scored by z against each name's own average block (drift-adjusted, with a vs-zero toggle), so more gain than random and more loss than random both surface with z, two-sided p and n. Several named time frames can be kept and every cell counted across them, without ranking any of them. Switch judging off and it returns to stating what happened and judging nothing. Reads hourly boundaries; 10-minute blocks live in /apps2/hourality. Ranked board with positional window strips + overnight win rate; overnight-edge tape with six coloring modes. Hyperliquid & Ostium: every tradable equity, last 6 months, five ways $1 was held — overnight · intraday · best hours · worst hours · hold only, never sell. Green = $1 grew, red = $1 shrank; more opaque = a bigger move. Price returns only — dividends (which land overnight), spreads and fees excluded, stated on-page. Never predicts. ● live 59 · ostiumTakes /apps2/hourality's range scan and asks the only question that
matters about it: after Ostium's real round-trip cost, does any of it make money? Shorts
the most bearish and longs the most bullish 10-minute windows on the 35 Ostium-tradeable
names, at the pair's maximum leverage, on a $10 margin per bet. Window statistics are
hourality's own engine.js output — nothing is recomputed. Costs are
measured, never assumed: fee from the Ostium subgraph and confirmed on real executed
orders, spread from the live builder feed, gas from real Arbitrum receipts, and the venue's real
09:31–15:59 stock hours with day-trading shut at 15:45.
Takes the 86 tickers tradeable on Hyperliquid and Ostium (the union from the house session-bias research — the two venues pick WHICH names are worth studying, they are not the price source) and studies their regular US market tape, 09:30–16:00 ET only: no pre-market, no after-hours, no futures. Each session is cut into 10-minute blocks of the New York clock, preceded by the last two hours of the previous session with a hard drawn break between them — 51 blocks at 10 minutes, 17 at 30.
The headline analytic is a contiguous RANGE search: all 1,326 possible runs of blocks are scored across the sessions in each window you define, against a length-matched baseline of that name’s own average block, so what surfaces is deviation rather than drift and a long run is not anomalous just for being long. z is two-sided, so more gain than usual and more loss than usual rank equally. You define the date windows — name as many as six — and every one scores the same grid, with a pane that lines them all up so a range that tops all your windows and one that tops a single window are both visible. Three flush canvases: the stacked tape (one segment per session, best and worst range shaded with their runners-up), the session×block heatmap with six colouring modes, and the windows×block z grid. SPY and QQQ cash ETFs on their own scale. Sortable tables of every candidate range and of every ticker by biggest anomaly, most bullish window or most bullish overall. Prices seeded from Yahoo 5-minute bars and deepened and extended by our own Alpaca lane, every bar tagged with its source; the store accrues, so the history only grows. Reports what was measured — n sessions, dates, counts past |z|=2 next to the count chance alone produces. Never predicts. The macro end of the same pair is /apps2/overnight — the same trading day cut into session-level segments for the top 20 most-popular US names, over a deeper daily history; both pages link to each other top and bottom. Renamed from/apps2/tape on 2026-08-25 (owner: “should be apps2/hourality”); the old path redirects here. The round-the-clock counterpart is /apps2/hourality/futures.
● live
60 · hourality/futures
The same question as /apps2/hourality, asked of the S&P 500 future (ES) across its ENTIRE live CME session. A session opens 18:00 ET and runs to 17:00 ET the next day — about 23 hours — with the 17:00–18:00 halt excluded because nothing trades in it (measured: zero bars in that hour), and the week opening Sunday 18:00 and closing Friday 17:00, so Friday evening to Sunday evening is not a block. That is 138 ten-minute blocks per session and 9,591 candidate stretches (46 blocks / 1,081 at 30 minutes, 23 / 276 at 60).
The four phases of the day are shaded and named on every chart — Asia 18:00–03:00, Europe 03:00–09:30, US cash 09:30–16:00, Post-close 16:00–17:00 — with a bold delineation at each boundary, a per-band table (each band end to end, plus its strongest and weakest run inside), and a plain-language reading of what the four bands say together. Every piece of the range maths is hourality’s ownengine.js,
imported, not reimplemented; only the frame and the matrix builder differ, and the baseline is
scoped to all 23 hours because this frame counts every clock time exactly once. NQ, the
Nasdaq-100 future, rides on its own scale and is one click from being the chart (there is no SPX
baseline here — the subject already IS the S&P 500 future). Data is the existing
/apps2/basics ES/NQ lane read over the house loopback — no new vendor
lane — seeded by walking the upstream store backwards and accrued forward, so depth grows
past both the 60-day vendor cliff and the upstream 200-day retention. Futures and overnight
hours are the subject here by owner order 2026-08-26; the no-futures rule on the parent page is
scoped to the parent and still stands there. Never predicts.
● live
60b · hourality/world
The third lane of the Ostium board, built to the owner’s 2026-08-30 order (“group tickers by trading hours … use correct 10 min ranges for each of 3 groups”): the eleven names whose hours are neither the New York share day nor the 23-hour CME day — bitcoin and ether (24×7), AUD · EUR · GBP · NZD · USD/JPY (24×5), and DAX 40 · FTSE 100 · Hang Seng · Nikkei 225 (each venue’s own cash session). Every frame here is MEASURED from that instrument’s own stored tape, never taken from published exchange hours: a ten-minute column joins the grid only if it printed on at least half the sessions, and the rarer prints are named as strays rather than left as permanently empty columns.
Measured 2026-08-30: bitcoin and the currencies 144 columns with NO gap column at all (a market that never closes has nothing to jump across), DAX 09:00–17:20 Frankfurt and FTSE 08:00–16:20 London, 51 columns + one overnight column, Hang Seng 09:30–12:00 + 13:00–16:00 Hong Kong, 35 columns + a lunch column + an overnight column, Nikkei 09:00–11:30 + 12:30–15:20 Tokyo, 34 + 2. The four foreign indices are read on their own city clock, because Hong Kong and Tokyo keep no daylight saving and New York does — on a New York grid the same Hong Kong minute would slide by an hour every autumn. You may hold a stretch across a gap column; you may never start on one, because its opening price is a closing print no order can obtain. Every piece of the range maths is hourality’s ownengine.js, imported, not reimplemented. A third pane lays
every selected instrument on ONE New York clock beside the S&P 500 and Nasdaq 100
futures, so markets on completely different sessions become directly comparable and you can see
exactly when each one is awake. Data is the existing /apps2/basics lane over the
house loopback — no new vendor lane. Depth is uneven and always on screen: 59–94
sessions at ten minutes, 490–765 at sixty, printed on every row and every play.
Never predicts.
● live
61 · dayality
The third view of the same clock, and the coarsest: the trading week cut at every bell into ten pieces laid out in clock order — five overnight legs (the market is shut, last close to next open) alternating with five intraday legs (the market is open, that day’s own open to its own close). The axis starts at the weekend leg Friday close → Monday open and ends at Friday open → Friday close, so every price move you ever experience lands in exactly one of ten columns. Same 86-name board as /apps2/hourality, read off the house split-adjusted daily store — about five years deep, roughly 260 occurrences per slot, where hourality holds 400 sessions.
Holidays are not silent: a Monday holiday makes the leg Friday close → Tuesday open, so it is bucketed by the day it OPENS INTO, flagged, and counted — and that week’s Friday→Monday and Monday-session slots record a MISSING occurrence rather than a quiet zero. A leftover split-shaped jump (near a canonical split factor AND corroborated by volume moving the opposite way) breaks the chain there, is counted, and restarts every cumulative line instead of being multiplied through. Three nulls side by side — the name’s drift spread evenly over the ten slots, judged flat, or spread per minute of clock time so the ~65-hour weekend leg is handed ten times a 6.5-hour session’s share — each with Student t at n−1 df, not a normal approximation, so a name with 16 sessions is scored honestly. You pick the window, by preset or by typing two dates; the page reports its arithmetic and never ranks one window against another. Four flush canvases: the ten-slot profile with SPY/QQQ on the same basis-point scale, the board tape (88 names × 10 slots, seven colouring modes), what one dollar did holding only one slot against held-every-session, and the week-by-week tape that shows whether a finding was steady or came from one stretch. Carries the five context baselines (all day · overnight only · AM · PM · never sold) beside the found slots, the test count and chance baseline (880 readings; chance alone lands ~44 past the 5-in-100 line), and a hidden copyable ASCII dump of every number on the page. Ships a versioneddow export block
with a per-(ticker, session-date) segment map for a later combiner app. Costs are not
modelled and dividends are not in the prices — both stated next to every dollar figure.
Never predicts. Its siblings: /apps2/hourality (which minute of
the day) and /apps2/overnight (overnight vs during the day).
● live
62 · trend
The state the other three clock pages get conditioned on. Every bullish and every bearish stretch each of the same 86 Hyperliquid / Ostium names has been in, at 1 hour, 1 day, 1 week and 1 month bars, with the bar each trend started on and the bar each one ended on, and a sortable table saying which condition ended it and at what price. The rule is stated rather than hidden: a Donchian channel state machine — plain words, the trend turns up when price closes above its recent high and turns down when it closes below its recent low — and every knob (both lookbacks, the confirmation delay, the price-chain threshold) is a slider on the page and is stamped into every exported row.
Because it has knobs, the page says so: a strong result found downstream is partly a function of where they were set, and that is not covered by any multiple-testing arithmetic. The constraint the whole thing hangs on is no look-ahead. Every label is what the rule said as of the prior close — a Tuesday label uses no Tuesday bar — so a combiner can ask “on Tuesdays during a confirmed uptrend, what did the overnight span do?” without the answer being confirmatory fiction. That is proved by truncation, not asserted: the suites delete every bar from a session onward, recompute, and require the published label to be byte-identical, at all four bar sizes and on the real store; a second test runs the rule over a random walk whose future is independent of its past and measures that no state predicts the next bar. Price chains are cut where a corporate action is detected — at 1 hour against the split-adjusted daily store, which is a genuinely independent reading; at daily and coarser there is no second source, so big jumps are listed rather than cut and the blind spot is stated instead of papered over. Four flush canvases: price with every trend shaded start to end and the channel edges drawn, what $1 did in-uptrend-only against held every session, the whole-board state tape (six colouring modes), and the trend-length distribution with the running trend pinned on it. Carries the five context baselines, a hidden copyable ASCII dump, and the export contract atdata.php?machine=1 — one row per
(ticker, session) with the state, when it began, how many bars it has held and the exact rule
settings. Consumers must check v >= 1, never === 1.
Never predicts. Its siblings: /apps2/hourality,
/apps2/overnight and /apps2/dayality.
● live
63 · seasonality
Twenty-five years of sessions sorted into the slot of the year they fell in, for the S&P 500, Nasdaq 100, gold and crude plus the 86-name board. Four axes — calendar date, session of the year, day of the month, month. Because ~366 buckets are scanned at once, the test count and the number chance alone puts past the line sit beside every answer, and the effects that already have a name (turn of month, witching, Santa, holidays) are labelled as context, not sold as discoveries. Siblings: /apps2/hourality, /apps2/dayality, /apps2/trend.
● live 64 · robotOne row per tradeable ticker, grouped by the hours it actually trades (23-hour futures first, then US stocks, then bitcoin and ether, then currencies). Four cycle columns — option-expiry, week of the month, day of the week, time of day — each with its x-axis drawn once at the top of the group and a red line running the full height marking where we are in that cycle right now; the cells carry no numbers at all, every figure is in a hover tooltip in plain words. The week opens at the Friday close, so the first leg is Fri close → Mon open — the weekend hold, named rather than buried. Then RSI and OBV divergence (the /obv engine verbatim — knowable one candle after the slope endpoint) and the timeportal advice slot.
Reads /apps2/dayality, /apps2/hourality + futures, /apps2/weekality and /apps2/rsiobv — a leaf reader that re-implements none of their maths. Every clock time is New York market time (ET). The bot consumes the same objects at data.php?machine=1 (floor version-check, v ≥ 1). Judges nothing, forecasts nothing. ● live 65 · weekalityEvery session sorted into the week of its month. A week is five trading sessions
by default — the way a market participant counts one — with calendar weeks as the
stated alternative; week 5 is a partial block (measured: it exists in 205 of 299 months)
and is never folded into week 4. Both legs are in the frame, and the overnight that
crosses a month boundary lands in week 1 of the NEW month. Score per session or
per whole week, where the comparison is length-aware so a short week 5 is judged
against a short ordinary stretch. Every ticker on the /robot board is here,
grouped by the hours it trades — futures, US shares, foreign listings, currencies
and crypto, each stating its own day boundary, because an “open” and a
“close” are different things in each. Turn of the month is named as a published
effect, not rediscovered. Five flush canvases, the five context baselines, a hidden copyable
ASCII dump, and data.php?machine=1 (floor version-check, v ≥ 1) carrying
both week keys per session. Never predicts. Siblings:
/apps2/opexality, /apps2/seasonality,
/apps2/dayality, /apps2/hourality.
Twenty-five years cut into monthly options-expiry cycles and each part measured against the others: the 5 sessions after an expiry, the middle, the 5 before expiry week, the run-up, and expiry day itself — or the same cycle one session at a time (+1 … 0 … −1). The anchor is measured, not assumed: expiry is the third Friday, moved back to the Thursday when the equity market was shut, applied against a real US equity session calendar — 7 of 301 expiries moved (six Good Fridays and Juneteenth 2026, which was also a witching) and every one is listed. Quarterly witching is a labelled subset (all / quad / ordinary-monthly), never blended into an average. Complete cycles only — a session in an unfinished cycle is counted as a stub and bucketed nowhere.
Same board and same grouping as /apps2/weekality; gold, crude, currencies and crypto keep their own expiry calendars, so their rows say plainly that this is a cross-asset question. Five flush canvases, the five context baselines, a hidden copyable ASCII dump, anddata.php?machine=1 (floor version-check,
v ≥ 1). Never predicts.
● live
67 · earnings
A results day is a scheduled jump: the stock moves on its own news, several times its ordinary size, in a direction nobody is entitled to know. So this is a randomness and volatility-risk flag first — inside the window, a name's day-of-week and time-of-day statistics are about to be swamped. Per stock on the /robot board: next date (vendor estimates flagged as estimates, absent dates rendered absent), BMO/AMC timing where the vendor timestamp passes a measured rule, the risk window, and the measured report-day jump vs an ordinary day on that name's own tape. Deep history is detected from the tape by volume only — picking by price move would select big movers and inflate the answer — and merges only after leave-one-out validation clears its declared bar; below it, confirmed dates only with the small n printed. Funds, futures, currencies and crypto each say why they have none. Names no direction.
● live 68 · fomcalityEight scheduled decision days a year, and the market knows every date in advance. This reads each instrument by its signed position relative to the announcement and by the week of the cycle. The calendar is read, never assumed: 215 scheduled decisions 2001–2027 built from the Fed's own pages, each past date checked against its own statement URL — including the seven meetings of 2020, because the Fed itself marks March 17–18 cancelled, and the 17 unscheduled calls that carried a real statement. Measured on 25 years: the decision day itself runs +22 bp on the S&P and +34 bp on the Nasdaq 100. It also prints the bar that judgment must clear — 51 positions are scanned, so about 2.6 would clear the 5-in-100 line by chance alone, and the count that actually clears is at or below that. The much-discussed drift into the announcement is not visible on daily bars. Foreign venues that shut before 14:00 New York are read on the next session, because that is when their tape can react.
● live 69 · fundAckman, Druckenmiller, Citadel, Jane Street and a seed set. Each 13F is a quarter-end photo of long US stocks and options only. Adds, exits and size changes versus the previous photo, sortable and filterable. Citadel and Jane Street default to the largest slice — those books are inventory, not conviction. The newest quarter is already up to 45 days old.
● livePut it on any chart. Weight the ticker by M2 or inflation three ways (ratio · real-deflated · relative-perf), re-adjust price into real / M2-deflated terms, time-offset M2/CPI to test whether it LEADS price, and an accel/decel oscillator of M2 & inflation. Click to download the Pine v6 source as a .txt — paste into TradingView's Pine Editor.
↓ download .txt pine v6 · ↓ .txtKnow M2's DIRECTION before the lagged monthly print. Overlays the faster-releasing leads — weekly M2 (WM2NS), Fed net liquidity (WALCL − TGA − RRP, unit-scaled to billions), and H.8 bank deposits (~80-90% of M2) — each normalized (z-score / %-from-start) and time-offsettable so you can slide a lead forward to front-run the monthly M2 (M2SL) reference line. A nowcast up/down read shades the background GREEN when the leads are rising / RED when falling. Click to download the Pine v6 source as a .txt — paste into TradingView's Pine Editor.
↓ download .txt