Methodology, without the black box.

Price, spread and flow

Coinbase, Binance, OKX, Kraken, Bitfinex and Bybit each retain venue identity and independent clocks for price, Level-2 book and volume. Every field declares one timestamp basis: source event when the venue publishes it, received for a WebSocket payload without a venue clock, and retrieved for a REST response without an observation clock. Aggregate price is the conventional median of finite positive fresh venue prices—averaging the two center values when the count is even. Best bid and ask remain raw; a non-atomic crossed aggregate is flagged instead of being rewritten into a manufactured zero-spread quote. Aggressive flow normalizes the venue's taker direction and uses complete 250ms buckets over the exact trailing 60 seconds.

Full Level-2 normalization

REST snapshots retain up to 50 normalized bid and ask levels per venue; browser WebSocket books can maintain 50–100 levels where the venue publishes them. Levels are price-sorted, zero or invalid sizes are rejected, and updates are last-write-wins by source event time. Comparable live imbalance sums bid and ask liquidity within ±25 basis points of the venue midpoint. Aggregated and per-venue books preserve size and source rather than reducing the market to top-of-book.

Continuity contract

A reconnect never intentionally clears the last verified price, book, tape, conviction or key level. Price, book, volume, flow, derivative fields and each on-chain metric advance independently and only with a verified observation at least as new as the field already mounted. A ticker cannot refresh an old book; a liquidation cannot refresh old funding; one new on-chain family cannot relabel another. The browser restores its last valid snapshot before a new request completes. A warm edge isolate can also retain a last-good snapshot, but that server cache is not a durable cross-region database and is never represented as one. Retained values keep their original timestamps, are labeled and are excluded from “fresh” counts.

Freshness and conflict rules

The event-to-interface engineering target is under 500ms, not a universal latency guarantee. A venue event is treated as live below 1.5 seconds, degraded above 1.5 seconds and stale above 15 seconds. REST fallback snapshots target a ten-second refresh and a 15-second freshness window. Older observations cannot overwrite newer venue state. Binance buffers differential events during snapshot retrieval, discards updates at or before the snapshot ID, requires the first applicable event to bridge snapshot ID + 1, and resynchronizes on any later gap. OKX enforces prevSeqId → seqId continuity. Bybit honors snapshots and update-ID reset 1, then accepts only monotonically newer deltas. Kraken and Bitfinex public books are snapshot/delta normalized but this client does not independently recompute their optional venue checksum; that limitation remains explicit. Invalid price, size, time and sequence frames are quarantined; missing values are not invented.

Each factor strength is clamped to −1 through +1. A fully bearish factor subtracts half its weight and a fully bullish factor adds half. Fixed weights sum to 100:

Regime adjustment occurs before contribution: ranging multiplies trend by 0.75; high-volatility multiplies aggressive-flow and book-imbalance strengths by 0.70. Buckets are fixed at 0–39 sellers, 40–60 mixed and 61–100 buyers. FINITE does not publish a new live conviction score until verified flow and book inputs exist; during a disruption it retains the last verified score with its continuity label. Other unavailable public factors contribute zero exactly as disclosed rather than being inferred. The plain-English ten-minute market brief is a separate explanatory layer and cannot silently change the score.

Exact live normalizations

• Trend = (up timeframes − down timeframes) ÷ available 5m/1h/4h/1D timeframes.
• Aggressive flow = clamp((trailing-60s taker-buy share − 50) ÷ 20).
• Book = clamp((fresh ±25bp bid-depth share − 50) ÷ 25).
• Momentum = clamp(current 1-hour percentage change ÷ 3).
• Derivatives = clamp(−mean funding × 2500 + mean mark/index basis percent ÷ 1.5); stale or retained venues are excluded.
• Location = clamp((24-hour range position − 0.5) × 1.2).
• On-chain = clamp(−latest daily labeled exchange netflow BTC ÷ 5000).

Support and resistance are observed window lows and highs for 5m, 1h, 4h and 1D structures. Trend is up when price is above the fast exponential average and that average is above the slow average; down is the inverse; otherwise it is range.

The 7-day volume profile divides the observed price range into 32 bins and assigns each candle's volume to its typical price, (high + low + close) ÷ 3. Point of control is the center of the highest-volume bin. The value area is the smallest highest-volume set reaching 70% of observed volume; its lowest and highest included prices are VAL and VAH. Because the source is OHLC volume rather than trade-by-trade volume-at-price, the profile is a clearly labeled approximation.

Durable on-device replay

The full replay stores one verified frame every 30 seconds for up to seven days on the current device, including unified price, score, factor contributions, flow, book imbalance, key level and captured trades. It survives page refreshes and browser restarts, is quota-bounded, and never leaves the device. It is not a global historical tape: frames exist only while FINITE was open, and network gaps or unseen events are never reconstructed.

Historical structure proxy

The historical panel uses walk-forward candles and measures the subsequent four-hour close-to-close move, normalized to the available interval when a fallback source is active. Forward outcome windows do not overlap. “Hit” means positive for 61–100, negative for 0–39 and within ±1.5% for 40–60. It reports sample size, hit rate, false-signal rate, mean and correctly even-sample median move, probability of an upward outcome and a 95% Wilson interval.

Derivatives

Binance, OKX, Bybit and Deribit rows expose mark, index, funding, open interest and perpetual basis when public sources return them. Each derivative field has independent availability, observation time and last-good retention; a successful mark, OI or liquidation request cannot zero or refresh a missing funding or basis observation. Native open-interest units are normalized to BTC with the current mark where required; venue source and unit treatment remain visible. Coinbase is explicitly spot-only.

Liquidation heatmap

The heatmap bins captured public liquidation events by venue and execution price, preserving reported size and time. It shows events actually observed during the live browser session—not estimated future liquidation levels, not hidden positions and not a complete global ledger. There is no durable server-side liquidation archive in the current public build.

On-chain

Exchange inflow, outflow, netflow and reserve data use Coin Metrics' provider-labeled exchange entity set. “Entity-adjusted” means the provider's label/cluster methodology, not proprietary wallet clustering by FINITE. Long- and short-term holder SOPR, total SOPR, supply in profit/loss and activity-age cohorts appear only when the Community source publishes them. Each family keeps its own observation time. A missing refresh retains only that field's last verified value and original timestamp; retained fields are disclosed and are never relabeled as current.

FLOWS

Venue cumulative delta is taker-buy BTC minus taker-sell BTC over the selected 1m, 5m, 15m or 1h window. Large prints are a user-selected size threshold, not proof that a wallet belongs to a whale. Venue divergence compares observed flow and depth; missed reconnect events are never fabricated.

INTEL

A story must be Bitcoin-primary, remain inside the 36-hour window and pass the 42/99 relevance threshold. Base relevance combines recency, source weight, explicit market-impact language and catalyst class. Contextual score adds bounded tape, conviction and active-theme confirmation. “Tape fit” is present correlation, not proof of headline causality.

NETWORK

Security budget per block equals protocol subsidy plus average fees in returned recent blocks. Fee contribution is fees ÷ security budget. Fee percentiles use visible projected-block ranges when available and explicitly fall back to recommendation tiers otherwise. Reachable node count is not total node population.

STACK

Realized price is realized capitalization ÷ current supply. UTXO profitability, SOPR, exchange reserve and holder-age families are requested independently, so one absent metric cannot erase valid neighboring data. “Inactive over one year” is an activity-age proxy, not an entity-adjusted long-term-holder label.

The public Status page validates value shape, expected venue presence, observation/retrieval age, depth availability, retained-state flags and each provider's reporting cadence. It distinguishes source-event, receipt and REST-retrieval clocks, and separates browser-to-edge response time from underlying data freshness. A 200 HTTP response containing missing or stale core data is degraded or unavailable, never green.

Market values are continuously visible through reconnects when a verified last-good value exists. The retained label, timestamp and source remain attached. No public browser can guarantee universal sub-500ms latency, uninterrupted background alerts or durable capture while the app is closed.

Saved chart templates, dashboard pins, watch selections and alert thresholds are device-local unless a future screen explicitly states otherwise. Browser alerts evaluate values while the page is active and may require notification permission; they are not server-guaranteed delivery. CSV and API outputs preserve timestamps and attribution. A shareable snapshot is a captured read, not a live promise after its timestamp.

• FINITE is market intelligence, not investment advice or a trade instruction.
• Public APIs can pause, throttle, change schema or differ by region.
• Stablecoin pairs carry depeg and venue-basis risk.
• Exchange, miner and holder labels are provider estimates and can be revised.
• Liquidation events do not reveal all leverage or future liquidation prices.
• Price causality cannot be proven from a headline, flow or on-chain correlation.