Library/Market Microstructure/Market-Depth Analytics

D11-F05-A06 / Complete engineering topic

Liquidity-Wall and Concentration Detection

A production-minded guide to Liquidity-Wall and Concentration Detection.

D11 · MARKET MICROSTRUCTURE
D11-F05-A06Canonical / Tested / Open
D11 / D11-F05

Screen visible depth for concentration and candidate walls while refusing to infer intent, persistence, or spoofing.

The decision this tutorial makes visible

One large price level can dominate a depth window. Share concentration, effective level count, and a transparent wall rule expose that structure.

The precise question is: Is visible depth broadly distributed, or is a declared level unusually dominant under explicit share and median-multiple thresholds?

This matters to two readers at once. A practitioner needs to know what the diagnostic does and does not justify. A builder needs a contract that can be reproduced from the same ordered inputs in Python, TypeScript, a visual, and a browser lab. This tutorial supplies both views and keeps the selected single-venue convention visible throughout.

Intuition before notation

HHI rises when quantity concentrates in fewer levels; the two-part wall rule prevents a small absolute outlier from passing on ratio alone.

The diagnostic is not a free-floating signal. Its meaning depends on the declared book or lifecycle scope, the event or snapshot clock, equality handling, the included price window, and the treatment of unavailable state. If any of those change, the result is a different estimand or engine rule even when the final field name looks similar.

Scope and nearby methods

The canonical detector applies fractional HHI to level depth shares and flags a level only when it clears both a median multiple and minimum share. DOJ antitrust thresholds are not reused.

VariantDefinitionBest useMain limitation
Fractional level HHISum squared decimal depth sharesComparable concentration geometryThreshold is application-specific
Largest-share screenUse only the maximum shareSimple monitoringIgnores remaining distribution
Spoofing surveillance modelCombine placement, cancellation, fills, intent evidence, and controlsCompliance investigationCannot be inferred from a snapshot wall

What is sourced, selected, synthetic, and derived

RoleMaterial claimEvidenceBoundary
Sourced factAuthoritative sources establish only the feed, message, formula-history, or empirical context recorded for this topic.S1, S2, S3They do not validate the synthetic fixture, future availability, prediction, or profitability.
Implementation choiceThe canonical detector applies fractional HHI to level depth shares and flags a level only when it clears both a median multiple and minimum share. DOJ antitrust thresholds are not reused.Frozen package definitionNearby venue, provider, consolidated, hidden, or event-flow variants are not silently substituted.
Synthetic teaching inputThe canonical and playground inputs are repository-authored teaching states.datasets/canonical-input.json and datasets/scenario-results.jsonThey are not observed exchange snapshots and carry no redistribution claim.
Author-derived calculationThe 450-share level is 50% of the 900-share window and 3.75 times the median 120, so it passes both wall conditions. Its HHI is calculated from all five level shares.Formula, canonical fixture, Python/TypeScript parity, and independent arithmeticDefinition fidelity does not establish causal, predictive, or trading value.

The authoritative sources define feed capabilities, protocol vocabulary, venue-specific rules, or published empirical context. They do not certify the synthetic numbers in this tutorial. The repository fixture is deliberately invented for auditability, and the displayed output is author-derived from that fixture under the selected implementation choice.

Formula, symbols, and numerical policy

Plain text
share_i=q_i/Σq; HHI=Σ share_i²; effective_levels=1/HHI; wall_i=(q_i≥m·median(q))∧(share_i≥s_min)
SymbolMeaningUnitPolicy
q_ivisible quantity at included level isharesnonnegative
w_ilevel share q_i divided by total quantityfractionsums to one when total is positive
Hfractional HHI sum of squared w_ifractionbetween 1/n and 1
N_effeffective level countlevels1/H
m, uwall multiple and minimum-share thresholdsratio and fractionboth must pass for a wall
  • Use fractional shares, so H ranges from 1/n to 1; do not import 0–10,000 antitrust scaling or thresholds.
  • Wall threshold equality is included, and both the median-multiple and minimum-share tests must pass.
  • A detected wall is a size/concentration label only; it cannot establish persistence, manipulation, or intent.

Read the formula in the same order as the algorithm. Validate identity, ordering, units, and supported state first. Apply the selected equality and window rules second. Calculate with unrounded numeric values. Round only at the declared presentation boundary, and preserve null as a diagnostic rather than coercing it to zero.

Build the algorithm

  1. Validate one ordered book side
  2. Normalize level quantities into shares
  3. Sum squared shares
  4. Calculate median multiples
  5. Flag only levels passing both thresholds and classify state

Production-minded operational checklist

  1. Freeze the included side, level window, and threshold values.
  2. Reconcile level shares to one and H to the selected fractional scale.
  3. Inspect both concentration and wall flags; neither substitutes for the other.
  4. Run threshold sensitivity around equality boundaries.
  5. Never convert a geometric flag into a spoofing or participant-intent claim.

The checklist is intentionally stricter than a charting demo. A plausible number from stale, incomplete, off-grid, misordered, or unsupported state is more dangerous than an explicit rejection.

Worked synthetic example

The canonical fixture is synthetic teaching data, not an observed venue snapshot or private order record. Its primary author-derived output, hhi_fraction, is 0.312839506173. The complete input and output are in datasets/canonical-input.json and datasets/expected-output.json.

The 450-share level is 50% of the 900-share window and 3.75 times the median 120, so it passes both wall conditions. Its HHI is calculated from all five level shares.

The structured result retains state and diagnostics in addition to the primary number. That makes the calculation independently reviewable and prevents a partial, null, rejected, or venue-bounded outcome from being mistaken for an unqualified value.

Boundary and counterexample workbook

The playground computes every scenario at 61 deterministic parameter states. The table shows the midpoint used for prose review; the full state ledger is in datasets/scenario-results.json.

ScenarioPurposeStatePrimary outputDiagnostic
Canonical sweepMove one declared driver around the canonical visible book.wall-detected0.3495effective levels 2.86; walls 1
Balanced comparisonCompare a deliberately symmetric or neutral state.distributed0.2000effective levels 5.00; walls 0
Bid-heavy boundaryIncrease visible bid-side pressure within the declared window.distributed0.2224effective levels 4.50; walls 0
Ask-heavy or partial caseStress the opposite side or available-depth boundary.wall-detected0.3341effective levels 2.99; walls 2
Thin or concentrated caseReduce breadth or concentrate depth to expose failure semantics.wall-detected0.6005effective levels 1.67; walls 1
Equality and null boundaryLand exactly on a classification, price, coverage, or null-producing boundary.wall-detected0.2426effective levels 4.12; walls 1
Adversarial interpretation checkCreate a valid but misleading-looking state that must retain its misuse guardrail.wall-detected0.4055effective levels 2.47; walls 2

These rows are not backtest observations. They are controlled counterexamples that expose how one driver changes the state, output, or reason code while the rest of the contract stays fixed.

Visualize the boundary

Liquidity-Wall and Concentration Detection annotated teaching map

Open this SVG at full size, or use the guided playground to compare canonical, equality, null, residual, adversarial, and cross-method scenarios.

The Mermaid flow answers where the selected calculation sits in the processing sequence. The SVG keeps the formula, output, decision boundary, and invariant visible together. The lab lets the reader step through the same structured states without changing the underlying definition.

Implementation walkthrough

The Python and TypeScript references begin with the same validation contract. They reject malformed and unsupported state before calculation, preserve best-to-worse or event-sequence ordering, apply the equality policy in the symbol table, and return a structured diagnostic object rather than one context-free number.

The main implementation branches are:

  • Level passes multiple and share — Add candidate wall diagnostic, because Both relative and absolute-window importance are required.
  • No wall but HHI ≥ threshold — Return concentrated, because Distribution is uneven without one qualifying wall.
  • A wall appears once — Do not infer manipulation, because Intent and persistence are unobserved.

Neither reference silently fetches data, mutates caller-owned inputs outside the declared engine behavior, guesses hidden state, or substitutes a provider default. Shared JSON fixtures make value, null, state, and reason-code drift visible across languages.

Testing and validation

Definition tests compare every canonical field, reject malformed state, and exercise the material boundary. Family validation recomputes every playground state from the reference function. Independent arithmetic is recorded beside the fixture rather than inferred only from implementation output.

The audit must preserve these invariants:

  • Fractional HHI is in [1/L,1] for L positive levels.
  • Effective level count is in [1,L].
  • Every wall includes its price, share, and median multiple.
  • The structured output retains hhi_fraction, state, and the diagnostics needed to reproduce its branch.

Passing definition and parity checks proves that the implementation matches the selected contract. It does not prove that displayed state remains available, that another venue uses the same rule, or that the diagnostic predicts a later price or fill.

Failure modes and misuse

  • Displayed quantity can be cancelled or executed before a real order arrives.
  • Hidden, reserve, midpoint, routed, and other-venue liquidity is outside the visible snapshot.
  • A deeper book does not by itself establish future price direction or executable capacity.
  • A visible wall is neither proof of executable persistence nor evidence of spoofing or intent.

Debugging order

When a result looks surprising, inspect the state in this order:

  1. Confirm instrument, venue, session, order type, and side.
  2. Confirm price, quantity, tick, clock, and sequence units.
  3. Confirm the included depth window or lifecycle-event boundary.
  4. Confirm equality, null, residual, and reset policies.
  5. Recalculate the invariant and midpoint counterexample before changing code.

Evidence and historical boundary

Historical decision: deferred. A named market-depth case would require instrument and venue identity, a sequence-complete licensed feed, clock and correction policy, visible/hidden scope, and redistribution permission. The synthetic case provides stronger public auditability for this build.

The primary sources are NYSE Integrated Feed, Nasdaq TotalView-ITCH 5.0, DOJ HHI definition. They support the source roles listed in the research ledger, not a redistributable historical observation, participant-intent claim, venue-conformance certification, profitability claim, or prediction claim.

Summary and next topic

You can now screen visible depth concentration without overclaiming intent. The learning flow is: Multi-Level Sweep Cost and Slippage → Liquidity-Wall and Concentration Detection → Depth Depletion and Replenishment. Carry the result forward only with its scope, clock, state, and evidence label.

Liquidity-Wall and Concentration Detection calculation flow

This flow identifies the selected calculation stages and the structured output.

Rendering system map…

Takeaway: Concentration is measurable; persistence, intent, and future execution are not visible in one snapshot.

ReferencesPrimary sources and evidence notes

Expand the source trail, evidence role, and limitations behind the engineering choices.

S1 — NYSE Integrated Feed

  • Organization or authors: New York Stock Exchange
  • Source type: Official exchange data-product description
  • Publication or effective date: Current page accessed 2026-07-30
  • Version: Current web version
  • URL or DOI: https://www.nyse.com/data-products/catalog/integrated-feed
  • Accessed: 2026-07-30
  • Jurisdiction: United States equities; NYSE Group venues
  • Supports: Order-by-order events, depth, trades, imbalances, security status, and sequence integration are available in an official depth feed.
  • Limitations: Does not define the package formulas, provide public redistributable observations, or reveal hidden liquidity.

S2 — Nasdaq TotalView-ITCH 5.0 Specification

  • Organization or authors: Nasdaq
  • Source type: Official exchange technical specification
  • Publication or effective date: Specification current at access
  • Version: 5.0
  • URL or DOI: https://nasdaqtrader.com/content/technicalsupport/specifications/dataproducts/NQTVITCHSpecification.pdf
  • Accessed: 2026-07-30
  • Jurisdiction: United States equities; Nasdaq
  • Supports: Add, execute, cancel, delete, and replace messages needed to reconstruct displayed order state.
  • Limitations: Feed messages are venue-specific; the specification does not make displayed depth firm through future latency.

S3 — Herfindahl-Hirschman Index

  • Organization or authors: U.S. Department of Justice, Antitrust Division
  • Source type: Official methodological description
  • Publication or effective date: Updated January 17, 2024
  • Version: Current web version
  • URL or DOI: https://www.justice.gov/atr/herfindahl-hirschman-index
  • Accessed: 2026-07-30
  • Jurisdiction: United States antitrust context
  • Supports: HHI is the sum of squared shares and rises as a distribution becomes more concentrated.
  • Limitations: Antitrust market thresholds do not apply to order-book price levels; this package reuses only the mathematical share concentration form.

Evidence boundary

The sources establish feed fields, venue rules, protocol states, or published microstructure definitions. They do not verify the repository-authored synthetic fixture, thresholds, empirical usefulness, execution probability, or profitability. The package-selected choices are labeled as implementation choices wherever they are used.

market-depth-analytics.ts
export type JsonObject = Record<string, any>;

function numberValue(name: string, value: unknown, options: {positive?: boolean; nonnegative?: boolean} = {}): number {
  if (typeof value !== "number" || !Number.isFinite(value)) throw new TypeError(`${name} must be a finite number`);
  if (options.positive && value <= 0) throw new RangeError(`${name} must be positive`);
  if (options.nonnegative && value < 0) throw new RangeError(`${name} must be nonnegative`);
  return value;
}

function integerValue(name: string, value: unknown, positive = false): number {
  const result = numberValue(name, value);
  if (!Number.isInteger(result)) throw new RangeError(`${name} must be an integer`);
  if (positive && result <= 0) throw new RangeError(`${name} must be positive`);
  return result;
}

type Level = {price: number; quantity: number};

function levels(raw: unknown, side: "bid" | "ask"): Level[] {
  if (!Array.isArray(raw) || raw.length === 0) throw new RangeError(`${side}s must be a non-empty list`);
  const seen = new Set<number>();
  const result = raw.map((item, index) => {
    if (!item || typeof item !== "object") throw new TypeError(`${side}s[${index}] must be an object`);
    const row = item as JsonObject;
    const price = numberValue(`${side}s[${index}].price`, row.price, {positive: true});
    const quantity = numberValue(`${side}s[${index}].quantity`, row.quantity, {positive: true});
    if (seen.has(price)) throw new RangeError(`${side} prices must be unique aggregated levels`);
    seen.add(price);
    return {price, quantity};
  });
  for (let index = 1; index < result.length; index += 1) {
    if (side === "bid" && result[index - 1].price <= result[index].price) throw new RangeError("bids must be strictly descending from best price");
    if (side === "ask" && result[index - 1].price >= result[index].price) throw new RangeError("asks must be strictly ascending from best price");
  }
  return result;
}

function book(bidsRaw: unknown, asksRaw: unknown): [Level[], Level[]] {
  const bids = levels(bidsRaw, "bid"), asks = levels(asksRaw, "ask");
  if (bids[0].price >= asks[0].price) throw new RangeError("visible book must be uncrossed");
  return [bids, asks];
}

function tickDistance(price: number, best: number, tick: number, side: "bid" | "ask"): number {
  const raw = side === "bid" ? (best - price) / tick : (price - best) / tick;
  const rounded = Math.round(raw);
  if (Math.abs(raw - rounded) > 1e-7) throw new RangeError("price is not aligned to the declared tick grid");
  return rounded;
}

export function cumulativeDepth(bidsRaw: unknown, asksRaw: unknown, tickRaw: unknown): JsonObject {
  const [bids, asks] = book(bidsRaw, asksRaw);
  const tick = numberValue("tick_size", tickRaw, {positive: true});
  const rows = (sideLevels: Level[], side: "bid" | "ask") => {
    let cumulative = 0;
    const best = sideLevels[0].price;
    return sideLevels.map(level => {
      cumulative += level.quantity;
      return {price: level.price, quantity: level.quantity, distance_ticks: tickDistance(level.price, best, tick, side), cumulative_quantity: cumulative};
    });
  };
  const bidRows = rows(bids, "bid"), askRows = rows(asks, "ask");
  const spreadRaw = (asks[0].price - bids[0].price) / tick, spreadTicks = Math.round(spreadRaw);
  if (Math.abs(spreadRaw - spreadTicks) > 1e-7) throw new RangeError("spread is not aligned to the declared tick grid");
  const bidTotal = bidRows[bidRows.length - 1].cumulative_quantity, askTotal = askRows[askRows.length - 1].cumulative_quantity;
  return {model: "visible-cumulative-depth", best_bid: bids[0].price, best_ask: asks[0].price, tick_size: tick, spread_ticks: spreadTicks, bid_levels: bidRows, ask_levels: askRows, total_bid_depth: bidTotal, total_ask_depth: askTotal, state: bidTotal > askTotal ? "bid-deeper" : bidTotal < askTotal ? "ask-deeper" : "balanced"};
}

export function topNDepthImbalance(bidsRaw: unknown, asksRaw: unknown, nRaw: unknown, thresholdRaw: unknown = 0.1): JsonObject {
  const [bids, asks] = book(bidsRaw, asksRaw);
  const n = integerValue("n", nRaw, true), threshold = numberValue("threshold", thresholdRaw, {nonnegative: true});
  if (threshold > 1) throw new RangeError("threshold must not exceed one");
  if (bids.length < n || asks.length < n) throw new RangeError("both sides must contain at least n visible levels");
  const bidDepth = bids.slice(0, n).reduce((sum, row) => sum + row.quantity, 0);
  const askDepth = asks.slice(0, n).reduce((sum, row) => sum + row.quantity, 0);
  const total = bidDepth + askDepth, imbalance = (bidDepth - askDepth) / total;
  return {model: "top-n-visible-depth-imbalance", n, bid_depth: bidDepth, ask_depth: askDepth, total_depth: total, imbalance, threshold, state: imbalance > threshold ? "bid-heavy" : imbalance < -threshold ? "ask-heavy" : "balanced-band", complete_window: true};
}

export function depthAtDistanceProfile(bidsRaw: unknown, asksRaw: unknown, tickRaw: unknown, maxRaw: unknown): JsonObject {
  const [bids, asks] = book(bidsRaw, asksRaw);
  const tick = numberValue("tick_size", tickRaw, {positive: true}), maximum = integerValue("max_distance_ticks", maxRaw);
  if (maximum < 0) throw new RangeError("max_distance_ticks must be nonnegative");
  const profile = (sideLevels: Level[], side: "bid" | "ask") => {
    const best = sideLevels[0].price, byDistance = new Map<number, number>();
    sideLevels.forEach(level => byDistance.set(tickDistance(level.price, best, tick, side), level.quantity));
    let included = 0;
    for (let d = 0; d <= maximum; d += 1) included += byDistance.get(d) ?? 0;
    let cumulative = 0;
    return Array.from({length: maximum + 1}, (_, distance) => {
      const quantity = byDistance.get(distance) ?? 0; cumulative += quantity;
      return {distance_ticks: distance, quantity, cumulative_quantity: cumulative, depth_share: included ? quantity / included : 0};
    });
  };
  const bidProfile = profile(bids, "bid"), askProfile = profile(asks, "ask");
  const bidTotal = bidProfile[bidProfile.length - 1].cumulative_quantity, askTotal = askProfile[askProfile.length - 1].cumulative_quantity;
  return {model: "same-side-best-tick-distance-profile", tick_size: tick, max_distance_ticks: maximum, bid_profile: bidProfile, ask_profile: askProfile, included_bid_depth: bidTotal, included_ask_depth: askTotal, state: bidTotal > askTotal ? "bid-profile-heavier" : askTotal > bidTotal ? "ask-profile-heavier" : "profile-balanced"};
}

function sweep(bidsRaw: unknown, asksRaw: unknown, sideRaw: unknown, quantityRaw: unknown, limitRaw: unknown): JsonObject {
  const [bids, asks] = book(bidsRaw, asksRaw);
  if (sideRaw !== "buy" && sideRaw !== "sell") throw new RangeError("side must be buy or sell");
  const side = sideRaw, requested = numberValue("quantity", quantityRaw, {positive: true});
  const limit = limitRaw == null ? null : numberValue("limit_price", limitRaw, {positive: true});
  const sideLevels = side === "buy" ? asks : bids;
  let remaining = requested, notional = 0;
  const fills: JsonObject[] = [];
  for (let index = 0; index < sideLevels.length && remaining > 1e-12; index += 1) {
    const level = sideLevels[index];
    if (limit != null && (side === "buy" ? level.price > limit : level.price < limit)) break;
    const fillQuantity = Math.min(remaining, level.quantity);
    fills.push({level_index: index, price: level.price, quantity: fillQuantity, notional: fillQuantity * level.price});
    notional += fillQuantity * level.price; remaining -= fillQuantity;
  }
  if (remaining <= 1e-12) remaining = 0;
  const filled = requested - remaining, partialVwap = filled ? notional / filled : null;
  return {side, requested_quantity: requested, filled_quantity: filled, unfilled_quantity: remaining, full_fill: remaining === 0, fills, fill_notional: notional, partial_vwap: partialVwap, worst_fill_price: fills.length ? fills[fills.length - 1].price : null, best_bid: bids[0].price, best_ask: asks[0].price, midpoint: (bids[0].price + asks[0].price) / 2};
}

export function expectedFillPrice(bids: unknown, asks: unknown, side: unknown, quantity: unknown, limitPrice: unknown = null): JsonObject {
  const result = sweep(bids, asks, side, quantity, limitPrice);
  return {model: "deterministic-visible-book-fill-estimate", ...result, expected_fill_price: result.full_fill ? result.partial_vwap : null, state: result.full_fill ? "full-fill" : result.filled_quantity ? "partial-fill" : "unfilled"};
}

export function sweepCostAndSlippage(bids: unknown, asks: unknown, side: unknown, quantity: unknown, benchmarkRaw: unknown = "midpoint", benchmarkPriceRaw: unknown = null, limitPrice: unknown = null): JsonObject {
  const result = sweep(bids, asks, side, quantity, limitPrice);
  if (!["midpoint", "best-quote", "explicit"].includes(String(benchmarkRaw))) throw new RangeError("benchmark must be midpoint, best-quote, or explicit");
  const benchmark = String(benchmarkRaw);
  const reference = benchmark === "midpoint" ? result.midpoint : benchmark === "best-quote" ? (side === "buy" ? result.best_ask : result.best_bid) : numberValue("benchmark_price", benchmarkPriceRaw, {positive: true});
  const vwap = result.partial_vwap, direction = side === "buy" ? 1 : -1;
  const slippagePrice = vwap == null ? null : direction * (vwap - reference);
  const slippageBps = slippagePrice == null ? null : slippagePrice / reference * 10_000;
  return {model: "visible-book-sweep-cost", ...result, benchmark, benchmark_price: reference, sweep_vwap: vwap, signed_slippage_price: slippagePrice, signed_slippage_bps: slippageBps, total_slippage_cost: slippagePrice == null ? null : slippagePrice * result.filled_quantity, state: result.full_fill ? "full-sweep" : result.filled_quantity ? "partial-sweep" : "unfilled"};
}

function median(values: number[]): number {
  const sorted = [...values].sort((a, b) => a - b), middle = Math.floor(sorted.length / 2);
  return sorted.length % 2 ? sorted[middle] : (sorted[middle - 1] + sorted[middle]) / 2;
}

export function liquidityWallConcentration(levelsRaw: unknown, multipleRaw: unknown = 2, minShareRaw: unknown = 0.25, thresholdRaw: unknown = 0.30): JsonObject {
  const parsed = levels(levelsRaw, "ask");
  const multiple = numberValue("wall_multiple", multipleRaw, {positive: true});
  const minShare = numberValue("minimum_share", minShareRaw, {nonnegative: true});
  const threshold = numberValue("concentration_threshold", thresholdRaw, {nonnegative: true});
  if (minShare > 1 || threshold > 1) throw new RangeError("share thresholds must not exceed one");
  const quantities = parsed.map(row => row.quantity), total = quantities.reduce((a, b) => a + b, 0), med = median(quantities);
  const shares = quantities.map(value => value / total), hhi = shares.reduce((sum, value) => sum + value * value, 0);
  const walls = parsed.map((level, index) => ({level_index: index, price: level.price, quantity: level.quantity, depth_share: shares[index], median_multiple: level.quantity / med})).filter(row => row.quantity >= multiple * med && row.depth_share >= minShare);
  let largestIndex = 0; quantities.forEach((value, index) => { if (value > quantities[largestIndex]) largestIndex = index; });
  return {model: "level-share-concentration-and-wall-screen", level_count: parsed.length, total_depth: total, median_level_quantity: med, hhi_fraction: hhi, effective_level_count: 1 / hhi, largest_level_index: largestIndex, largest_level_share: shares[largestIndex], walls, wall_multiple: multiple, minimum_share: minShare, concentration_threshold: threshold, state: walls.length ? "wall-detected" : hhi >= threshold ? "concentrated" : "distributed"};
}

export function depthDepletionReplenishment(seriesRaw: unknown): JsonObject {
  if (!Array.isArray(seriesRaw) || seriesRaw.length < 2) throw new RangeError("series must contain at least two snapshots");
  const parsed = seriesRaw.map((item, index) => {
    if (!item || typeof item !== "object") throw new TypeError(`series[${index}] must be an object`);
    const row = item as JsonObject;
    return {timestamp_seconds: numberValue(`series[${index}].timestamp_seconds`, row.timestamp_seconds, {nonnegative: true}), quantity: numberValue(`series[${index}].quantity`, row.quantity, {nonnegative: true})};
  });
  for (let i = 1; i < parsed.length; i += 1) if (parsed[i - 1].timestamp_seconds >= parsed[i].timestamp_seconds) throw new RangeError("timestamps must be strictly increasing");
  let depletion = 0, replenishment = 0;
  const changes = parsed.slice(1).map((current, index) => {
    const previous = parsed[index], delta = current.quantity - previous.quantity;
    replenishment += Math.max(delta, 0); depletion += Math.max(-delta, 0);
    return {timestamp_seconds: current.timestamp_seconds, previous_quantity: previous.quantity, quantity: current.quantity, delta_quantity: delta, classification: delta > 0 ? "replenishment" : delta < 0 ? "depletion" : "unchanged"};
  });
  const net = replenishment - depletion;
  return {model: "snapshot-visible-depth-change-decomposition", snapshot_count: parsed.length, start_quantity: parsed[0].quantity, end_quantity: parsed[parsed.length - 1].quantity, gross_depletion: depletion, gross_replenishment: replenishment, net_depth_change: net, replenishment_to_depletion: depletion ? replenishment / depletion : null, changes, state: net > 0 ? "net-replenishing" : net < 0 ? "net-depleting" : "net-flat"};
}

export function marketDepthHeatmap(snapshotsRaw: unknown, tickRaw: unknown, binRaw: unknown, maxRaw: unknown): JsonObject {
  if (!Array.isArray(snapshotsRaw) || snapshotsRaw.length < 2) throw new RangeError("snapshots must contain at least two regular observations");
  const snapshots = snapshotsRaw as JsonObject[], tick = numberValue("tick_size", tickRaw, {positive: true}), binSize = numberValue("time_bin_seconds", binRaw, {positive: true}), maximum = integerValue("max_distance_ticks", maxRaw);
  if (maximum < 0) throw new RangeError("max_distance_ticks must be nonnegative");
  const times = snapshots.map((item, index) => {
    if (!item || typeof item !== "object") throw new TypeError("each snapshot must be an object");
    return numberValue(`snapshots[${index}].timestamp_seconds`, item.timestamp_seconds, {nonnegative: true});
  });
  for (let i = 1; i < times.length; i += 1) if (times[i - 1] >= times[i]) throw new RangeError("snapshot timestamps must be strictly increasing");
  const intervals = times.slice(1).map((time, index) => time - times[index]);
  if (intervals.slice(1).some(interval => Math.abs(interval - intervals[0]) > 1e-9)) throw new RangeError("canonical heatmap requires regularly sampled snapshots");
  const origin = times[0], buckets = new Map<string, number[]>(), binSet = new Set<number>();
  snapshots.forEach((snapshot, snapshotIndex) => {
    const [bids, asks] = book(snapshot.bids, snapshot.asks), binIndex = Math.floor((times[snapshotIndex] - origin) / binSize);
    binSet.add(binIndex);
    const observed = new Map<number, number>();
    bids.forEach(level => { const d = tickDistance(level.price, bids[0].price, tick, "bid"); if (d <= maximum) observed.set(-(d + 1), level.quantity); });
    asks.forEach(level => { const d = tickDistance(level.price, asks[0].price, tick, "ask"); if (d <= maximum) observed.set(d + 1, level.quantity); });
    const coordinates = [...Array.from({length: maximum + 1}, (_, i) => -(maximum + 1) + i), ...Array.from({length: maximum + 1}, (_, i) => i + 1)];
    coordinates.forEach(coordinate => { const key = `${binIndex}|${coordinate}`, values = buckets.get(key) ?? []; values.push(observed.get(coordinate) ?? 0); buckets.set(key, values); });
  });
  const cells = [...buckets.entries()].map(([key, values]) => {
    const [binIndex, coordinate] = key.split("|").map(Number);
    return {time_bin_index: binIndex, time_bin_start_seconds: origin + binIndex * binSize, book_coordinate: coordinate, mean_quantity: values.reduce((a, b) => a + b, 0) / values.length, max_quantity: Math.max(...values), observation_count: values.length};
  }).sort((a, b) => a.time_bin_index - b.time_bin_index || a.book_coordinate - b.book_coordinate);
  const peak = cells.reduce((best, cell) => cell.mean_quantity > best.mean_quantity ? cell : best);
  return {model: "regular-snapshot-inside-relative-depth-heatmap", snapshot_interval_seconds: intervals[0], time_bin_seconds: binSize, max_distance_ticks: maximum, time_bin_count: binSet.size, book_coordinates: [...Array.from({length: maximum + 1}, (_, i) => -(maximum + 1) + i), ...Array.from({length: maximum + 1}, (_, i) => i + 1)], cells, peak_cell: peak, state: peak.book_coordinate < 0 ? "bid-peak" : "ask-peak"};
}

export function calculate(topicId: string, inputs: JsonObject): JsonObject {
  if (!inputs || typeof inputs !== "object" || Array.isArray(inputs)) throw new TypeError("inputs must be an object");
  if (topicId === "D11-F05-A01") return cumulativeDepth(inputs.bids, inputs.asks, inputs.tick_size);
  if (topicId === "D11-F05-A02") return topNDepthImbalance(inputs.bids, inputs.asks, inputs.n, inputs.threshold ?? 0.1);
  if (topicId === "D11-F05-A03") return depthAtDistanceProfile(inputs.bids, inputs.asks, inputs.tick_size, inputs.max_distance_ticks);
  if (topicId === "D11-F05-A04") return expectedFillPrice(inputs.bids, inputs.asks, inputs.side, inputs.quantity, inputs.limit_price);
  if (topicId === "D11-F05-A05") return sweepCostAndSlippage(inputs.bids, inputs.asks, inputs.side, inputs.quantity, inputs.benchmark ?? "midpoint", inputs.benchmark_price, inputs.limit_price);
  if (topicId === "D11-F05-A06") return liquidityWallConcentration(inputs.levels, inputs.wall_multiple ?? 2, inputs.minimum_share ?? 0.25, inputs.concentration_threshold ?? 0.30);
  if (topicId === "D11-F05-A07") return depthDepletionReplenishment(inputs.series);
  if (topicId === "D11-F05-A08") return marketDepthHeatmap(inputs.snapshots, inputs.tick_size, inputs.time_bin_seconds, inputs.max_distance_ticks);
  throw new RangeError(`unsupported topic_id: ${topicId}`);
}
Full-height labplaygroundOpen full screen