Saltar a contenido

🧪 Risk UI: Correlation and Dashboard Risk Tabs

This page covers the UI rules of two risk surfaces. The Correlation tab of the Assets page — the Asset Global lab — runs the risk engine on a hand-picked selection of assets ({kind: "asset_set", asset_ids: [...]}): how the selection is built and kept, how the engine's eligibility verdicts are used, which requests each section sends, and how each section draws and discloses its results. The Risk tab of the Dashboard and of Broker Detail runs it on a portfolio ({kind: "portfolio", broker_ids: [...]}) as four levels: see Dashboard Risk Tab.

It does not repeat the engine. The request flow, the analytics, the eligibility rules and thresholds, the panel controller and the shared benchmark store are in Risk Engine; the mathematics is in Risk Metrics; what users see is in the Correlation Tab and Risk Tab user pages.

Paths are relative to frontend/src/lib/components/risk/ unless stated otherwise. The lab never shows an amount of money: a selection has no weights.

🗺️ Source Map

Source Responsibility
frontend/src/routes/(app)/assets/+page.svelte Mounts the lab on the Correlation tab with the page's full asset list; hides the list filters there (l.1310); its toolbar's Sync selection and Reload All call the lab's openSync() and reload() (l.1509, l.1524); onfitperiod moves the toolbar range (l.1609-1614); targetCurrency is the instance default currency (l.1618).
AssetSetRiskPanel.svelte The lab: selection card, eligibility, holdings command, benchmark picker snippet, banner and notice, and the four sections (l.847-864).
assetSetSelection.ts Opening selection, persistence, bulk actions, labBenchmarkId(), labL3Waits().
eligibility.ts Reads and words the engine's verdicts; batches; fitPeriodOffer().
LabAssetPicker.svelte The +: AssetPickerPanel in multi mode.
AssetSetCorrelationSection.svelte, CorrelationHeatmap.svelte, CorrelationPairsList.svelte, correlationHelpers.ts, matrixMetadata.svelte.ts Correlation section.
AssetSetComparisonLevels.svelte, AssetSetLossComparisonSection.svelte, AssetSetRiskReturnSection.svelte, RiskReturnLevel.svelte, assetSetLevels.ts, assetSetTable.ts L1° (losses) and L3° (risk and return).
AssetSetReplaySection.svelte, levels/l4/L4Replay.svelte, levels/l4/TornadoChart.svelte L4° (historical replay).
levels/RiskLevelSection.svelte, levels/RiskPartialNotice.svelte, levels/partialNotice.ts, levels/levelMetadata.ts, syncTargets.ts Section frames, the shared notice, banner actions.

The UI strings live under risk.assetSet.* (lab-only) and risk.levels.* (shared with the Dashboard).

🧺 Selection

🚪 Opening Selection

resolveInitialSelectionWithSource() (assetSetSelection.ts:137) picks the first rung that yields ids, and the card publishes it as data-selection-source (AssetSetRiskPanel.svelte:707):

  1. persisted — the last selection, stored in localStorage under lf_<userId>_assetGlobal.riskSelection.v1 (assetSetSelection.ts:39, l.64) and intersected with the page's list, so deleted or merged ids drop out.
  2. mine — the assets held on dateEnd in every accessible broker. heldAssetIds() reads a light fetchReport() without the daily and allocation histories (AssetSetRiskPanel.svelte:509), asks once more when it answers null (l.510-516), and keeps only ids of the page's list.
  3. fallback — the FALLBACK_SELECTION_SIZE (6) active assets with the most transactions, ties by id (assetSetSelection.ts:37, l.165).

Every rung is capped at MAX_SELECTED_ASSETS (100, assetSetSelection.ts:31), which is the API's limit for the scope. A change the user makes while rung 2 is loading wins: the late seed is dropped (selectionTouched, AssetSetRiskPanel.svelte:466). Every later change is written back (l.479).

🧰 Bulk Actions and Holdings

  • applyBulkAction() (assetSetSelection.ts:198) acts on candidates: the page's list minus the ineligible assets (AssetSetRiskPanel.svelte:347). Neither the + filters nor the Assets-tab filters narrow it. all adds every candidate, inactive ones included, up to 100; none empties the selection, parked assets included; invert swaps the candidates and leaves parked assets where they are.
  • My assets runs runHoldingsPreset(brokerId | null) (AssetSetRiskPanel.svelte:537): it replaces the selection with the assets held on dateEnd (all accessible brokers, or one), sliced to 100 (l.554), ineligible ones included. It is a command, not a filter: a date change does not run it again. A failed report sets risk.states.loadFailed (l.547); nothing held leaves the selection as it was and shows risk.assetSet.preset.noneHeld (l.551).
  • The + is LabAssetPicker.svelte, a thin wrapper around AssetPickerPanel in multi mode: rows in name order, a search over the name, the currency and the type label (searchText, LabAssetPicker.svelte:60), and room = 100 − selection size (AssetSetRiskPanel.svelte:368). The panel's own behaviour (Select visible, filters, blocked section) is documented there.

✅ Eligibility in the Lab

The rules behind POST /api/v1/risk/eligibility — the reasons, the 20-quote floor, the seven-day lateness, common_range and suggested_range — are in Analysis Eligibility. The lab uses the verdicts in three ways:

  • The whole list is asked, in batches of ELIGIBILITY_BATCH (500, eligibility.ts:47), after a 300 ms debounce (AssetSetRiskPanel.svelte:246), and again when the list, the period or the currency changes; a superseded answer is dropped. A failed call leaves no verdict (l.294-295): every asset stays selectable and the card shows risk.assetSet.eligibilityFailed.
  • An ineligible asset is parked: it stays in selectedAssetIds as a dashed chip, with the reasons in its tooltip, but only analysedIds (l.357) reaches the sections. The counter shows analysedIds.length of candidates.length (l.791). A warning chip is analysed.
  • The selection is asked again on its own (selectionVerdicts, l.316-340), so that unselected assets cannot narrow the period offered. fitPeriodOffer() (eligibility.ts:109) returns the answer's suggested_range when at least one selected asset is not eligible and is not no_price_history. The ranges survive only a single-batch answer, which a selection of at most 100 ids always is. The strip's button calls onfitperiod; the page clears its preset badge and applies the range to the toolbar (+page.svelte:1609-1614).

🔌 Requests and the Shared Window

Each section owns a controller (createRiskPanelController), and all four are mounted under the panel's {#if analysedIds.length > 0} (AssetSetRiskPanel.svelte:847), so an empty selection never reaches the API. Which analytics each section asks for is the table in By Level.

Section Controller Benchmark
Correlation AssetSetCorrelationSection.svelte:64 Never
L1° How much did each of these hurt? lossController, includeAssetSetLossLevels (AssetSetComparisonLevels.svelte:114-123) Never
L3° What did each of these pay for its risk? RiskControllerHost, includeAssetSetPaidLevels (AssetSetComparisonLevels.svelte:244) assetSetBenchmarkId (l.138)
L4° What if…? AssetSetReplaySection.svelte, replay on demand Never
  • Rates. Every lab controller sends appliedRiskFreePercent: 0 (AssetSetCorrelationSection.svelte:72, AssetSetComparisonLevels.svelte:120, l.136, AssetSetReplaySection.svelte:84), and the KPI and the comparison send target_annual_return: 0 (riskAnalysisHelpers.ts:334, l.368-371). The lab has no control for either.
  • Bad month. MONTHLY_VAR_HORIZON_DAYS = 30 calendar days (riskAnalysisHelpers.ts:280, used at l.340). The engine turns them into observations at the observed frequency — about 21 on a trading-day calendar, 30 on an every-day one — as explained in Value at Risk.
  • One calendar per request. The engine prepares one joint calendar per request from the scope, the window and the currency (Request Flow, step 4). Requests that agree on those three share it, so a matrix cell, an L1° row and an L3° dot cover the same dates (AssetSetComparisonLevels.svelte:20-29). A benchmark joins only L3°'s request, so only L3°'s window can move with it; its period line says so. How a short history narrows the calendar is in Alignment.
  • L3° waits. labBenchmarkId() (assetSetSelection.ts:228) passes a benchmark only in state set. labL3Waits() (l.245) keeps L3°'s controller unmounted while the picker is pending, or while a chosen benchmark waits for the lab's verdicts, so L3° asks once and never tries a benchmark that turns out blocked. The Beta and Correlation columns follow the answer, not the choice: benchmarkApplies = benchmarkId !== null && comparison.status === 'ok' (AssetSetComparisonLevels.svelte:207).
  • Sync and reload. openSync() opens a PageSyncModal on the selection's prices and the configured FX pairs that convert them (buildSyncTargets(), syncTargets.ts:43). After an accepted sync, and on reload(), the panel calls invalidateRisk() and bumps refreshVersion for every section (AssetSetRiskPanel.svelte:200-228); the replay also forgets its answer (AssetSetReplaySection.svelte:99).

🕸️ Correlation Section

  • Triangle. Only the strict lower triangle is drawn: the first asset has no row and the last has no column (CorrelationHeatmap.svelte:172, lowerTrianglePoints() in correlationHelpers.ts:439). Fewer than two assets draw nothing (l.424). A cell prints ρ with two decimals (l.506); the tooltip gives ρ with three, its band, the observations and coverage of the pair, and each asset's three largest sectors and countries plus a folded rest (l.343, l.487-489). The colour scale runs red at −1, neutral at 0 and blue at +1 (l.128-129).
  • Bands and lists. correlationBand() (correlationHelpers.ts:69-82): high above 0.7, inverse at −0.3 or below; near-identical at 0.9 or above (NEAR_IDENTICAL, l.268). topPairs() (l.290) lists only high pairs (The most alike) and inverse pairs (The ones that offset), five each; past PAIR_LIST_THRESHOLD (20) assets it lists eight, and the lists move ahead of the matrix on lg screens (CorrelationHeatmap.svelte:290, l.616). The opposite tag starts at OPPOSITE = −0.7 (CorrelationPairsList.svelte:47); a pair's colour follows the heatmap's polarity (l.62).
  • Orderings. similarity, the default, is clusterOrder() (correlationHelpers.ts:148): average-linkage clustering on 1 − |ρ|, a missing pair at distance 1 (l.124-127), ties broken by position. type follows the asset-type menu order, then the name. sector and region use dominantExposure() (l.362): the key holding at least DOMINANT_SHARE (0.5, l.337) of the classified weight, Other left out; exposureOrder() (l.396) then sorts named groups, Diversified, Other only and Unclassified. A mode is offered only when its data exist (CorrelationHeatmap.svelte:93). The sector and country distributions come from one bulk GET /api/v1/assets?asset_ids=… (matrixMetadata.svelte.ts:85); a failed read only hides those two buttons.

📉 Loss Table (L1°)

  • Columns by horizon: Bad day, Bad month, Worst fall (with its lasted N d line), Below peak now, Rise to peak (AssetSetLossComparisonSection.svelte:178-224). The order is the point: a one-day loss printed beside a multi-year drawdown must not read as something to add up.
  • Signs. The VaR/CVaR contract is positive and the drawdown contract negative; both are drawn as falls with U+2212 (fall(), l.106) and sorted as drawn (drawnLoss(), l.120), an unmeasured cell last. Every loss uses one red (l.124): no per-asset colour, no ranking.
  • Rows. Rows come from the selected ids, so every analysed asset has one and a blank is a dash whose tooltip explains it (figureCell(), assetSetTable.ts:25).
  • Widths. Each figure column opens at its title's measured width, which is also its minimum (titleWidth(), l.137; headerWidth(), riskReturnLevel.ts:106).
  • Mixed outcomes. asset_set_drawdown needs 2 observations and the other four analytics 20, so a short window returns one ok beside four unavailable (AssetSetComparisonLevels.svelte:39-47; floors in The Analytics).

⚖️ Risk and Return (L3°)

  • Shared level. RiskReturnLevel.svelte also renders the portfolio's L3. The lab declares capabilities={{ratios: true, benchmark: benchmarkApplies}} and no weight (AssetSetRiskReturnSection.svelte:198), so Beta and Correlation exist only with a measured benchmark (RiskReturnLevel.svelte:261).
  • Benchmark row. withBenchmarkRow() (assetSetLevels.ts:296) adds a ref-<id> row for a benchmark outside the selection; a selected benchmark keeps its own row. referenceRowsFirst() opens the table with it (RiskReturnLevel.svelte:95).
  • Chart and line. The chart needs two placeable points (l.101). capitalMarketLineAnchor() (frontend/src/lib/components/charts/scatterChartHelpers.ts:160) anchors the line on a benchmark with positive volatility, else on the portfolio, which a selection never has: no benchmark, no line. The notes come from riskReturnNotes() (riskReturnLevel.ts:174): above and line only with a line, return and priceOnly always. The line starts at LAB_RISK_FREE_RATE = 0 (AssetSetRiskReturnSection.svelte:137).
  • Period line. assetSetCalculationWindow() (assetSetLevels.ts:391) reads the first result with metadata — from end − calendar_days + 1 to end — and sets narrowed when that start is more than NARROWED_AFTER_DAYS (7, l.368) after the toolbar's start, or that end more than 7 days before the toolbar's end. The length is spelled in calendar units by calendarLength() (l.436).
  • Selection. Rows and dots share the table's single selection; a selected asset's dot is drawn larger in the selection green, while the benchmark keeps its colour (the same scatterChartHelpers.ts, l.101).

⏮️ Replay (L4°)

  • One rung. AssetSetReplaySection.svelte hands L4WhatIf the replay snippet only (l.135): no hypothetical shock and no simulation on an unweighted scope. The frame is collapsible and loads the scenario catalogue on first open (l.133); showMoney={false}.
  • Controls. L4Replay.svelte puts No preset first (l.95). A preset sets the dates; a quick range or a typed date drops the preset; both reset the answer (l.151-170). The quick ranges exclude MAX (l.231). The composition total is withheld when portfolio_return is null, as it is on every unweighted scope (l.269), and TornadoChart drops the weight, contribution and amount columns when no row carries them (levels/l4/TornadoChart.svelte:57-60).
  • Lifetime. A finished replay is discarded when the base signature moves, i.e. the selection or the dates (The Panel Controller), and after an accepted sync (AssetSetReplaySection.svelte:99). Coverage, exclusion reasons and the suggested replay period are in Replay Coverage.

🚩 Notices and Frames

  • One notice. partialNotice() (levels/partialNotice.ts:40) runs over the results of the correlation section and the two levels, read through each section's qualitySource() (AssetSetRiskPanel.svelte:625). The replay hands it no results (AssetSetReplaySection.svelte:127) and keeps its own status. Only partial results become badges; the two VaR instances are labelled Bad day and Bad month (VAR_LABELS, AssetSetComparisonLevels.svelte:158). The notice is blue, titled Holdings without prices…, only when every reason is no_price_source, and titled Worth knowing… when nothing is partial (levels/RiskPartialNotice.svelte:42, l.54).
  • One banner. The four controllers' dataQualityIssues are merged by mergeQualityIssues() (AssetSetRiskPanel.svelte:633) and rendered by DataQualityBanner in grouped mode. labQualityAction() (syncTargets.ts:78-82) maps any sync_* action to the lab's own sync, and navigate_asset, navigate_fx and add_fx_pair to navigation.
  • Frames. RiskLevelSection keeps under each section only what did not come back (levelErrorHealth(), levels/partialNotice.ts:49), in an amber alert (levels/RiskLevelSection.svelte:179), plus answer_discarded (ANSWER_DISCARDED_CODE, frontend/src/lib/stores/risk/riskPanelController.svelte.ts:57) once the controller's re-asks are exhausted — the bound and the wording are in The Panel Controller. The section bodies offer Retry (loadBase(true)) when they have no figure to show, for example AssetSetCorrelationSection.svelte:139, l.153.
  • Calculation details. The frame's <details> (levels/RiskLevelSection.svelte:229) lists the observations, coverage, annualization factor and return basis. levelMetadata() (levels/levelMetadata.ts:54) collapses analytics that agree into one row, and leaves out method.
  • Manual links. Each frame's book icon (docsPath) opens the theory: the correlation page, the risk metrics index for L1° and L3°, the historical replay page for L4°.

🛡️ Dashboard Risk Tab

The Risk tab of the Dashboard and of Broker Detail mounts levels/RiskLevelsPanel.svelte: one createRiskPanelController, a header (levels/RiskPanelHeader.svelte), the shared notice and four levels, L1–L4. The two hosts pass the same props with a different scope. What users see is in the Risk Tab user page; which analytics each level asks for is in By Level.

Host Scope What differs
frontend/src/routes/(app)/dashboard/+page.svelte:1040-1054 broker_ids: ownedBrokerIds (l.222): the brokers the user owns with a share above 0 (getOwnedBrokers(), frontend/src/lib/stores/reference/brokerStore.ts:262) The broker filter is not applied: subtitle becomes risk.dashboardFullPortfolio while it is on (l.1048), and scopeValue is null then (l.1046), because the summary follows the filter. assetIds are the summary's holdings (l.1045).
frontend/src/routes/(app)/brokers/[id]/+page.svelte:680-694 broker_ids: [broker.id] scopeValue is that broker's net worth (l.686); internalSubset prints risk.internalSubset (l.688).
  • No owned broker, no request. The Dashboard host mounts the panel only when canAsk (l.226); otherwise it shows common.noData (l.1056), or a spinner while the brokers load (l.1059).
  • Waves. The controller opts into includeDrawdownSummary, includeMonthlyVar and includeCurrentCompositionRiskReturn (levels/RiskLevelsPanel.svelte:85-103; the options are in The Panel Controller). The historical wave reads the portfolio's TWRR (The Primary Series), which is why deposits and withdrawals are neither gains nor losses here. Neither host sends a slice, so the backtest notice (backtestDeclared(), l.249, l.322) never shows on these pages.

📉 How Much Can It Hurt?

levels/L1HowMuchItHurts.svelte, over historical_var (two instances), drawdown_summary and historical_kpi (L1_CODES, levels/RiskLevelsPanel.svelte:125).

  • Cards. buildHurtRows() (levels/levelHelpers.ts:374) builds day → month → worst. The day and month cards lead with the CVaR at 95 % (l.384); the VaR is computed as secondaryLoss but never drawn, and appears only as the histogram's cut. The 30-day horizon is the lab's: see Requests, Bad month. A rung whose analytic failed is omitted, never zero-filled (l.369).
  • Worst fall. Depth and duration come from drawdown_summary, falling back to historical_kpi (levelHelpers.ts:410). The recovery needed line is requiredRecovery() (l.356: −50 % needs +100 %), drawn at L1HowMuchItHurts.svelte:169 and l.197.
  • Current drawdown. buildCurrentDrawdown() returns null at a peak (levelHelpers.ts:437), so the card exists only below one (L1HowMuchItHurts.svelte:190).
  • Tail measures as second rows. The worst realization sits under the day card; drawdown at risk and conditional drawdown at risk under the worst fall (L1HowMuchItHurts.svelte:174, l.180; buildTailMeasures(), levels/l1/l1Helpers.ts:227). The ulcer index is the caption of the underwater chart (levels/l1/UnderwaterChart.svelte:69).
  • Histogram. buildReturnHistogram() (l1Helpers.ts:136) flags the bin holding the VaR cut (holdsCut, l.92), drawn amber, and the bins wholly beyond it (belowCut, l.101), drawn red; the straddling bin is neither (levels/l1/ReturnHistogram.svelte:51). A bar's title gives its range (l.79).
  • Money. lossMoney() is scopeValue × loss (L1HowMuchItHurts.svelte:86), shown only for a positive scopeValue (l.65): on the Dashboard, never while a broker filter is on.
  • Labels and links. Each card prints a technicalName line (frontend/src/lib/components/ui/display/RiskMetricCard.svelte:107, keys risk.levels.l1.technical.*) and links its own metric page (DOC_PATHS and TAIL_DOCS, L1HowMuchItHurts.svelte:117, l.127).

🧩 Diversification

levels/L2Diversification.svelte, over risk_contribution (current composition, portfolio scope only: The Analytics) and correlation (historical wave).

  • Rows. buildDivergenceRows() (levels/levelHelpers.ts:476) drops a row without weight or contribution (l.485) and sorts on the signed gap contribution − weight (l.496). Eight rows show before Show all (visibleRows, L2Diversification.svelte:77, l.338).
  • Lead. leadDivergence() returns the top row only when its gap reaches 5 points (minimumGap, levelHelpers.ts:507); levels/RiskLevelsPanel.svelte:307 words it.
  • Cards. buildConcentration() (levelHelpers.ts:556) returns the effective number of assets and the diversification ratio together or not at all. The first card's caption compares the index with the measured rows — above, below or equal (effectiveAssetsReading, L2Diversification.svelte:149). uncoveredWeight() (levelHelpers.ts:601) splits cash_weight into cash and excluded_weight, worded by uncoveredCaption (L2Diversification.svelte:172). A card whose metric the scope lacks is not drawn (l.243).
  • Matrix. The shared heatmap (Correlation Section) receives only output and assetLabels (L2Diversification.svelte:363): no type, sector or region data, so only similarity and name are offered and no exposure badges show. When no pair reaches NEAR_IDENTICAL, a line says so (hasRedundantPair, l.114, l.359). On both surfaces, a click on a listed pair reveals, highlights and opens its cell, and a second click closes it (selectPair(), CorrelationHeatmap.svelte:393).

⚖️ Being Paid

levels/L3Benchmark.svelte (the picker) and levels/L3RiskAdjusted.svelte (table and chart), over the current-composition historical_kpi and asset_risk_return and the on-demand comparison.

  • Perimeter. selectKpiWave() (levels/l3Helpers.ts:64) reads historical_kpi from the current-composition wave, falling back to the historical one; the title appends risk.levels.l3.perimeter.* from the answer's metadata.mode (L3RiskAdjusted.svelte:59-61).
  • Benchmark. Picker and store are Benchmark and Eligibility. Instead of the lab's labL3Waits() (Requests): measuredAssetIds={[]}, so a held asset may be the benchmark (L3Benchmark.svelte:158); an effect asks the comparison once per baseEpoch|benchmark when the picker reports set and the catalogue is ready (l.83-93); any other state bumps the generation and drops the answer (l.101-107).
  • Rates. The comparison runs on current_composition with the page's risk-free rate and target_annual_return: 0 (L3Benchmark.svelte:133). That rate is appliedRiskFreePercent = 0, with no control (levels/RiskLevelsPanel.svelte:77): the same zeros as the lab (Requests, Rates).
  • Table and chart. The shared RiskReturnLevel (Risk and Return (L3°)), with capabilities={{weight: true, ratios: true, benchmark: benchmarkApplies}} (L3RiskAdjusted.svelte:113). benchmarkApplies is a measured beta or a benchmark row (l.97), not the lab's status === 'ok'. The portfolio gets a reference row from the KPI (l.93) and a dot from asset_risk_return's portfolio figures (levels/l3Helpers.ts:142-151), so with no measured benchmark the line runs through the portfolio instead of vanishing. weight: true adds the weight column and the dot-size note (riskReturnLevel.ts:187); widths and visible columns persist under storageKey="risk-l3" (L3RiskAdjusted.svelte:117), and the column menu sits in the level's header (levels/RiskLevelsPanel.svelte:342-346). Only a click on a column title sorts (RiskReturnLevel.svelte:14). Dashes: a blank figure has the blankNote tooltip (l.227), an uncalculated ratio a bare dash (l.224), the benchmark's own beta and correlation referenceItself (l.212), and a benchmark nobody holds notHeld under the weight (l.216).

🔮 What If…?

levels/L4WhatIf.svelte with three snippets — L4Replay, L4Shock, L4Simulation (levels/RiskLevelsPanel.svelte:366-376) — in a collapsible frame that loads the scenario catalogue on first open (l.364).

  • Tool selector. With more than one snippet the level is selectable (L4WhatIf.svelte:88): one Add: button per tool, always in L4_TOOLS order (l.7). The open set is read at mount from risk.l4.openTools in user storage (l.13, l.90, l.100), so the Dashboard and every broker page share it, and written by add() and close() (l.109, l.113); close() also resets that tool's analysis. Folding the frame unmounts the body (levels/RiskLevelSection.svelte:170) and keeps the controller's answers.
  • Replay. Controls and lifetime are the lab's (Replay (L4°)); coverage, exclusions and the suggested period are in Replay Coverage. On a weighted scope the excluded weight becomes a zero-return residual (backend/app/services/risk_plugins/stress/plugin.py:562), worded by excludedHeader() (levels/l4/L4Replay.svelte:205). showMoney follows the result's scope (scopeKind === 'portfolio', l.100–101) unless the parent passes showMoney, so the total sentence (l.272–273) and the amounts show here.
  • Shock. A preset click adopts the scenario's dimension and bucket shocks and runs it (levels/l4/L4Shock.svelte:82); nothing runs without buckets (l.73); editing a bucket clears the preset and the answer (l.91); a note shows when classification_coverage < 1 (l.170).
  • Simulation. Defaults and the five modes are in Simulation Parameters. In the UI the sampling select exists for GBM only (levels/l4/L4Simulation.svelte:62), the seed input is hidden on QMC (l.64), and SOBOL_START_INDEX is fixed at 0 (l.45). The answer shows the terminal mean, the probability of loss and the P5–P95 band (l.216-222), the drift-uncertainty note (l.225) and SimulationProvenance (l.244). The beta banner and the model warning are drawn in the simulation box only (L4WhatIf.svelte:156-160).
  • Disclosure. L4 keeps its own: every degraded status, named per step (L4_STEP_LABELS, RiskLevelsPanel.svelte:229), its reasons (l.239), its errors with discarded runs (l.240) and its metadata (l.241). The replay enters through replaySectionView() (l.214), so its warnings are read in the replay block alone.

🚩 Notices and Refresh

  • Header. levels/RiskPanelHeader.svelte draws the title and subtitle, Sync and Refresh (l.84-104), the load error (l.109-113) and the DataQualityBanner in grouped mode over the controller's merged issues (l.116). Instead of the lab's labQualityAction(), handleQualityAction() (l.66) opens the header's own PageSyncModal for any sync action and navigates for navigate_asset, navigate_fx and add_fx_pair.
  • Sync targets. syncAssets are the host's assetIds — on the Dashboard the summary's holdings, which follow the broker filter — and syncFxPairs the configured pairs between their currencies and the target (RiskPanelHeader.svelte:38-64); the button is disabled when both are empty (l.89). After a sync, handleSynced() marks the answers stale and the host's onsynced reloads its own data (Dashboard host l.1050, Broker Detail host l.690).
  • Notice. The shared notice is described in Notices and Frames. Here it reads the L1 results, risk_contribution, correlation and the L3 results (levels/RiskLevelsPanel.svelte:154), groups each partial badge under the first level that renders it (levelOf, l.162), and names the two VaR instances and L3's KPI as the levels show them (L1_LABELS, l.128; l.141). The L1–L3 frames keep only unavailable and failed (levelErrorHealth, l.334-346); L4 keeps its own (above). Calculation details are per level (l.191-193, l.241), as in Notices and Frames.
  • Refresh and retry. There is no Retry button. The header's Refresh is loadBase(true) (levels/RiskPanelHeader.svelte:98); the Dashboard's own Refresh bumps refreshVersion (Dashboard host, l.695-698), which the controller turns into loadBase(true) (frontend/src/lib/stores/risk/riskPanelController.svelte.ts:647-656). Re-asks, answer_discarded, the page cache and loadError are in The Panel Controller. Here onrefreshfailed toasts risk.states.loadFailed (RiskLevelsPanel.svelte:82), setTweenHydration() stops cached cards from counting up (l.106), and loadError replaces the notice and the levels (l.328).
  • Signature. A new period, currency or scope discards the on-demand answers and relaunches only those in flight (The Panel Controller); L3's comparison is asked again by its own effect (levels/L3Benchmark.svelte:84-92).
  • Manual links. Every level's book icon opens the risk-metrics index without an anchor, which would break on the translated sites (levels/RiskLevelsPanel.svelte:331-333); cards and second rows link their own metric pages (levels/L1HowMuchItHurts.svelte:117, l.127; levels/L2Diversification.svelte:258, l.268, l.284).