Wireframe phaseDummy data onlyUnified arm views

Resolution Yield Operator Dashboard

This version is about structure, not real data yet. The goal is to review the layout as an operator tool first, with the right panels, comparisons, and drill paths already in place before we connect the reporting database.

Screen behavior

One interface for normal operation and A/B review. If only one arm is active, the same cards and charts collapse down cleanly.

Active viewCommand center

Command CenterExperiment ReviewFunnelExecution & RiskLanesReplay

Principle

Trust the numbers

Every key number on the dashboard should come from the bot's canonical SQLite reporting store, with clear caveats when a metric is still proxy-derived.

Principle

See the funnel clearly

The dashboard should make it obvious where opportunities are dying: hard reject, duplicate hygiene, planning, confirmation, execution, or exits.

Principle

Compare arms naturally

Arm comparison should live inside shared charts and tables rather than in a separate experiment-only interface.

Principle

Track unlocks, not just outcomes

We need to watch the specific operational unlocks that matter for this bot: duplicate churn, soccer late-funnel friction, NBA protection, execution quality, and replay trust.

Top navigation

Dashboard skeleton

Command CenterFunnelExecution & RiskLanesReplay

Window: 24h

Mode: paper

Arm view: all active arms

Category: all

Subtype: all

Health-first overview

Command Center

These cards are placeholders for the exact first-screen information we want: status, PnL, exposure, trust, and clear warnings.

Live later:
DB-backed cards, recent update timestamp, and alert count.

Bot status

Healthy

green

Heartbeat, orderbook freshness, and core monitoring all clear in this placeholder view.

99.4% healthy cycles

Realized PnL

$684.12

yellow

Stand-in figure for rolling operator review, not a real live number yet.

+$28.7 / 24h

Open exposure

$14.8k

yellow

Should later show book size, open positions, and crowding sensitivity together.

100 open positions

Replay trust

89%

green

Placeholder trust headline for replay-vs-live alignment and config coverage.

fresh batches improving

Shared multi-arm chart

PnL over time

mainAB

MonTueWedThuFriSatSun

Live alert stack

Attention queue

Confirmation friction rising in soccer winner

12h

This area should eventually link straight into the subtype drill-down and blocker table.

Replay config coverage below target

24h

A trust issue should feel first-class, not buried in a secondary report.

Protective exits stable

Healthy signals should sit beside warnings so the first screen does not feel purely alarming.

Flow diagnosis

Funnel and blockers

This section should make it obvious where opportunities die and whether B is improving the lanes it is meant to improve.

Stage flow

Wireframe bars

Discovered

All candidate opportunities seen by the bot

94,368

100%

Qualified

Survived hard reject and quality gating

5,982

6.3%

Planned

Made it through duplicate and planner controls

188

3.1% of qualified

Confirmed

Confirmation succeeded and trade remained viable

153

81.4% of planned

Executed

True late-funnel success

20.9% of confirmed

Top blocker reasons

Needs drill-down

PLANNING_MARKET_ALREADY_OPEN_OR_PLANNED

Duplicate-thesis pressure still visible in winner-side lanes.

418

down in B

RECENT_MARKET_ATTEMPT

Should improve if cooldown handling is now cleaner in the retuned lanes.

256

mixed

EXPECTED_FILL_GT_CAP

Late-funnel friction to watch closely in soccer winner flow.

watch

CONFIRMATION_FAILED

Needs eventual breakdown by source and subtype.

flat

Quality and downside

Execution and risk

These should eventually become real charts and tables for fills, entry quality, concentration, and protective exits.

NBA winner protection

4 checks

Real lane, but concentration-sensitive and vulnerable to ugly downside if we overtrust it.

Q -> P -> X conversionprotective exitscategory / cluster concentrationrealized PnL variance

Execution realism and fill quality

4 checks

If execution assumptions are wrong, tuning decisions get misleading quickly.

avg fill completion pctavg entry gap bpsexpected fill vs actual fillfee-aware execution economics

Replay trust

5 checks

Replay should become a real tuning tool, but only if it lines up with live truth.

qualification match ratequality reason match ratequality score match rateconfig snapshot coveragelate-funnel replay/log consistency

Reference map

KPI framework

Outcome KPIs

Realized PnL

yellow

Positive on rolling 24h, 7d, and experiment-since-start windows.

Total PnL

yellow

Realized plus unrealized should not be hiding large drawdowns behind open positions.

Protective-exit drag

yellow

Protective behavior should stay useful without dominating the bot's outcome shape.

Throughput KPIs

Qualified -> Planned

green

Should improve when duplicate/planner hygiene improves, without fake improvement from over-filtering.

Planned -> Confirmed -> Executed

green

Especially important for soccer winner and execution-friction review.

Trades per active arm

red

Comparison views should make thin samples obvious without needing a separate arm page.

Efficiency KPIs

Duplicate blocker rate

green

Should fall where planner hygiene improves, especially in winner-side lanes.

Expected-fill blocker rate

yellow

Should fall where we are deliberately reducing late-funnel friction.

Fee-aware economics

yellow

Executed trades should remain worthwhile after fees, not just before them.

Trust KPIs

Replay qualification match

green

Near-perfect. If replay cannot reproduce qualification, tuning trust stays low.

Replay config snapshot coverage

yellow

Fresh batches should all have exact config provenance attached.

Canonical reporting coverage

green

Important analysis should come from SQL first, not Telegram archaeology.

Deep dives

Lane snapshots

sports_nba / winner_moneyline

Duplicate/planner hygiene and downside protection

watch closely

Q -> P -> Xprotective exitscategory concentrationPnL variance

sports_soccer / winner_moneyline

Late-funnel friction and fill quality

priority lane

planned -> confirmed -> executedentry gap bpsconfirmation failuresfill completion

crypto threshold / range

Transport reliability, confirmation source mix, and controlled sizing

guard rails

source fallbackfee-aware flowconfirmation agesized executions

Replay trust

Validation panel

Qualification match rate

97.8%

Later this should link to replay validator output and exact windows.

Config snapshot coverage

84%

Fresh batches should carry exact provenance; older ones may remain fallback-derived.

Late-funnel consistency

watch

Replay versus log or DB consistency should become easy to read at a glance.

Operator table

Loss review placeholder

Dummy rows

MarketArmExit typePnLHold

NBA: Knicks vs Heat winnerBinvalidation-$42.103.8h

Soccer: Milan vs Roma winnerAstop_loss-$31.886.1h

Crypto: BTC > 72kmaincritical_exit-$18.251.4h

Metric hierarchy

From summary to drill-down

Top level

8 items

bot statusrealized PnLunrealized PnLopen exposureprotective exitsPnL by armtrades by armtop blocker rates

Diagnostic

6 items

full funnel countsconversion ratesconfirmation failures by sourceexecution qualityfees on qualified/planned/confirmed/executed flowcategory and subtype leaders/laggards

Drill-down

7 items

subtype round review metricscluster concentrationresolution bucket exposureentry-price bucketsloss review by exit typeposition lifecycle and hold timereplay validator stats