User Guide · every surface, every metric

GolfIQ is a comprehensive golf-intelligence dashboard built on a complete mirror of a live tour-stats feed. Every page reads from a Postgres database that refreshes every 15 minutes during live events and daily otherwise.

1.1 The sticky banner

At the top of every page, you’ll see a glowing header with:

• Eyebrow tag — tour + event status, e.g. PGA · ACTIVE EVENT
• Event title — the current tournament name and (in cyan) the course name
• Snapshot timestamp — when the model last refreshed
• “Live mirror” badge — confirms data is flowing
• Tour pills — switch between PGA, DP World, Korn Ferry, Opposite, Alt, LIV. Pills only light up for tours that currently have data.

1.2 Conditions strip

Thin metadata bar under the header showing tour · event · date. Holds the slot where live wind / rain / temp / bias will land once OpenWeather is wired (v3.1).

1.3 Narrative hero

The single biggest model-vs-market edge among the current event's top contenders, picked automatically. Headline names the player. Sub-line shows MODEL % · MARKET % · EDGE pp. Below: a one-sentence GolfIQ "why" and an Open profile → CTA.

1.4 Field at a glance (3-col grid)

• Pre-tournament decomposition — recharts stacked horizontal bars for the top 15: BASELINE + HIST + FIT + MAJOR + OTHER contributions to final win prob. Hover for the per-component breakdown.
• Risers / Fallers — two columns, 5 rows each. Player + current rank + delta + inverted rank sparkline (up = improving). Last 14-day window.
• Course-fit movers — plain-SVG slope chart, baseline rank → fit-adjusted rank. Green up = course suits the player, red down = course fades them.

1.5 Course block

Wraps the existing hole heat map with a 5-pill SG weights row (OTT/APP/ARG/PUTT/Scrambling) and a top-3 "watch holes" card grid. Self-resolves the slug from the active event so the section just works for any tournament.

1.6 Event mode module

Pre-tournament vs live, picked by event.isLive:

• Pre-tournament — pairings table (group, tee, hole, players).
• Live — top-20 leaderboard slice (pos/total/thru/ win%/SG total) + live contender trajectories (multi-line win% chart, 6 contenders, time on x-axis) + cut-line tracker (projected cut score + ±1 bubble strip during Thu-Fri).

1.7 The Bet Sheet

Horizontal snap-scroll strip of 8 cards. Each card: player name · market · DG fair % + American odds · best book + implied % · edge pip · 4-book chip grid · 5-pip confidence score.

1.8 GolfIQ on this week

Four chip questions seeded for the current event. Tapping any opens the existing GolfIQ chat surface with the question pre-filled — same chat trigger that fires from the insight cards.

1.9 Off-week mode

When no tour event is active, the HUB swaps to: a recap card for last week (winner + link to the model report card) · risers / fallers · "course of the week" pulled deterministically from course_archetypes. The Bet Sheet hides until there's a market to bet on.

Reached by clicking any player name across the app. Sections, top to bottom:

Reached from the Courses directory or by clicking course names in the History tab or live event banner.

Header

• Event name (current event, if any) + course name
• Stats strip — active event: field size, avg adjustment, max boost, biggest fade. Always: events on file, total rounds, avg score, year span. Plus derived Par + Yardagepills summed from the heat map's most-recent round.

Skill importance (legacy bar view)

Bar chart with four horizontal bars: OTT, APP, ARG, PUTT — each showing what fraction of the spread between players at this course is attributable to that skill (variance-based). Kept alongside the radar below as a quick scan view.

This week's best/worst fits

Two side-by-side tables. Best fits = largest positive total adjustment from baseline. Worst fits = largest negative. When there's no active event for this course, both tables fall back to all-time leader / laggard rows from the per-course history aggregate.

Course DNA radar

5-axis radar — OTT / APP / ARG / PUTT / Scrambling — with a dashed tour-average reference polygon at 20% per axis. Auto-generated sentence below the chart calls out the dominant skill weight at this course (e.g. "this course rewards approach play 2.3× more than average"). Confidence pip (low / medium / high) based on events on file.

Course over time

Single-line chart, x = year, y = field-avg SG vs par. Hover for the per-year round count. Reads from mv_course_difficulty_history.

Hole-by-hole scorecard

18-cell grid (3 cols on mobile, 6 sm, 9 lg). Each cell: hole # · par · vs-par SG · difficulty rank pip. Diverging red↔green background by vs-par sign (over par = red, under par = green), clamped to ±0.5 strokes. Per-hole drawer with scoring distribution + best/worst players queued for v3.1.

All-time leaders

20-row table from mv_player_course_career (rounds ≥ 4). Player headshot, rounds, events, avg SG/round, first/last year played. Click any name to open their profile.

Baseline → Final slope (current event only)

Slope chart from this course's active event — top 10 movers in baseline → fit-adjusted rank. Green up = course suits the player, red down = course fades them. Same primitive as the HUB's course-fit movers panel.

Historical leaders + nemeses (legacy)

Kept alongside the new all-time leaders table. Bottom 5 players with 2+ starts — venues that have consistently chewed them up.

Reached by clicking any SG ▸ row in the History tab. The single best way to put any individual historical event under a microscope.

Model report card

4-tile grid + reliability diagram, computed live from the full leaderboard's predicted_win × fin_num == 1:

Reliability diagram

Recharts composed chart in the right column. Predicted win % bucketed into 10 deciles on the x-axis, observed win frequency on the y-axis, with a dashed 45° "perfect calibration" reference line. Hover any bucket to see how many players were in it.

Leaderboard

Full historical leaderboard (all players, not just 80): finish position, name + headshot, predicted win %, round-by-round scores, SG breakdown averaged across rounds.

Model accuracy plot

Scatter chart: predicted win % on X-axis, actual finish position on Y-axis. Each dot is a player. Top-10 finishers in green, the rest in cyan. A well-calibrated model clusters dots in the upper-right (favorites finished high).

Closing-line value

Top 10 players by model edge vs. bet365 close (DG fair probability − closing-line implied probability). Columns: player, open/close odds, edge in pp (capped at ±15pp for sanity), finish, bet result.

Past-dated events never appear in Upcoming any more — NULL or stale event_date rows get filtered out and logged. Clicking any event opens its full detail page (/dash/tournaments/[event_id]) with the live or pre-tournament leaderboard, betting markets, and tee pairings.

On the detail page — live sections

When event.isLive is true, two extra cards mount between the leaderboard and the odds module:

• Live contender trajectories — multi-line win-% chart for the top 6 contenders, sourced from dg_in_play_history (which is fed by the 5-min snapshot cron during Thu-Sun 11-23 UTC).
• Cut-line tracker— latest snapshot's projected cut score + ±1 bubble strip with player names. Hides until 65+ players have posted live scores.

The odds module never lands on "No win market." any more — when dg_outrights is empty, each card falls back to a 20-row DG-fair list derived from the model probabilities and labels itself "DG fair only · book lines unavailable".

Press ⌘K (Mac) or Ctrl K (Win/Linux) from anywhere in the app, or click the floating logo at the bottom-right. A slide-out drawer opens with:

• The current page path shown as a context chip in the header
• 4 suggested prompts you can click to send
• A free-text input
• Open full link if you want the full-page chat at /golfiq
• Esc button (or just press Esc)

What you can ask

Tips for getting the best answers

• Be specific about the player. “Scheffler” works, “Scottie” works. Two-word ambiguous queries (“Brown”) may resolve to the wrong player — GolfIQ will list matches.
• Reference the current page. Because the chat knows your pathname, you can ask “explain this player’s edge” on a player profile and it will use that context.
• Ask follow-ups. The chat maintains conversation history. “Now compare him to Rory” works.
• Use it during research. Hit ⌘K from any page — faster than navigating back and forth.

Strokes Gained (SG)

The fundamental performance metric in modern golf. SG measures how a player performs vs. the PGA Tour field average on the same shot. Positive = above average, negative = below.

Model terms

Betting terms

DFS terms

• DK / FD — DraftKings / FanDuel
• Salary efficiency — projected fantasy points per $1,000 of salary. Higher = better value.

Glossary of DG markers

• sg_categories=‘yes’ — archived event has full SG breakdown available
• traditional_stats=‘yes’ — driving accuracy, GIR, proximity, scrambling recorded
• archived_preds=‘yes’ — pre-tournament predictions backfilled for this event

• Click any row in the History tab. Drills into the archived leaderboard with model accuracy.
• Click any course name in This Week’s decomposition table. Goes to the venue profile.
• Use the tour pills. The home dash is fully tour-scoped — pick LIV or DPWT to see those tours’ predictions.
• Sort the Players tab by Δ. Top-of-the-table = biggest weekly rank improvers. The forming bubble.
• Sort DFS by Value. Anything above 5.5 points per $1K is high-leverage; the value column is the single best DFS sort.
• Look at the calibration chart on the Backtest page. If a bucket’s green bar is taller than its cyan bar, the model under-predicted that group’s win rate (those longshots are real value). Inverse if green is shorter.

• Off-week dashboard — recent winners, biggest skill movers, archive featured event when no live tournament
• Shot-level visualizations from the historical-rounds payload
• User accounts + watchlists — pin players to a personal home view
• Multi-book historical odds — DK, FD, Pinnacle closing lines alongside bet365
• Auto-refresh of the Live tab during active rounds (no page reload required)