story-lines

name: story-lines description: Detect ELO-based narrative story angles for F1 races and seasons. Use when the user asks about driver momentum, surprise performances, teammate battles, hot streaks, slumps, or wants story lines written to the workspace. allowed-tools: Bash(pitlane *), Read, Write

F1 Story Lines

You detect and synthesize ELO-based narrative story angles for F1 races. Story lines are grounded in the calibrated endure-ELO model and represent statistically significant deviations from expectation.

Detecting Story Lines

For a specific race

pitlane stories detect --year 2026 --round 3

Returns JSON saved to {workspace}/data/stories_2026_3.json.

Options:

• --session-type R (race, default) or S (sprint)
• --trend-lookback 3 — how many prior races for momentum delta (default 3, use 6 for longer view)

For a full season

pitlane stories season --year 2025

Returns the top 10 signals (by |value|) across the entire season, saved to {workspace}/data/stories_2025_season.json.

Options:

• --top-n 15 — change how many top signals to write (default 10)

Use pitlane stories detect --year Y --round R to get full signal detail for any individual round.

Reading the Results

After running the command, read the output file:

Read {workspace}/data/stories_2026_3.json

Before Interpreting Surprise Signals

After running pitlane stories detect, use the f1-analyst skill to load the race session before analyzing surprise signals. You need to know the actual grid order, retirements, safety cars, and lap count context before judging whether a surprise result is a story.

Signal Types

Each signal has a signal_type. Interpret them as follows:

hot_streak

• What it means: A driver's ELO rating has risen significantly over the last N races (default: 3).
• value: ΔR̂ (positive). Higher = stronger momentum.
• Threshold: 0.5 ELO
• Story angle: Career breakthrough, form revival, new team adaptation succeeding, regulation beneficiary.
• Example narrative: "Norris is on the hottest streak in the field — +0.72 ELO over 3 races heading into Monaco."

slump

• What it means: A driver's ELO has dropped significantly over recent races.
• value: ΔR̂ (negative). More negative = deeper slump.
• Threshold: −0.5 ELO
• Story angle: Car reliability crisis, driver under pressure, team strategy failures, form collapse.
• Example narrative: "Alonso has shed 0.61 ELO over 3 races — the deepest slump on the grid right now."

surprise_over

• What it means: A driver finished far ahead of their predicted position (SurpriseScore < −2.0).
• value: SurpriseScore (negative = better than expected).
• context.expected_position: Model's predicted finish rank.
• context.actual_position: Where they actually finished.
• Story angle: Giant-killing, tactical brilliance, wet-weather specialist, Safety Car timing.
• Note: Cross-reference with race control data to check if a Safety Car or unexpected incident explains the result before calling it a genuine story.

surprise_under

• What it means: A driver finished far behind their predicted position (SurpriseScore > 2.0).
• Story angle: Shock retirement, car failure after strong qualifying, strategic disaster.
• context.status: Raw retirement string from the data (e.g., "Engine", "Collision", "Retired"). This is your first clue. Note that dnf_category is unreliable for 2022+ data — all retirements may be classified as "retired" regardless of cause. Do not use it as the final word.
• Always use the web-search skill to find the actual reason behind any DNF: search for [driver] [grand prix year] DNF to find FIA reports, team radio, and post-race statements. A "Retired" status with no further context is not a complete story.

teammate_shift

• What it means: Within-team ELO gap has been consistently one-sided (or just flipped).
• context.current_delta: Current gap (positive = driver_id leads teammate).
• context.historical_deltas: Gap over the last N races (newest first).
• Story angle: Intra-team power shift, new driver establishing hierarchy, team's primary driver under threat.
• Flip signal: When historical_deltas are all one sign but current_delta is the opposite, it's a power reversal story.

Writing Story Lines to Workspace

Once you have the signals, synthesize them into narrative prose and write to the workspace:

Write {workspace}/data/story_lines_2026_3.md

Good story line format:

# Story Lines: 2026 Round 3

## 1. [Driver] — Hot Streak (Hot_streak, ΔR̂ +0.72)
[2-3 sentences contextualizing the ELO trend with race results. What happened
over the 3 races? What does this mean for the next race? Is this a car story
or a driver story?]

## 2. [Driver] vs [Teammate] — Internal War Tipping Point (teammate_shift)
[Explain the gap direction and what it means for team dynamics.]

Contextual Checks

Before finalizing a story line, consider:

1. Is this race on a wet or street circuit? Check context fields — wet races and street circuits can inflate/deflate performance deviations.
2. Did a Safety Car or red flag affect the result? Use the race-control skill to cross-check surprise signals against race incidents.
3. Is the driver new or recently changing teams? A high k-factor (uncertainty) means the surprise threshold is wider — check pre_race_k in context.
4. DNF reason: For surprise_under signals, check context.status for the raw string, then always web-search for the actual cause. dnf_category is not reliable for 2022+ data.

Signal Priority

When writing story lines, prioritize signals in this order:

1. teammate_shift with flipped gap — rarest and most dramatic
2. Strong surprise_over (SurpriseScore < −3.0) — unexpected winner or giant-killing
3. hot_streak with ΔR̂ > 1.0 — multi-race dominance
4. surprise_under where status is not a mechanical failure — crash or strategy disaster is more of a story
5. slump with ΔR̂ < −1.0 — crisis narrative

Notes

• Story signals are sorted by |value| descending in the JSON — the first signal is the strongest.
• The narrative field in each signal is a starting point, not a final story line. Enrich it with race results context.
• Signals reflect pre-race ELO (what the model expected going in), making them true out-of-sample assessments.
• The underlying model is calibrated endure-ELO (k_max=0.8665, φ_race=0.999) — see docs/F1_ELO_Story_Detection_System_Design.md for full methodology.