fct_driver_isolation_lap

Driver isolation per Ω lap with the rolling rating: the lap’s three ratings, its tier-3 peer aggregates and identity split, and a trailing window per rating (the Ω laps of the same stint with lap_number in [l - 4, l]; never forward, T44), each NULL below var(‘isolation_window_min_laps’) contributing laps, with SE, lambda, a shrunk value, confidence_pct and trust_label.

Part of the Marts family.

Driver isolation per Ω lap with the rolling rating: the lap’s three ratings, its tier-3 peer aggregates and identity split, and a trailing window per rating (the Ω laps of the same stint with lap_number in [l - 4, l]; never forward, T44), each NULL below var(‘isolation_window_min_laps’) contributing laps, with SE, lambda, a shrunk value, confidence_pct and trust_label. The 5-lap window is about half signal: the ‘is he on it right now’ number, not the ‘who is better’ number; verdicts belong at stint or race grain. Positive = faster. Leakage: a function of the residual trajectory; no ML-contract mart may depend on it (T48, ml/tests/test_manifest_contract.py). WI-16a, _roadmap/_fixes/wi/WI-16-cumulative-driver-isolation.md.

Lineage

Overview

Property	Value
Layer	marts
Family	Marts
Contract enforced	No
Tags	driver_isolation
Upstream	int_driver_isolation_lap_values, fct_driver_isolation_pair_lap, int_lap_telemetry_aggregates

28 tests guard this model - 23 generic, 5 singular.

Columns

Column	Type	Nullable	Tests	Description
lap_id		Yes	not_null, unique	Lap identifier (int_lap_residual_decomposed.lap_id).
stint_id		Yes	not_null	Stint identifier (int_stint_geometry.stint_id).
race_year		Yes	not_null	Season.
race_id		Yes	not_null	FastF1 event id, e.g. 2021_8.
driver_id		Yes	not_null	Driver (three-letter code). In the pair mart, the focal driver whose gains the row reports.
constructor_id		Yes	not_null	Constructor (stg_laps.constructor_id).
stint_number		Yes		Stint ordinal within the driver’s race (int_stint_geometry).
lap_number		Yes	not_null	Race lap number. Tiers 1 and 3 match cars on it, which is why 2020_1 is excluded (W18).
compound		Yes	not_null	Dry compound of the stint (INTERMEDIATE and WET laps are not in Ω).
age_in_stint		Yes	not_null	Tyre age in laps (int_stint_geometry).
valid_lap_in_stint		Yes		Ordinal among the stint’s valid laps; early vs mid phase splits at 5/6.
laps_past_cliff		Yes		Laps beyond the compound seed’s cliff onset (0 before it); drives tyre_phase = cliff. Never NULL in Ω.
era		Yes		Regulation era: pre<era_boundary> or post<era_boundary> (var era_boundary, 2022).
tyre_phase		Yes	accepted_values	early (laps_past_cliff = 0, valid_lap_in_stint <= 5), mid (laps_past_cliff = 0, >= 6) or cliff (laps_past_cliff > 0), from the seed’s onset, so never selected on the residual.
is_recovery		Yes		TRUE on the first var(‘isolation_recovery_laps’) valid laps of a non-dirty-air run directly after >= var(‘isolation_recovery_min_dirty_run’) consecutive dirty-air valid laps in the stint. Backward-looking only.
stint_phase		Yes	not_null, accepted_values	’recovery’ when is_recovery, else tyre_phase (T45).
air_state_dominant		Yes		int_lap_air_state.air_state_dominant (context; dirty_air drives the recovery overlay).
is_dirty_air_lap		Yes		TRUE when air_state_dominant = ‘dirty_air’.
field_n		Yes		Ω laps on this (race, lap); >= var(‘isolation_min_field_n’) by construction (T40).
lap_time_s		Yes		Raw lap time, s.
compound_component_s		Yes		C: the compound seed’s expected cost of this tyre state, s, positive = slower (int_lap_residual_decomposed). Never COALESCEd: an unknown tyre is not in Ω.
dirty_air_tax_s		Yes		D: dirty-air tax, s, positive = slower (int_lap_residual_decomposed).
y_s		Yes		x_s minus the lap’s Ω field median of x_s, s. POSITIVE = SLOWER than the field- typical car and driver on the lap. Invariant to the field base, fuel, rubber and ambient terms, which depend on (race, lap) only (T41).
car_iso_s		Yes		Isolation car term, s, NEGATIVE = FASTER than the race’s lap-weighted car (int_constructor_car_fe_isolation). NULL when not identified.
car_term_source		Yes	accepted_values	Where car_iso_s came from: driver_id (fitted per-era), unidentified (multiple components in race), not_estimated (pyfixest singleton), or no_fit_row (no row in the fit).
tyre_offset_vs_field_s		Yes		compound_component_s minus the lap’s Ω field median of it, s; positive = this tyre state costs more than the field’s. The strategy position on the lap: team-dominated context, not a rating.
lift_coast_share		Yes		int_lap_telemetry_aggregates.lift_coast_share (context; fuel- and tyre-saving are indistinguishable in it).
lift_coast_excess_share		Yes		lift_coast_share minus the lap’s Ω field median of it (context for WI-16b’s V2d; aggregates are means).
pace_isolated_gain_s		Yes		p = -(y_s - car_iso_s), s/lap. POSITIVE = FASTER than the field on an equal car, tyre state, traffic and fuel: the driver plus noise, everything modelled removed. NULL without a car term (never a car term of 0).
pure_skill_gain_s		Yes		Tier 1, s/lap, POSITIVE = FASTER: pace_isolated_gain_s plus the declared season offset (int_driver_isolation_lap_values.pure_season_offset_gain_s; 2018 +0.37 s, W33). Within a season it ranks and differences exactly like p. NULL without a car term. On cliff laps it is not a driver measurement (pure_is_extrapolated) and pure aggregates skip it.
pure_is_extrapolated		Yes		TRUE on cliff-tyre laps (tyre_phase = ‘cliff’): the lap time there is dominated by the tyre, so pure_skill_gain_s is not a driver measurement; aggregates exclude these laps.
season_sigma_w_s		Yes		Pooled within-stint SD of p around its stint line for the season, s.
season_neff_factor		Yes		f = (1 - rho+)/(1 + rho+), rho+ = GREATEST(season_rho1, 0): the effective-sample factor every SE uses. NULL if rho cannot be measured.
n_peers		Yes	not_null	Same-strategy peers on this lap (pair-mart rows with this lap_id); 0 when none.
n_teammate_peers		Yes	not_null	Of n_peers, how many are teammates.
relative_pace_gain_s		Yes		Tier 3 on this lap: mean relative_pace_gain_s over its peers, s/lap, positive = faster than same-strategy peers. NULL with no peer.
relative_pace_raw_gain_s		Yes		Mean raw gap over this lap’s peers, s/lap. Equal to relative_pace_gain_s since W63 design 2 (2026-10-03).
identity_n_pair_laps		Yes		Pair-laps with both car terms, over which the identity_* contributions are computed.
identity_relative_pace_gain_s		Yes		Mean relative_pace_gain_s over the identity_n_pair_laps rows, s/lap. = the three identity_* contributions plus the group’s mean tyre-age bias (the seed’s pricing of the peers’ age gap, at most the match tolerance; not published, 0 when every pair is on the same tyre age). W63 design 2, 2026-10-03.
identity_pace_gap_gain_s		Yes		Contribution: SUM(pace_gap) over rows with both car terms / identity_n_pair_laps, s/lap.
identity_car_advantage_gain_s		Yes		Contribution: mean car_advantage_gain_s over identity rows, s/lap; > 0 = the driver’s car faster.
identity_traffic_advantage_gain_s		Yes		Contribution: mean traffic_advantage_gain_s over identity rows, s/lap; > 0 = he lost less to dirty air.
window_first_lap_number		Yes		First lap number the window could span (the earliest Ω lap in it).
window_n_laps		Yes	not_null	Ω laps in the trailing window (same stint, lap_number in [l - (var(‘isolation_window_laps’) - 1), l]).
window_mixed_phase		Yes	not_null	TRUE when the window’s laps span more than one stint_phase; the window is labelled with this lap’s.
window_n_pure_laps		Yes	not_null	Non-extrapolated laps with pure in the window; pure_skill_5lap_gain_s is NULL below var(‘isolation_window_min_laps’).
window_n_pair_laps		Yes	not_null	Pair-laps in the window (SUM of n_peers); relative_pace_5lap_gain_s is NULL below the floor.
window_n_relative_laps		Yes	not_null	Laps in the window with at least one peer.
pure_skill_5lap_gain_s		Yes		Pure pace skill, mean over the window’s non-extrapolated laps at trailing 5-lap window grain, s/lap, POSITIVE = faster / better (raw, before shrinkage).
pure_skill_5lap_se_s		Yes		SE of pure_skill_5lap_gain_s, s/lap (see the model header for the formula).
pure_skill_5lap_lambda		Yes		Reliability tau^2 / (tau^2 + SE^2) in [0, 1]; tau^2 per (season, rating) at this grain. 1 when SE = 0.
pure_skill_5lap_shrunk_gain_s		Yes		lambda * pure_skill_5lap_gain_s: shrunk toward 0 (the ratings are field- or peer- centred), s/lap.
pure_skill_5lap_confidence_pct		Yes		ROUND(100 * lambda * method_score); NULL until WI-16b validates the method (seed driver_isolation_method_scores).
pure_skill_5lap_trust_label		Yes	accepted_values	solid (>= 75), indicative (50-75), thin (25-50), suppress (< 25, grade F or no lambda), unvalidated (no method score yet). NULL without a raw value.
relative_pace_5lap_gain_s		Yes		Relative pace vs same-strategy peers, SUM(relative * n_peers) / SUM(n_peers) at trailing 5-lap window grain, s/lap, POSITIVE = faster / better (raw, before shrinkage).
relative_pace_5lap_se_s		Yes		SE of relative_pace_5lap_gain_s, s/lap (see the model header for the formula).
relative_pace_5lap_lambda		Yes		Reliability tau^2 / (tau^2 + SE^2) in [0, 1]; tau^2 per (season, rating) at this grain. 1 when SE = 0.
relative_pace_5lap_shrunk_gain_s		Yes		lambda * relative_pace_5lap_gain_s: shrunk toward 0 (the ratings are field- or peer- centred), s/lap.
relative_pace_5lap_confidence_pct		Yes		ROUND(100 * lambda * method_score); NULL until WI-16b validates the method (seed driver_isolation_method_scores).
relative_pace_5lap_trust_label		Yes	accepted_values	solid (>= 75), indicative (50-75), thin (25-50), suppress (< 25, grade F or no lambda), unvalidated (no method score yet). NULL without a raw value.