PUBLIC RESULTS. EXPLAINED.

A score with
its workings shown.

New here? Read the quick guide →

Manager Pressure Score (MPS) combines wins, draws and losses, adjusts for pre-match opponent strength, and adds club dismissal history. It is an experimental index, not a dismissal probability, an assessment of a manager’s absolute ability or a statement about a club’s intentions.

The overall score

MPS = round(clamp(0.60R + 0.30S + 0.10C, 0, 100))

Each component runs from 0 to 100. Higher scores mean greater pressure. Fifty is the centre of the performance transformation and small-sample correction. It does not mean results matched expectations or equal the average manager’s score. The club-history prior gives C=50. MPS is not a dismissal probability.

Pre-match strength and expected points

Elo begins at 1500 unless verified, permitted prior results or initial values are supplied. At each season boundary, ratings regress: 1500 + 0.85 × (previous − 1500). Promoted clubs without ratings use the previous league’s 25th percentile when available; otherwise 1500. These are weak starting assumptions.

Every expectation uses information available before the match. Matches at the same kickoff time share one pre-match rating snapshot. Cup matches, friendlies, incomplete results and missing scores are excluded.

D = Elo(team) − Elo(opponent) + H
r = 10^(D / 800), ν = 2 × 0.26 / (1 − 0.26)
Z = r + 1/r + ν
P(win) = r/Z, P(draw) = ν/Z, P(loss) = (1/r)/Z
Expected points = 3 × P(win) + P(draw)

H is +60 at home, −60 away and 0 at a neutral venue. An equal neutral match gives 0.37 win, 0.26 draw, 0.37 loss and 1.37 expected points. Elo changes by 20 × (actual result − expected result), where a draw counts as 0.5. The two changes sum to zero.

Three components

Start with each match

A win has a base pressure of 0, a draw 50 and a loss 100. Expected points x come from the same pre-match Elo calculation for both R and S. The result carries 80% of each match score, while opponent strength carries 20%.

q = 0.80 × result base + 0.20 × (100 × x / 3)

Wins score 0–20, draws 40–60 and losses 80–100 before aggregation. Opponent strength cannot reverse that ordering. Better results lower pressure when the other inputs are held constant.

R · Recent results · 60%

Use up to eight eligible league matches in the same club, league and continuous manager spell. Newest first, the weights are 0.85^i. The weighted mean of their match scores q is r.

S · Tenure performance · 30%

Use all available eligible league matches from that same continuous spell, including previous seasons. Newest first, the weights are 2^(−i/38): a match 38 eligible games ago has half the weight; 76 games ago has one quarter. Their weighted mean is s.

Season openings do not reset the aggregate. Time passing without a new match does not change these weights. Previous managers, other clubs, separate spells after a reappointment and other leagues do not contribute. Both R and S use the same changing pre-match Elo; there is no fixed season-opening or appointment anchor in this model.

Separate form and moderate small samples

The transformation spreads strong and poor form further apart, then pulls small samples toward 50. Let n be all available eligible matches in this continuous spell, including past seasons.

F(z) = 50 + 50 × tanh((z − 50) / 25)
ρ = n / (n + 1)
R = 50 + ρ × (F(r) − 50)
S = 50 + ρ × (F(s) − 50)

The first match produces a calculated score. Recent and tenure results overlap by design; R gives more weight to short-term changes. The formula does not target a fixed number of managers in any score band.

C · Club Dismissal Index · 10%

C is fixed from a dated research baseline. The observation periods, league prior and time weights are calculated once and saved with the resulting value. Daily match and manager updates do not recalculate it. A sourced correction or an explicit new review creates a dated revision; earlier score records keep their original values.

Confirmed in-season club dismissals across at most ten seasons are weighted by 0.5^(age in years/3). D is weighted dismissals and Y is verified club-season exposure. Unknown periods are not counted as dismissal-free seasons; lower-league history stays separate.

rate = (D + 5μ) / (Y + 5)
C = clamp(100 × rate/(rate + μ), 0, 100)

μ is the leave-one-club-out league rate with a floor of 0.05. Estimation requires at least five other clubs, 20 observed club-seasons and 90% verified coverage. Otherwise μ defaults to a provisional 0.5. No history gives C=50 and a history-unavailable flag. Fifty represents a prior mean, not a dismissal probability.

Position, losing runs, tenure and heavy defeats receive no separate points. They may help explain the results but do not add another penalty for the same poor performance.

Reading the categories

These are descriptive categories, not guarantees of safety or predictions of an imminent dismissal. Cross-league scores are not calibrated to the same absolute scale.

Missing data and manager changes

An unverified current manager receives no score. A confirmed manager spell with no eligible matches receives 50, labelled “Neutral baseline” and “Insufficient data”. All three components use 50 regardless of the available club-history value. This is a placeholder, not a measured assessment. The first eligible match starts the calculation; scores from 1 to 7 matches are provisional. Caretakers and interim managers appear separately. Tenure boundaries are half-open: [startAt,endAt). When only the appointment date is known, eligible matches that day are attributed to the new manager by assumption; the exact time remains unverified. Broken recent expected-point inputs stop scoring.

Fetch freshness, result timestamp, Elo warmup, adopted tenure periods, season counts and missing history are shown separately. Older results are added only when source access, stable IDs and match dates can be verified. Missing periods are never replaced with losses or zero results; scores use the available eligible matches. Coverage is not statistical confidence. Missing components keep their original weights. Neutral baseline entries are retained in the records but excluded from pressure-history charts, score-change comparisons and backtests. Available club-history research remains unchanged. Previous-manager scores remain attached to their own spells, and model revisions and additions of older match data are distinguished from new-match changes. Older score records retain their original model, season summaries and anchor information.

Pressure history is shown at match granularity: each eligible completed league match contributes one chronological calculated point, and the points are connected by a line. Missing or unconfirmed matches do not create zero or loss points. The chart is reconstructed from available results and distinguishes locally generated observations from historical reconstruction; it is not an archive of every score that was published on that date.

Confirmed matches when the appointment date is unknown

A confirmed current club–manager assignment may use individually evidenced completed league matches. R uses the latest eight confirmed matches; S uses all confirmed matches for that same assignment. The existing formula, weights, decay and small-sample correction apply. Games without explicit evidence for this assignment are excluded; no exact appointment date or uninterrupted tenure is inferred.

These scores use the same ranking, race selection, profile and Last 5 presentation as other current managers. Their separate score basis, adopted count, period and sources remain in the detailed data-quality record; the full-tenure denominator is unknown, and continuous losing or winless runs are not claimed. A confirmed identity with no eligible match evidence has no score. Score changes and history comparisons require the same club, manager, assignment and score basis; a verified tenure starts a separate comparison series. Partial histories are excluded from backtest performance claims.

Validation and limitations

Automated tests verify calculations, temporal boundaries and data handling with fictional fixtures. The backtest pipeline uses chronological splits, purges 30-day label overlap, separates competing departures and incomplete follow-up, and compares simple recent-PPG and position baselines. Fictional test output is not predictive evidence.

Injuries, contracts, finances, cup runs, dressing-room relationships and private club decisions are not included. Elo can change with form, and incomplete older history can affect tenure scores and sample correction. Wins and losses dominate by design; expectations are a smaller adjustment. Weights, sensitivity, decay, shrinkage and club-history assumptions need real-data validation. MPS is not a calibrated probability and is not assigned a Brier score.

Model changelog

mps-v2.1-experimental — adds explicit confirmed-match scoring for verified current managers with unknown appointment dates. The primary UI uses the standard manager presentation while the assignment basis remains available in data-quality details and separate from verified-tenure comparisons. The result-led formula and 60/30/10 weights are unchanged.

mps-v2.0-experimental — result-led match scores (80% result, 20% expected points), a more responsive transformation, and tenure performance across past seasons with a 38-match half-life. Both components use pre-match Elo. Scoring and ranking start at one match; only zero matches receive a neutral placeholder. Fixed club history and 60/30/10 weights remain.

mps-v1.5-experimental — confirmed managers with zero to two eligible matches receive a neutral 50, with an explicit insufficient-data label. Normal scoring starts at three matches. Baseline entries do not count as calculated pressure observations or predictive evidence.

mps-v1.4-experimental — date-only appointments include that day’s eligible matches under the new manager, with the assumption flagged. Fewer than three matches is labelled “Insufficient data”. The formula and 60/30/10 weights are unchanged.

mps-v1.3-experimental — C uses an explicitly researched, fixed baseline. Its effective date and missing-history reasons remain visible. R/S and the 60/30/10 weights are unchanged. A new baseline is never applied to a date before it took effect.

mps-v1.2-experimental — recent expected-points shortfall (60%), anchored season shortfall (30%) and club dismissal history (10%). Both performance components shrink toward 50. Position, streak, tenure and heavy-defeat additions are retired. Earlier score records retain their model version and recorded contributions; changes across versions are not match-driven changes.

mps-v1.1-experimental — seven-component model with three-result expectations, manager-spell boundaries, coverage-adjusted dismissal history and small-sample shrinkage. Earlier records use 0.35R + 0.20E + 0.15S + 0.10M + 0.15C + 0.05T + K, followed by clamping and rounding.