Reliability benchmark
E currently anchors the reliable end of the portfolio with a 60s P95 tail.
Research review
Copenhagen mobility resilience, turned into a decision-facing review.
This page packages the repository's committed structure, robustness, and reliability outputs into one narrative: network shape, failure sensitivity, line-level delay exposure, and early transfer-risk decisions. It is meant to sit beside the atlas and benchmark as the evidence layer for people who need both presentation quality and methodological traceability.
Intervention priority
Re 4516 should be the first candidate for targeted intervention, with a 780s P95 tail.
Coverage caveat
The committed 24h Task A summary shows 7.3% coverage, so operational completeness should still be read alongside the seven-day conclusions.
Executive View
High-level performance and exposure signals for management review.
Static network size
36,871
50,141 directed edges in the GTFS graph
Lines tracked
44
10 in the leadership band
Worst line tail
780s
Re 4516 is the current upper-tail outlier
Median P95
120s
Portfolio midpoint across ranked lines
Watchlist exposure
20
Lines above a 240s P95 or approaching it
Coverage caveat
7.3%
24h Task A snapshot, use as operational evidence not service truth
Targeted attack penalty
42.1x
Random vs targeted LCC ratio at 15% removal
Leadership story
E leads the portfolio with a 60s P95. The company-facing message is that the network already has visibly dependable services worth using as operational benchmarks.
Immediate risk
Re 4516 reaches a 780s P95 tail. That is where delay communication, schedule padding, and intervention planning should start.
Structural fragility
At 15% node removal, targeted failures shrink the largest connected component from 0.600 to 0.014. The network is bridge-sensitive, not evenly resilient.
Interactive Line Portfolio
Filter the reliability ranking, focus a single line, and compare its tail delay against the broader portfolio.
Portfolio context
The ranking chart responds to the active filter. Click a row to pin a line.
| Line | Band | Rank | P50 | P90 | P95 | Average | Samples |
|---|
Geographic Exposure
The map is a schematic coordinate view built from GTFS stop locations. It highlights where connectivity concentration and vulnerability overlap.
Static network map
No external basemap is used. This view intentionally stays offline and reproducible inside the repo.
Selected node
Critical bridges
Top hubs
| stop | type | metric | value |
|---|---|---|---|
| Skive Trafikterminal | hub | Degree | 42 |
| Odder Busterminal | hub | Degree | 35 |
| Randers Busterminal | hub | Degree | 35 |
| Hadsten St./Østergade | hub | Degree | 27 |
| Middelfart Station (Middelfart Kommune) | hub | Degree | 22 |
| Haderslev Busstation | hub | Degree | 22 |
| Hobro Busterminal (H.I. Biesgade) | hub | Degree | 20 |
| Rosen Allé (Ribe) | hub | Degree | 17 |
Decision Support Layer
Static fragility, observed delay patterns, risk-model confidence, and route trade-offs are shown together so the research can support operational decisions.
Failure sensitivity
Largest connected component retention under random and targeted removals.
Observed delay tail by hour
P95 delay profile in Copenhagen local hour.
Mode-level risk model
Current model remains mode-level with explicit fallback logic when sample sizes thin out.
| line | mode | p95 | CI low | CI high | evidence | source |
|---|---|---|---|---|---|---|
| C | ST | 120s | 0s | 0s | low | global |
| F | ST | 120s | 0s | 0s | low | global |
| H | ST | 120s | 0s | 0s | low | global |
Router trade-off table
Sample candidate paths show the travel time versus robustness trade-off.
| od | path | travel time | transfers | miss prob | CVaR95 | confidence |
|---|---|---|---|---|---|---|
| CPH_H_CPH_AIRPORT | path_a | 24.0 min | 1 | 0.083 | 4.0 min | medium |
| CPH_H_CPH_AIRPORT | path_b | 28.0 min | 0 | 0.083 | 4.0 min | medium |
| NORREPORT_LYNGBY | path_a | 22.0 min | 1 | 0.083 | 4.0 min | medium |
| NORREPORT_LYNGBY | path_b | 26.0 min | 0 | 0.083 | 4.0 min | medium |
Week 1 connectivity profile
Static connectivity footprint of the national GTFS graph used as the baseline.
Week 2 targeted vs random
Published robustness curve comparing random failures to betweenness-targeted attacks.
Week 3 line reliability
Committed seven-day line ranking by P95 delay.
Week 3 daypart pattern
Delay tail by Copenhagen local hour from the committed BQ-derived analysis.
Provenance
The dashboard is generated from committed markdown, CSV, JSON, and PNG artifacts so it remains auditable and safe to open locally.
week1_summary
../../docs/week1_summary.mdrobustness_summary
../../results/robustness/summary.mdrobustness_nodes
../../results/robustness/top10_vulnerable_nodes.csvweek3_summary_json
../../data/analysis/reports/week3/dt=2026-03-02/summary.jsonweek3_reliability
../../data/analysis/week3_line_reliability_rank.csvhour_quantiles
../../data/analysis/week3_hour_dow_quantiles.csvrouter_table
../../data/analysis/router_pareto_table.csvrisk_model
../../data/analysis/risk_model_mode_level.csv