Where Line 51A Loses Time on Broadway

What the data says

Street friction by day. Missing service by night.

The named downtown crossings are real hotspots, although they are not the only slow blocks on the corridor. The evening result is sharper: buses that do appear move faster, but nearly one weekday evening trip in five is not observed at central Broadway.

+79% PM slowdown

20th Street to Grand

Rockridge-bound median time reaches 1.78 minutes in the PM peak, compared with a 1.00-minute low-traffic reference. Midday is 70% slower.

13–23 min middle half

Evening intervals spread out

The weekday evening IQR is 12.8–22.9 minutes Rockridge-bound and 14.1–21.1 minutes Fruitvale-bound. The p90 reaches 34 and 31 minutes.

80.7% observed

Scheduled trips disappear

Of 1,619 scheduled weekday evening trips at the two screenline stops, 1,307 have an arrival. Another 304 have no vehicle-position observation anywhere on the route.

Evening speed recovery does not translate into evening reliability.

Across the four focus segments, median times fall 14%–33% from the PM peak to evening. Yet the median arrival interval rises to 17.7 minutes, p90 rises above 30 minutes, and roughly 19% of observed intervals cross one or more missing scheduled slots.

Study design

A block-by-block view of Oakland Broadway

The corridor runs from the downtown turn at 8th/9th Street to College Avenue. Both directions are analyzed using their actual stop patterns; the patterns are not perfectly symmetric because stops fall on different blocks.

Rockridge-bound screenlineBroadway & 15th Street

Fruitvale-bound screenlineBroadway & 17th Street

Primary day typesWeekday, Saturday, Sunday/holiday

Primary time bandsAM, midday, PM, evening, late night

Schematic of the Rockridge-bound Broadway stop pattern from 8th Street to College Avenue, highlighting 11th to 15th and 20th to Grand. — Rockridge-bound schematic. The reverse direction uses its own GTFS stop pattern, including 17th–12th and Grand–19th as the closest opposing-direction counterparts.

Daily Line 51A trip and stop observation percentages for June and July 2026, with June 14 excluded. — June 14 is omitted because the previous static feed had expired and its replacement reached this tracker at 10 p.m. PT. That is a data-coverage failure, not a service result.

Daytime street performance

The arterial crossings slow sharply—but they are part of a wider pattern

On regular weekdays, Broadway’s slow blocks emerge early and remain visible through the PM peak. The two Rockridge-bound focus segments are substantially slower than their own 5–7 a.m. reference: 11th–15th is 66% slower at midday, while 20th–Grand is 79% slower in the PM peak.

Two heatmaps showing weekday median stop-to-stop times by Broadway segment and hour in both directions. — Median elapsed time for valid consecutive-stop observations. Redder cells are slower in absolute seconds. Focus segments are marked in red type.

Focus-segment results

Direction and segment	Low-traffic ref.	AM median	Midday median	PM median	Evening median	PM slowdown	PM p90
Rockridge · 11th → 15th	1.39 min	2.05	2.30	2.10	1.48	+51%	2.98
Rockridge · 20th → Grand	1.00 min	1.30	1.69	1.78	1.31	+79%	2.78
Fruitvale · 17th → 12th	1.48 min	2.13	2.58	2.79	1.88	+88%	4.20
Fruitvale · Grand → 19th	1.64 min	1.98	2.11	2.30	1.98	+40%	3.51

Low-traffic reference is the segment’s median on regular weekdays from 5–7 a.m. Each focus-period cell contains 485–1,209 observations. Elapsed time includes upstream dwell as well as movement, queuing, and signals.

Box plots of weekday elapsed times for the four focus segments across AM, midday, PM, and evening periods. — The distribution matters as much as the median. Fruitvale-bound 17th–12th reaches a 4.20-minute p90 in the PM peak, even though its median is 2.79 minutes.

Hourly median and 90th-percentile elapsed times for the four focus segments. — Solid lines are medians; dashed lines are p90. The shaded region begins at 7 p.m.

The focus blocks are not the only hotspots

Across all Broadway stop pairs, several northern blocks post larger percentage changes from their own early-morning references. That makes a corridorwide TSP screening more defensible than treating two intersections as the whole problem.

Ranked horizontal bars of weekday PM slowdown for every Broadway segment and direction. — Percentage change in median PM elapsed time from each segment’s own weekday 5–7 a.m. median. Red bars are the four focus segments.

Interpretation: the spatial pattern is consistent with intersection and cross-traffic friction, but these data do not include signal phases, traffic volumes, passenger loads, or door-open times. They cannot prove that a red signal caused a particular delay.

Evening service reliability

The street clears before the service does

From 7 p.m. to midnight on regular weekdays, the scheduled median is 15 minutes in both directions. Actual medians are 17.6–17.7 minutes. More importantly, the upper tail expands: p90 reaches 31–34 minutes and p99 approaches one hour.

Weekday Line 51A headway box plots for AM, midday, PM, and evening periods in both directions. — The plot is capped near 60 minutes so the central distributions remain visible. Weekday evening maxima are 99.7 minutes Fruitvale-bound and 98.9 minutes Rockridge-bound.

Weekday evening	Scheduled median	Actual IQR	Actual median	p90	p95	p99	>2× schedule	Mean rider wait
Rockridge-bound	15.0 min	12.8–22.9	17.7	33.9	40.3	59.5	12.2%	12.8 min
Fruitvale-bound	15.0 min	14.1–21.1	17.6	31.0	39.0	59.6	11.0%	12.6 min

Mean random-arrival wait is Σh² ÷ 2Σh. Evenly spaced 15-minute service implies 7.5 minutes; the observed distribution raises that to 12.6–12.8 minutes.

Curves showing the percentage of Line 51A intervals exceeding each duration at midday, PM peak, and evening. — Evening headways remain above a given long-gap threshold more frequently than midday or PM headways.

What is behind the gaps?

At the screenline, 19.9% of scheduled Rockridge-bound evening trips and 17.6% of Fruitvale-bound trips have no vehicle-position observation anywhere on Line 51A. Only 0.5% in each direction are seen elsewhere on the route but not at the screenline. The dominant signature is therefore whole-trip non-observation, not a bus routinely disappearing midway along Broadway.

Hourly stacked bars showing the weekday percentage of scheduled Line 51A trips observed at the screenline, observed elsewhere only, or not observed on the route. — The vertical scale begins at 75% to show hour-to-hour variation. “No route observation” may mean a trip did not operate, or that it operated without a usable GTFS-Realtime vehicle-position trace.

Weekday evening screenline	Scheduled trips	Observed at stop	Observed elsewhere only	No route observation	Stop observation
Rockridge · Broadway & 15th	831	662	4	165	79.7%
Fruitvale · Broadway & 17th	788	645	4	139	81.9%
Combined	1,619	1,307	8	304	80.7%

Calendar heatmap of the longest evening Line 51A headway on each June and July service day. — Long gaps recur rather than concentrating on a single incident. The two-month maximum is 151.8 minutes on Saturday, July 11.

Weekend extension

Weekend service is the sharper reliability warning

Saturday and Sunday/holiday street times are not uniformly worse than weekdays. Service delivery is. Only 52.6% of scheduled Saturday evening trips and 64.7% of Sunday/holiday evening trips are observed at the two central screenline stops.

Day type	Days	Route trips observed	Evening screenline observed	Median evening interval	p90 evening interval	Longest evening gap
Regular weekday	44	87.8%	80.7%	17.7 min	32.3 min	99.7 min
Saturday	7	75.4%	52.6%	24.2 min	68.8 min	151.8 min
Sunday / holiday	9	81.1%	64.7%	23.1 min	58.0 min	112.9 min

Grouped bars comparing median Broadway segment elapsed time by day type and time band in both directions. — Weekend roadway conditions do not explain the much larger weekend service-delivery gap.

Priority follow-up: validate a sample of the worst weekend dates against dispatch, operator, and AVL-health records. Vehicle-position data alone cannot distinguish a dropped run from a bus whose tracking equipment or trip assignment failed.

Joint diagnosis

Traffic and operations move independently

The chart below treats each service hour as two coordinates: how much slower Broadway segments are than their low-traffic reference, and how irregular the arrival intervals are. Evening hours move left as street delay falls, but often remain high on spacing variation.

Scatter plots comparing hourly segment slowdown with headway variability, with evening hours in purple and marker size showing missing scheduled trips. — Labels are hours; purple points are 7 p.m. or later. Larger markers indicate a greater share of scheduled trips without an arrival at the direction’s screenline.

The schedule change is not driving the street result

The focus-segment PM medians are similar across the two GTFS versions. Rockridge-bound 11th–15th is 2.15 minutes before the June 14 schedule shift and 2.08 afterward; 20th–Grand is 1.81 versus 1.77. The opposing-direction changes are mixed rather than uniformly worse.

Evening delivery did change

At the Rockridge-bound screenline, weekday evening observation falls from 83.7% in the June 1–13 feed era to 78.5% from June 15–July 31. Fruitvale-bound falls from 88.3% to 79.9%. This is descriptive, not evidence that the schedule change caused the decline.

Engineering and operations agenda

Treat the two problems with different tools

Signal studyCollect phase, split, queue, and traffic-volume data at 14th, Grand, MacArthur, and the northern hotspots.

TSP screeningModel conditional priority first where both median and p90 segment penalties remain high.

Run-delivery auditReconcile the 304 unobserved weekday evening trips with dispatch and operator records.

AVL healthSeparate trips that did not operate from vehicles that ran without a valid realtime trace.

Weekend controlReview pull-outs, reliefs, recovery time, and supervision on the dates with 60–150 minute gaps.

TSP can reduce a bus’s exposure to cross traffic. It cannot replace a bus that never enters the data—or guarantee that unevenly spaced buses leave a terminal evenly. Line 51A needs both a street-running program and a run-delivery program.

Methodology and limitations

How the analysis was constructed

Schedule denominator

Every scheduled 51A trip and stop was reconstructed from the GTFS version valid on that service date.
S1000251 covers June 1–13; S1000252 covers June 15–July 31.
June 14 is excluded because the tracker did not archive the newly effective feed until 10 p.m. PT.
GTFS calendar and calendar-date exceptions determine weekday, Saturday, and Sunday/holiday service.

Realtime observations

Vehicle positions are projected onto the scheduled shape; stop-crossing timestamps are linearly interpolated between adjacent probes.
Live finalization duplicates are removed by service date, trip, and stop sequence, preferring a row with an arrival and then the latest ingestion.
512,860 raw 51A rows become 356,115 deduplicated rows—a 30.6% reduction.
All observed trip IDs in the study match the date-correct static schedule.

Street-time measures

A segment requires valid arrivals at consecutive scheduled stops and positive elapsed time from 10 seconds through 30 minutes.
123,081 of 145,016 scheduled Broadway segment opportunities meet those requirements.
The low-traffic reference is each segment’s regular-weekday 5–7 a.m. median.
Confidence intervals in the underlying result tables use a service-day bootstrap so repeated trips on one date do not masquerade as independent days.

Headway measures

Headways are ordered actual arrivals within the same service date and direction; overnight non-service is never counted as a gap.
Screenline stops are Broadway & 15th Rockridge-bound and Broadway & 17th Fruitvale-bound.
A missing slot means one or more scheduled screenline trips fall between consecutive observed trips.
Random-arrival wait weights long headways in proportion to the time riders are exposed to them.

What this analysis cannot establish

Stop-to-stop elapsed time includes upstream dwell. Without door-open or passenger-load data, it is not pure roadway running time.
No signal-phase or traffic-volume data are available, so intersection causation and TSP benefits require a separate engineering study.
“No route observation” is a probable service-delivery failure, not a confirmed cancellation. A missing or incorrect GTFS-Realtime trip assignment can create the same signature.
The results are trip- and stop-weighted, not passenger-weighted.
These two months describe June and July 2026; they do not establish a permanent seasonal pattern.

Sources

AC Transit public GTFS schedules and GTFS-Realtime vehicle positions archived by the AC Transit performance tracker.
Archived schedule feeds S1000251 and S1000252.
Deduplicated observations from transit-203605.actransit.trip_observations.
Reproducible query, calculations, result tables, and plotting code in this project’s analysis/line_51a_2026_08_19 directory.