Part 5 of 6 in The Good Side of 311
Most of What the Streets Department Does, It Does Well.
Volume-weighted verdict distribution across all 19 named Streets service codes. Each segment is both a percentage and a ticket count.
I wanted to know what a normal Philadelphian would think of the Philadelphia Streets Department, not what the city publishes about timing. The Streets Department handles the largest workload of any 311 agency in the city — 293,532 closed tickets, roughly 60% of everything 311 routes. The question is whether a reasonable person, looking at the work, would consider the service acceptable.
So I gave an LLM the median and p90 close time for each of the 19 named Streets service codes, along with what each code represents, and asked it: based on the kind of work the code describes, would a normal Philadelphian consider this service acceptable? The LLM judged on what a reasonable person would expect — not on whether the city had a published commitment it hit or missed. Four buckets: ahead of expectation, meets expectation, misses expectation, disappoints.
The headline: on the citizen's yardstick, 76.6% of Streets volume meets what a normal Philadelphian would consider acceptable. 19.6% disappoints.
What the 76.6% actually looks like
The bulk of the agency's work — rubbish collection, illegal dumping, street lighting, salting, paving, complaint response — runs at a pace a normal person would consider acceptable. Most of what Streets does, it does well, at scale.
Rubbish collection is the highest-volume operation, at 84,131 tickets over the analysis window. The median close is 2 business days. The LLM classified this as “meets expectation” — on the largest single workload in the city's 311 system, the agency delivers at the pace a reasonable person would consider acceptable.
Illegal dumping is the second-largest code, at 51,158 tickets. The median close is 3 business days. The LLM classified this as “meets expectation” — on its second-largest workload, the agency is faster than a reasonable person would expect.
Street light outage runs 19,233 tickets with a 9-business-day median. Salting runs 17,921 tickets with a 2-business-day median. Both codes landed in “meets expectation” on volume big enough that the rank isn't a small-sample artifact. Street paving runs 1,860 tickets with a 7-business-day median. These are operations the city has chosen to staff, and the staffing holds.
The rest of the 76.6% — alley light outage, dead animal pickup, manhole covers, line striping, dangerous sidewalks, sanitation violations, the catch-all “Other (Streets)” code — adds another 31,877 tickets at what the LLM classified as acceptable timing. These are the day-to-day operations Philadelphians don't lose sleep over. They are doing what a reasonable person would consider reasonable.
Where the agency runs ahead of expectation
One code lands in the “ahead of expectation” bucket. SR-ST08, “Complaint (Streets),” is the catch-all for generic street-related concerns that don't fit a more specific code — a citizen calling about a vague street problem with no clear defect. The published timing commitment on this code is 30 business days; the median close is 3.9 calendar days. The LLM classified this as “ahead of expectation” — the city is resolving complaints far faster than a reasonable person would assume based on the kind of work involved. That is the kind of result the city ought to be advertising.
The 19.6% the agency needs to fix
The bulk is fine. The slice that disappoints is small in volume but concentrated on codes where citizen tolerance is zero. Four codes land here, and they are not random — they are the safety-and-sanitation fundamentals where a reasonable person's expectation is fast and the city's actual delivery is not.
Street defect (potholes) is the biggest single gap in the operation. On 34,660 tickets, the median close is 23 calendar days. The LLM classified this as “disappoints” — a reasonable person reporting a pothole would expect a much faster response, and three weeks is not it.
Traffic signal emergency is the safety concern. On 11,710 tickets, the median close is 2 business days. The LLM classified this as “disappoints” — a citizen reporting a malfunctioning signal at a busy intersection is reporting something where someone could be killed, and a reasonable person would expect a same-day response.
Snow shoveling is the structural problem. On 6,365 tickets, the median close is 4 business days. The LLM classified this as “disappoints” — a reasonable person reporting uncleared sidewalks after a storm would expect a much faster response, and four times that expectation is not it.
Abandoned bike is the absurdity. On 157 tickets — the lowest-volume code in the analysis — the median close is 388 business days, or 553 calendar days — a year and a half. The LLM classified this as “disappoints” — the city's timing on abandoned bikes is so slow that even a very patient person would consider it unreasonable.
These four codes together are 19.6% of named-code Streets volume. They are not most of the agency's work. They are the codes where the agency's failure mode is loudest and where a citizen forms their opinion of the agency. The agency should pay attention to this slice — not because it is large, but because the codes are safety-critical and citizen tolerance is zero.
Why I asked an LLM and not a fixed cutoff. A reader looking at the data could pick any number of thresholds — “the median should be at most X days,” or “the p90 should be at most 2x the median.” Those thresholds are arbitrary. They encode one opinion about what counts as “acceptable” and apply it to every code regardless of what the code is. I wanted a verdict that accounted for the kind of work the code represents — a pothole response is different from a paving response, and a reasonable person's expectation for each is not the same. The LLM gives me that, with the usual caveat that the verdict is one reasonable read rather than a ground truth.
The bottom line
Most of what the Streets Department does, it does well, and on a volume big enough that the headline isn't a hedge. 76.6% of Streets 311 volume meets what a normal Philadelphian would consider acceptable. That is 206,180 of the 269,340 named-code tickets in the analysis window — more than three of every four — where a reasonable person is satisfied with the work the agency delivered.
The 19.6% of tickets where the city underperforms vs a reasonable person (or reasonable AI's) expectations are not insignificant. And they tend to be concentrated in areas we see (broken traffic signals, potholes) or viscerally feel as inconveniences (e.g., unploughed snow right after a snowstorm) - but dwelling on them is unfair to the city and the Streets department.
Three out of four times the Streets department solves the problems it encounters in a reasonable period of time - let's be honest, that's a pretty high bar most of us probably don't consistently meet in our professional or personal lives.
Notes, Sources, and Methodology
All data is the OpenDataPhilly 311 Service Requests table public_cases_fc. The Streets Department's closed-ticket universe is 293,532 tickets filed 2024-01-01 through mid-2026. The 19 codes covered here are the codes with at least 100 closed tickets; together they account for 269,340 tickets, which is the denominator behind every percentage in this article. Information Request rows are excluded, and the 24,190 null-service_code Streets rows are audited separately and never counted toward a verdict.
Methodology. For each code I extracted the median and p90 actual close times in calendar days, what the code represents, and (for context) the city's own published timing commitment and met-rate under its unit arithmetic. I sent that profile — code name, what it covers, median close, p90 close, and the published commitment — to an LLM via an Anthropic-compatible API with a classification prompt asking for one of four verdicts on whether a typical Philadelphian would consider the service satisfactory: ahead_of_expectation, meets_expectation, misses_expectation, or disappoints. The volume-weighted verdict share is computed by summing ticket counts per verdict category and dividing by the cohort total.
Limitations. The LLM verdict is one reasonable read of what a typical Philadelphian would think — not a ground-truth survey. Different prompt framings, or a different model, could shift individual codes between the “meets” and “misses” buckets, or between “misses” and “disappoints.” The headline 76.6% figure is robust to those kinds of boundary moves; the precise share in each bucket is not. The volume weighting means that SR-ST03 Rubbish Collection (84,131 tickets) and SR-ST02 Illegal Dumping (51,158 tickets) dominate the headline by themselves; if either of those verdicts flipped, the headline would move meaningfully.