Where every figure on a city page comes from, what we refuse to print, and what each number means at the glass.
A figure appears on a city page only if we can name the document it came from. For the water table that is the operator's own publication: its annual water quality report (the Consumer Confidence Report every US community water system must publish each year), a lab sheet, or a water-quality page the utility itself runs. The source and treatment entries can also come from the federal record for that water system in EPA's Safe Drinking Water Information System (SDWIS).
Water-softener dealers, filter sellers, aggregator sites and search snippets are not sources, however often their numbers are repeated. Where an operator publishes no figure, the page says so and prints none: no estimate stands in for a measurement.
The build enforces this. A captured field without a value and a named source fails the build, and so does city prose that quotes a figure its own water table contradicts.
Each table covers seven items across eleven fields: total hardness; pH, alkalinity and a corrosion index; iron and manganese; sodium, chloride and total dissolved solids; the disinfectant; the source and treatment; and whether results are reported for the whole system or per plant, zone or source.
Every field is either captured with the number the document prints, or listed under “Not in the report” after the document has been read. That line is a finished answer, not a gap we forgot to fill.
Where a report prints a range, the table shows the range; where it prints an average, the table shows that. Units are converted only where the conversion is unambiguous: micrograms to milligrams, or grains per gallon to milligrams per litre at 17.12.
Some reports cannot be read at all: the site refuses automated access, or the figures sit behind an address lookup. Those cities are recorded as closed, with the reason and every attempt listed, and they reopen when a readable copy turns up.
Calcium and magnesium, reported as milligrams per litre of calcium carbonate (mg/L as CaCO₃). One grain per gallon is 17.12 mg/L. Hardness is what a drying drop leaves as a white ring; the bands are in the table above.
| BAND | MG/L | GRAINS/GAL | AT THE GLASS |
|---|---|---|---|
| Soft | 0–60 | 0–3.5 | Wash and rinse with tap water. |
| Moderately Hard | 61–120 | 3.6–7 | Tap is fine except on dark or hot glass. |
| Hard | 121–180 | 7.1–10.5 | Wash with tap, rinse with distilled. |
| Very Hard | 181–250 | 10.6–14.6 | Distilled rinse mandatory; deposits accumulate. |
| Extremely Hard | 251+ | 14.7+ | Tap water cannot touch the glass. DI or pure-water system. |
The first three bands and the 180 mg/L line follow the USGS classification; this site splits the very hard end at 250 mg/L, because above it tap water cannot be left to dry on glass at all. A range that crosses a boundary is described at both ends rather than averaged into one band.
The same bands label every city in the city water rankings, which order the published hardness and dissolved-solids figures and list the cities that use chloramine.
In each city's table a short note sits under some numbers, such as “Below the staining level.” It links here. These thresholds are this site's working rules for glass, not health limits.
pH 7 is neutral. EPA's secondary (cosmetic) guideline for drinking water is 6.5 to 8.5, and utilities often run toward the top of that range, or above it, to protect pipes.
PH 9 OR ABOVE · Strongly alkaline. Strongly alkaline finished water: do not let it dwell on anodised aluminium.
The water's buffering capacity, how strongly it resists a change in pH, reported as mg/L of calcium carbonate. With pH and hardness it decides whether water tends to leave scale or to dissolve it.
Utilities report the Langelier Saturation Index or a similar measure (the Aggressive Index, calcium carbonate precipitation potential, or the Ryznar index); the table names which. On the Langelier scale a positive value means the water tends to deposit calcium carbonate scale and a negative value means it tends to dissolve it; the other indexes use their own scales.
EPA's secondary (cosmetic) standard for iron is 0.3 mg/L, the level at which iron begins to stain.
0.3 MG/L OR MORE · At the staining level. At or above the 0.3 mg/L staining threshold: rust-coloured film that acid cleaners do not shift.
BELOW 0.3 MG/L · Below the staining level. Below the 0.3 mg/L staining threshold.
EPA's secondary (cosmetic) standard for manganese is 0.05 mg/L.
0.05 MG/L OR MORE · At the staining level. At or above 0.05 mg/L: brown-black oxidative staining is possible on sills and frames.
BELOW 0.05 MG/L · Below the staining level. Below the 0.05 mg/L staining threshold.
Sodium stays behind when water dries, as part of the film on the glass.
60 MG/L OR MORE · Salt-film level. Enough to leave a salt film that pulls moisture back from the air.
Chloride residue is hygroscopic: it holds moisture on a surface that looks dry. EPA's secondary standard, 250 mg/L, is set for taste; this site flags a lower level because of what it does to frames.
100 MG/L OR MORE · Elevated for frames. Elevated: hygroscopic residue and faster corrosion on aluminium frames and screens.
Everything that stays behind when the water evaporates, so it predicts spotting better than hardness alone. EPA's secondary standard is 500 mg/L.
300 MG/L OR MORE · Spots on dry-down. Spots on dry-down; a pure-water final rinse earns its keep.
100 MG/L OR LESS · Often dries clean. Low enough that tap-rinsed glass often dries clean.
Utilities disinfect with free chlorine or with chloramine, chlorine combined with ammonia. Free chlorine fades from water left standing; chloramine lasts.
CHLORAMINE · Lasts in the bucket. Chloramine will not gas off in a bucket, and it wears pure-water resin faster.
From the utility's own report or, for the treatment steps, from EPA's federal record for that water system.
SOFTENED AT THE PLANT · Softened at the plant. Softened at the plant: treatment, not geology, sets this hardness.
From the US Census Bureau's American Community Survey, 2020–2024 five-year estimates: table B25034 (year structure built) and B25035 (median year built), read from the Census summary files. Each city maps to its Census place, or to its county subdivision where the city is a town or township.
Where a page covers two places (Wasilla-Palmer, Kailua-Kaneohe, Kahului-Wailuku) the counts are summed and no median is shown. A median the Census reports only as 1939 or earlier is printed that way.
The notes beside each band (wood sash, aluminium sliders, low-E coatings) describe what those decades usually installed. They are general guidance, not a survey of that city’s windows.
From NOAA's 1991–2020 US Climate Normals. Each city uses the nearest station with a complete year of normal highs, lows, rain days and freezing nights, preferring a first-order station within 15 km. The station and its distance are printed under the chart.
A frost month has three or more nights below 32°F on NOAA’s count. The best months are the frost-free months whose rain days are at or below the median for that city’s frost-free months. A rain day has 0.10 inch of precipitation or more, and a hot month has a normal high of 95°F or above.
From the US EPA Air Quality System annual summary for 2025. Each city uses the nearest monitor with a complete year of data within 25 km of its Census point, one monitor per site; where a region's monitors are not in that year's file yet (the Bay Area's 2025 data among them), the previous complete year is used and the page prints its year. Coarse particles (PM10) are mostly soil and road dust; fine particles (PM2.5) are mostly smoke, exhaust and industrial haze.
The figure is the annual mean in micrograms per cubic metre, with smoke and dust-storm days left in, because that is the air the glass actually sat in. The flag beside it compares that mean with every complete US monitor of the same kind that year: the median across 507 PM10 sites was 18 and across 907 PM2.5 sites 7.3.
At the glass, more PM10 means panes dull faster between washes, worst on the side facing the prevailing wind; more PM2.5 leaves a greasier film that plain water smears, so it wants a surfactant before the rinse.
A monitor measures the air where it stands, so a site 20 km away or beside a highway describes its own neighbourhood better than yours; the distance is printed for that reason. Where no complete monitor sits within range, the page says how far the nearest one is and prints no figure.
The distance from the city's Census point to the nearest salt water in the Census Bureau's TIGER/Line 2024 files: the coastline, plus water areas classed as bay, estuary, gulf or sound, or ocean, so a city on a bay reads as close to the sea as it really is. The coastline alone closes bays off; it would have put Warwick 26.5 km from salt water instead of half a kilometre.
Airborne salt falls off steeply in the first kilometre or two from the shore, so the distance matters more than the name of the water. Salt film is hygroscopic: it pulls moisture back out of the air, which is why coastal glass looks wet and smeared again after a humid night.
The Great Lakes are fresh water. Where one lies within 10 km, the page names it with its distance: expect spray and slow-drying damp on the facing walls, but no salt. A distance in the hundreds of kilometres simply means salt plays no part.
The Great Salt Lake is the reverse: an inland lake several times saltier than the sea, which the coastline measure above cannot see. For the three Utah cities within 25 km of its shore as the Census Bureau's 2024 water layer maps it (Salt Lake City, West Valley City and West Jordan), the page adds its distance beside the ocean figure. The shoreline moves with the lake's level, so treat the distance as the mapped edge, not today's water line.
From the EPA Toxics Release Inventory for 2024: plants in the industries the inventory covers report the pounds of listed chemicals they release to the air each year, as fugitive emissions and from stacks. A city counts every plant inside its Census boundary, or within 2 km of it, that reported any air release, and names the largest by pounds.
The inventory covers listed toxic chemicals, not dust or soot by weight, so it maps industrial activity near the glass rather than measuring what lands on it; the particle figures above measure the air itself. Glass downwind of a large reporting plant tends to carry an oily or gritty film that needs a surfactant or degreaser before the rinse.
From NOAA's Hazard Mapping System, where analysts outline the smoke they can see on satellite images each day and grade it light, medium or heavy. A city counts a day when a medium or heavy outline covers its Census point. The page gives the average for 2023–2025 and the worst of those years, because one fire season can dominate: the Canadian smoke of 2023 doubled the count across much of the Midwest and Northeast.
Satellites see smoke anywhere in the air column, and much of it passes overhead without reaching the ground, so this counts smoke days, not days of dirty air at street level; the fine-particle figure above measures what was breathed. When smoke does come down, often with rain, it leaves a greasy grey film that plain water smears: wash it with a surfactant, and wait until the smoke has cleared before booking.
From the USDA's 2022 Census of Agriculture: acres of cropland in the county that contains the city's Census point, as a share of that county's land (Census Gazetteer). Where the Census withholds a county's figure to protect individual farms, or publishes none for it (Alaska's boroughs, Virginia's independent cities, Baltimore and St. Louis), the page says so and prints no share.
Worked fields shed fine dust at planting and harvest, and spray drift travels with the wind, so a city ringed by farmland needs its glass done after harvest rather than before it. A county is a coarse unit: a city can sit in a county of farms while its own streets are miles from the nearest field, so read the figure as the farming around it, not next door.
From the USA National Phenology Network's Extended Spring Index: the 1991–2020 average day on which lilacs and honeysuckles first bloom at the city's Census point, modelled from temperature. It is the standard marker of spring's arrival. The heavy tree-pollen wave (oak, pine, birch) builds around that date, and early trees such as maple, elm and juniper start weeks before it, so wash glass after the wave rather than into it.
In frost-free places (South Florida, the Gulf coast, Southern California) the date comes very early and means less: something is flowering most of the year. The model covers the lower 48 states only, so Alaska and Hawaii print no date.
Rules that touch window cleaning, each read from the city's or water utility's own page and put in plain English: hose and water-use rules (runoff, shut-off nozzles, washing hard surfaces, drought stages), rules on wash water reaching storm drains, state rules for high-rise and rope-access work, and approval needed to replace glass in historic districts. Every item names its source and the month it was checked. State law (working at height) is printed once, on the state page, and each city page links to it.
Water rules change with drought stages, so treat the checked date as part of the rule and confirm with the utility before a big job. Where a source was checked and has no such rule, the page says so rather than leaving a gap.
From the Census Bureau's American Community Survey (2020–2024 5-year estimates) for the city's own Census place: median household income, median value of owner-occupied homes, the share of occupied homes lived in by their owners, the share of all housing that is detached houses, and the share in buildings of 20 or more units. Where a city is two Census places added together (Wasilla-Palmer, Kailua-Kaneohe, Kahului-Wailuku), the shares are summed but no median is printed, because medians do not add.
The firm count is from County Business Patterns 2023: establishments in NAICS 561720, janitorial services, the industry window cleaning is filed under, in the county around the city's Census point, with the rate per 10,000 county residents. It counts every cleaning business with a payroll, most of which clean inside buildings, so read it as how crowded the local cleaning trade is, not as a count of window cleaners. One-person outfits with no employees are not in it.
Cleaner pay is the median hourly wage of janitors and cleaners (occupation 37-2011) in the city's metropolitan area, from the Bureau of Labor Statistics' Occupational Employment and Wage Statistics for May 2025. It is employee pay, before overheads, so a firm's hourly rate runs well above it; read it as what local labour costs relative to other cities. Window cleaners are not published as an occupation of their own. Covered cities outside every metropolitan area show no figure: the survey's rural areas group several counties at once and change shape between releases, so we do not stretch one onto a single town.
What the figures mean for the work: detached houses are route work priced by the house and the season, while buildings of 20 or more units mean commercial contracts, lifts and rope access. Owner-occupied streets rebook year after year; renter-heavy ones book at move-out. Income and home value set what a household will spend on glass, not what the glass needs.
Each city's opening line, its explanation of the water, its paragraph on what sets the cleaning clock and its paragraph on doing the job yourself are written for that city from its own record.
Build checks refuse a number in an opening line that does not appear in the city's record, and prose that contradicts the water table. Where older prose and a newer report disagree, the report wins and the prose is rewritten.
Under each city's water table sits a short panel of method calls. Nothing in it is new data: each call is worked out from figures already printed on that page (the utility's own hardness, dissolved solids, iron, manganese, sodium, chloride and disinfectant, and NOAA's normals), using the thresholds this page sets out above. Each cell names the figure it rests on, and where a utility publishes nothing usable the panel says so rather than guessing.
Final rinse. Total dissolved solids decide it when the utility publishes them, because everything dissolved stays behind as the water dries: 300 mg/L or more calls for a pure-water final rinse, 100 or less usually dries clean from the tap, and anything between works if the rinse is squeegeed off before it dries. With no dissolved-solids figure the call falls back on hardness: soft water is a tap rinse, moderately hard water is squeegeed off wet, and hard water or harder gets a pure-water rinse. With neither figure, the page suggests letting one rinsed pane dry in the sun; a white ring settles it.
Residue. Iron at 0.3 mg/L or manganese at 0.05 mg/L, EPA's cosmetic staining levels, outranks everything else: that film is metal staining, and the acid that shifts a white mineral ring does not touch it. Next come sodium at 60 mg/L or chloride at 100 mg/L, which leave a salt film that draws moisture back from the air. Otherwise, hard water or harder leaves calcium and magnesium scale, which a mild acid lifts while it is fresh. 'Nothing flagged' is printed only when the utility published those figures and every one sits below its flag.
Pure-water gear. Deionising resin is used up by what is dissolved in the water it treats, so the panel compares the city's dissolved solids with water at 100 mg/L: at 400 mg/L a cartridge yields roughly a quarter as much pure water. Chloramine is flagged separately because, unlike free chlorine, it does not gas off from a bucket and it wears resin faster. This is a planning ratio, not a cartridge rating; manufacturers rate capacity in their own terms.
Hot months. Listed only where heat changes the job: months whose normal high is 90°F or more, in cities whose rinse call is pure water. A loaded rinse dries on sunlit glass before it can be squeegeed, so shaded sides go first and sunlit sides wait. The best months for exterior work are in the climate strip; this cell does not repeat them.
The month strip on each city page is also a cleaning calendar. It adds two things to the NOAA normals: the month of first spring bloom from the USA National Phenology Network's 1991–2020 Extended Spring Index, and the months when wildfire smoke keeps coming back, from NOAA's Hazard Mapping System. Hazard Mapping System analysts outline the smoke they see from satellite each day and grade it light, medium or heavy; a day counts for a city when a medium or heavy outline covers its point. That is smoke in the air column, not a ground reading, so the calendar uses it for timing only. A month is marked as a smoke month when it had two or more such days in at least two of the three years 2023 to 2025, so one extreme season (the Canadian smoke of June 2023) does not mark a month for good.
WASH marks up to three months, and every one of them is already one of the city's best months (frost-free and no wetter than the median of its frost-free months). The first comes after the pollen: the first best month after the bloom month. The second comes after the smoke: the first best month after the last month of the city's longest run of smoke months. The third closes the year: the last best month before the city's frost season starts, skipping smoke months, because that wash has to last the winter. Where two rules land on the same month it is marked once. Cities with no bloom date, no recurring smoke and no frost season get no picks, and the best months stand on their own. These are timing rules, not a service interval: how often glass needs cleaning still depends on the address.
Every water table names its document and edition, so a figure’s age is visible. US utilities publish a new report every year, due by July 1; when a newer edition is read, its figures replace the older ones.
Figures we have had to correct, and what they read before, are listed in the corrections log.
If a figure is wrong, or a newer report is out, tell us through the contact page and name the document. The record changes first, and the build checks run again before anything ships.
The wider rules for articles, bylines and corrections are in our editorial standards.