Methodology v1.2 · The Will Johnson Team
How the Middle Tennessee Town Index is measured
Eighteen Middle Tennessee municipalities, measured on how much of something each one has — restaurants per resident, park acres per resident, permits per thousand existing homes. Each measurement carries its unit, its source, the date we pulled it and the written test that decided what counted. Beside it sits a 1–10 band, which is a bucket on a ruler published in advance and identical for every town. Nobody assigns a band; it is arithmetic. There is no total, no average and no overall score, and there never will be. The index measures what a place has, never what a place is.
A 10 means the town has at least a published amount of something — a threshold set in advance and the same for every town. It does not mean best. Several of these run opposite to each other: the town with the most restaurants per resident will almost never be the town with the biggest lots. Read the number, not the band.
How to read a band
A band is a bucket on a ruler, not a placing. Each category publishes nine edges in its own unit, set in advance and identical for every town, and the band is arithmetic: count how many edges the measurement clears. Nobody assigns it. Two people working from the same measurement land on the identical integer, not within a point of each other.
Because the edges are absolute, a 10 is a statement about the place — “at least this much of this thing” — and not a statement about the other seventeen towns. Ties are honest. A blank in one town moves no other town. Re-measuring one town changes exactly one number.
The measurement is printed larger than the band on purpose. Read the number.
The five design rules
- 1
Measure availability, never desirability.
Every category answers “how much of this does the town have?” — a countable, sourced quantity. A 10 never means best.
- 2
No composite. Ever.
No total, no average, no weighted index, no leaderboard, no overall score, no default sort but A–Z. This is a permanent design decision, not a current preference, and the reason is published below.
- 3
The excluded categories are published.
Schools, crime and safety, community feel, demographics of any kind, religion, who lives here, prestige, exclusivity — none of these exists in this system, and the cards below say why, in public, with somewhere useful to go instead.
- 4
Every measurement shows its work.
The inputs, the number, the unit, the source, the date, and the written test that decided what counted. A figure without provenance cannot be represented in our data model, let alone published.
- 5
Measurements describe places, never people.
No number here may be justified, explained, or captioned by who lives somewhere.
Why there is no overall score
A single number summarizing a town is the artifact that places one town above another, and a licensed agent publishing it is steering. That is the whole reason. There is no total field to render, no average, no weighted index and no leaderboard — not hidden, not disabled, not present in the data model at all, and a build guard fails the deploy on the vocabulary.
Eighteen measurements in eighteen different units are also not commensurable. Adding minutes to acres to permits would be meaningless arithmetic even if it were safe.
One ordering does survive, and it is governed rather than assumed: a category page shows one measurement, in one unit, and a reader may sort that single column. Refusing to let anyone sort one column of acres would be theatre, not compliance. Four rules hold it: A–Z is what the page loads in and what a crawler sees; no superlative appears in the page title, the heading, the description or the structured data — the page is named for the measurement, never for its extreme; one category is sortable at a time and never two together; and the two-ended label prints above the column, not below it.
The honest statement of what that leaves: those rules stop us publishing a ranking. They do not stop a reader from reading one. We accept that for single, sourced availability measurements in one unit, and nowhere else — and nobody on this team may repeat one of those orderings back to a client as a recommendation.
Every category prints both ends
Direction is exactly where desirability leaks in. A 10 on drive time means closest, and a meaningful share of the people moving here are moving for the distance. So every category carries a printed sentence naming what a 10 means, what a 1 means, and that neither end is better. It is required, not optional: the build fails on a category without one.
Where “neither end is better” would not be credible, we do not say it. Seventeen of the eighteen are genuine matters of preference — someone wants the big-box stores close and someone is moving here to be farther from them. Hospital bed counts are not one of them, and printing the boilerplate beside a bed count would teach a reader to discount the other seventeen labels. In that case the label names both ends and then says what the number is not a judgment about. Every category on this site prints that second sentence too.
What we do not measure, and why
Schools and crime lead, because those are the two a reader is actively wondering about, and each card says where to go instead. Pointing a reader somewhere useful is what separates a principled exclusion from a dodge.
Schools and school quality
School ratings are the most widely used proxy for familial status and race in American real estate, and an agent measuring districts is steering with an extra step. We will name the district or the assigned school for a specific address on request — that is a factual answer — and we measure nothing about quality. Our default here is silence, and it is a house rule that predates this index. No private rating company is linked or named anywhere in this index.
The official school data sources, by district and by address
Safety, reported-crime statistics, and police data
Reported-crime statistics correlate so strongly with race and income that measuring them functions as steering regardless of intent. They are also distorted by reporting and enforcement differences between jurisdictions, so the figures are not comparable across towns on their own terms. We do not measure them, summarize them, or characterize them, and we do not use the adjectives that usually travel with them about any place. The FBI publishes guidance asking users not to rank or compare places using its data; we follow it.
Any composite, total, average, weighted index, or overall score
Permanent, and structural rather than promised. See the section above: a single number summarizing a town is the sentence a steering complaint quotes back, and the eighteen measurements are in units that cannot be added anyway.
Community feel, vibe, atmosphere, family-friendliness
No source, no unit, no denominator — an opinion wearing a number. Pairing “friendly” with a household type is specifically a familial-status characterization. Anything real underneath a vibe claim is either one of the eighteen measured categories or an opinion, and we publish only the first.
Demographics of any kind
Race, ethnicity, national origin, age, income, household composition, disability status — and population growth or change, which is a demographic series wearing a growth label. These are protected classes or direct proxies for them. We use Census data heavily and pull only housing-stock, land-area, population-count, building-permit and journey-to-work tables; the demographic tables in those same datasets are off-limits in the pull scripts, not merely in judgment. Population appears in this index only as a denominator.
Religion, houses of worship, and religious character
A count of churches looks innocent and is not. A planned annual-event count was cut for the same reason, and has its own card below.
Prestige, exclusivity, desirability, status, or “nicest”
Pure ranking language with no measurable referent, and what a steering complaint quotes back. The site's compliance gate already blocks that adjective family next to a place noun, and “town” is one of the nouns it watches.
Median home price as a 1–10
The dollar figure itself is published once the live MLS feed is on, with its source and date, because a buyer needs it. It is never converted to a band: a band adds nothing over the dollar figure while adding an ordinal ordering of eighteen towns by wealth inside a grid whose grammar teaches that higher is better. The certified property-tax rate is handled the same way — published, never banded.
Future price movement, investment outlook, market timing
Banned by house rule and by the compliance gate; it would turn a descriptive index into unlicensed advice. The one category that touches construction is backward-looking only: permits already issued, housing already built.
Third-party livability rankings as inputs
Every one of them bakes school ratings, crime scoring and demographic composition into its method, so importing their numbers would import our excluded categories through the back door. This is also why our figures will sometimes disagree with theirs.
Proprietary walkability and livability indices
We cannot publish their method, so the number could not show its work — and republishing a licensed score is a licensing problem besides.
Nashville itself, and any neighbourhood-level measurement
Nashville-Davidson is a consolidated city-county of roughly 690,000 people across 525 square miles; normalizing it against towns of 10,000 to 180,000 produces arithmetic, not information. Breaking any city into sub-areas and measuring them against each other is the comparison Fair Housing treats most severely, and the town-set rule excludes it structurally. Goodlettsville is measured because it is a separately incorporated municipality; Madison, Hermitage, Donelson, Antioch, Bellevue and Old Hickory are not municipalities and are not measured.
Municipalities below 10,000 residents
By the published floor, and each is named with its figure in the town-set section rather than quietly dropped. An omission with a number beside it reads as arithmetic; an omission without one reads as a verdict.
Nightlife, “scene,” and anything describing a crowd
Durable code for age and household composition. An earlier draft tried to make this measurable as a count of venues with live-music calendars; no enumeration source exists, so it is not measured at all.
Industry mix, occupation categories, and wage data
One word away from the class-and-race proxies the compliance gate blocks by name. The jobs category publishes one ratio and the two counts behind it, and stops.
Annual event counts
Cut. In practice it was a proxy for the card above: a pilot count found a third of one town's events were Christmas programming. Two further defects sank it on measurability alone — the two pilot towns were counted off different classes of source, and whether an event actually recurs was never verified. What a town is known for locally still appears on a town's own page as named things with sources, never as a count.
Walkable-centre storefront counts
Cut. The rule measured a quarter-mile circle around a “centre” that fell back to town hall where no historic square exists, which put one pilot town's centre in a big-box corridor and the other's on a courthouse square — the number measured the fallback rule, not the towns. The only neutral rewrite needs a storefront inventory no free source provides, and its per-address inclusion test was the one measurement where two careful scorers genuinely disagreed, which breaks the rule that nobody assigns a band.
Pharmacy counts
Dropped for measurability rather than for steering: chain locators are blocked to scripts and the free map data missed half the stores in the pilot. Said here so the omission is explained rather than noticed.
The categories that publish today
Residential units authorized per 1,000 existing housing units
How much residential construction is underway in this town, relative to the housing that already exists?
- Unit
- units authorized per 1,000 existing housing units
- Band edges (edge table v1.1, basis B+O)
- 5 · 10 · 15 · 20 · 30 · 40 · 50 · 65 · 80
- What a 10 and a 1 mean
- 10 = the most construction underway. 1 = the most settled housing stock. Neither end is better than the other.
- What this number is not a judgment about: it counts permits already issued against housing already standing. It says nothing about what the homes here are worth, what they will be worth, what the existing housing is like, or who lives in it.
- What was counted
- Residential units authorized by building permit inside the municipality, summed across the four unit-size classes the Census Building Permits Survey publishes (1-unit, 2-unit, 3–4-unit, 5-or-more-unit) in its place-level annual file, for 2023, 2024 and 2025; averaged over the three years; divided by the place's housing units from the American Community Survey 5-year table B25001, latest released vintage; multiplied by 1,000. A place-year counts only where the survey's months-reported field is 12. Where any of the three years reports fewer than twelve months, the cell is not published and no figure is estimated in its place.
- Sources of record
- U.S. Census Bureau, Building Permits Survey — place-level annual files, South Region (so2312y, so2412y, so2512y) · U.S. Census Bureau, American Community Survey 5-year, table B25001 (2024 release, table-based summary file) · U.S. Census Bureau, American Community Survey 5-year, table B25034 (2024 release) — printed as context, never banded
- Where it could go wrong
- Growth language is where this category leaks. Everything published here is past tense: permits already issued, housing already built. No forward-looking statement of any kind attaches to this number, no expectation about prices, and the share of housing built 2010 or later is never paired with a description of the housing that already stands or of anyone living in it. The new-versus-old housing contrast has carried coded meaning for seventy years; the compliance gate blocks the euphemisms by name, and this category measures new supply and contrasts it with nothing. No builder is named, compared or ranked here.
- Notice about these edges
- Part of this ruler was set from the range observed across Middle Tennessee rather than from a physical anchor or a published external benchmark, and that observation base was partial when the edges were set. Under the methodology's edge-governance rule an edge can change only as a published version bump that re-bands every town at once — never a per-town edit, never within twelve months of the last change in this category, and never while a town's measurement here is pending. The methodology separately schedules a single, announced re-set of observed-range edges after the first complete eighteen-town pass, and names the categories that carry it; this one is not among them today. If that ever changes, it is announced here before any band moves.
12 of 18 towns publish a figure · measured 2026-09-02 · refreshed on the A-tier cadence · pulled by the index-automation role.
Jobs located in the town per 100 employed residents
How many jobs are located inside this town, relative to how many of its residents are employed?
- Unit
- jobs per 100 employed residents
- Band edges (edge table v1.1, basis P+O)
- 40 · 55 · 70 · 85 · 100 · 120 · 140 · 170 · 200
- What a 10 and a 1 mean
- 10 = the most jobs inside the town relative to its working residents — a town people drive into. 1 = the fewest — a town people drive out of. Neither end is better; it is which way the traffic goes in the morning.
- What this number is not a judgment about: it is a ratio of two counts and nothing else. It says nothing about what kind of work is done here, what it pays, or who does it.
- What was counted
- Jobs located in the place, from the Census LEHD LODES8 Workplace Area Characteristics file (worker segment S000, job type JT00 — all jobs, column C000), divided by employed residents of the place from the matching Residence Area Characteristics file (same segment and job type), multiplied by 100. Both sides are aggregated from 2020 census blocks to the municipality using the Tennessee LODES crosswalk field `stplc`, so the figure follows the municipal boundary rather than a ZIP or a county. Reference year 2023, the latest LODES8 release. Above 100 the town has more jobs than working residents; below 100 it has fewer.
- Sources of record
- U.S. Census Bureau, LEHD LODES8 — Tennessee Workplace Area Characteristics, S000 JT00, 2023 · U.S. Census Bureau, LEHD LODES8 — Tennessee Residence Area Characteristics, S000 JT00, 2023 · U.S. Census Bureau, LEHD LODES8 — Tennessee geography crosswalk (tn_xwalk)
- Where it could go wrong
- The sharpest trap in the index: employment talk slides one word into the class-and-race proxies the compliance gate blocks by name. This category publishes one ratio and the two counts behind it, and stops. No industry mix, no occupations, no wages, no employer-type breakdown is scored or published here. A named major employer may appear elsewhere on the site as a landmark fact; it is never an input to this or any band.
- Notice about these edges
- Part of this ruler was set from the range observed across Middle Tennessee rather than from a physical anchor or a published external benchmark, and that observation base was partial when the edges were set. Under the methodology's edge-governance rule an edge can change only as a published version bump that re-bands every town at once — never a per-town edit, never within twelve months of the last change in this category, and never while a town's measurement here is pending. The methodology separately schedules a single, announced re-set of observed-range edges after the first complete eighteen-town pass, and names the categories that carry it; this one is not among them today. If that ever changes, it is announced here before any band moves.
18 of 18 towns publish a figure · measured 2026-09-02 · refreshed on the A-tier cadence · pulled by the index-automation role.
Defined, not yet published
The other sixteen categories are written, edged and owned; they render as blanks until they are measured. Honesty about the gaps is cheaper than the gaps being discovered.
- Dining Density — Waiting on a point-of-interest dataset with a measured capture rate of at least 90% against Census ZIP Business Patterns. The free option captured 40% and 34% of the two pilot towns.
- Dining Variety — Same dataset dependency as Dining Density.
- Grocery Access — Store-locator pages are blocked to scripts, so the count is browser work by a person. Scheduled for the first January block.
- Retail Anchor Depth — Hand-verified browser work, and each town's count must clear the two-sided distance test in METHODOLOGY §3.4 before it can band on a free routing engine.
- Parkland Supply — Acreage lives in adopted parks master plans and county GIS, not on city web pages. Second researcher block.
- Trails and Greenways — Mileage must come from the municipality's own published figure and pass a written inclusion test. Second researcher block.
- Water Access — Needs the state agency's geocoded boat-access layer extracted. A floor is not a count.
- Golf Access — Needs broker ratification of the public-play inclusion test, and the routing key.
- Movie Screens — Operator pages are blocked to scripts. First January block.
- Arts, Museums and Theaters — Needs a full visitor-bureau directory sweep per town, verified at each venue's own site.
- Downtown Nashville Drive Time — Needs a traffic-model routing key. A free-flow engine is a specification deviation and is not banded.
- Rush-Hour Swing — Unmeasurable on any free engine — it is the difference between two departure times on a traffic model.
- BNA Airport Drive Time — Needs a traffic-model routing key.
- Healthcare Capacity — State licensure records are a per-facility lookup by hand. First January block.
- New-Construction Availability — Blocked on the live MLS feed. Every cell stays blank until then; an aggregator figure is never substituted.
- Typical Lot Size — Blocked on the live MLS feed.
Which towns are in, and which are out
The rule is mechanical, published, and applied without discretion. An incorporated Tennessee municipality is in the index if its Census population estimate is at least 10,000 and its town hall is within a 55-minute off-peak drive of one fixed downtown Nashville coordinate. Both thresholds are arbitrary-but-fixed; moving either is a version bump that re-admits or re-excludes towns in public. The population floor exists because nine of the eighteen categories divide by residents, and per-capita rates are unstable at small denominators.
| Town | County | Population at admission |
|---|---|---|
| Brentwood | Williamson | 45,525 |
| Clarksville | Montgomery | 185,690 |
| Columbia | Maury | 48,812 |
| Dickson | Dickson | 16,878 |
| Franklin | Williamson | 89,142 |
| Gallatin | Sumner | 52,489 |
| Goodlettsville | Davidson and Sumner | 17,433 |
| Hendersonville | Sumner | 63,947 |
| La Vergne | Rutherford | 41,430 |
| Lebanon | Wilson | 51,501 |
| Mount Juliet | Wilson | 44,066 |
| Murfreesboro | Rutherford | 168,387 |
| Nolensville | Williamson | 15,809 |
| Portland | Sumner | 13,651 |
| Smyrna | Rutherford | 60,302 |
| Spring Hill | Williamson and Maury | 59,398 |
| Springfield | Robertson | 19,842 |
| White House | Sumner and Robertson | 16,463 |
Census Population Estimates Program, vintage 2024, July 1 2024 estimate, read at the whole-municipality summary level. Population is used here as a denominator and as the published admission threshold, and nowhere else.
Named exclusions, with their figures
- Fairview — about 9,500 residents. Below the published 10,000 floor.
- Thompson's Station — 7,485 at the 2020 Census. Below the floor. Enters automatically the year the estimate crosses 10,000, without anyone deciding to add it.
- Belle Meade — below the 10,000 floor. Residential-only municipality under the floor.
- Forest Hills — below the 10,000 floor. Residential-only municipality under the floor.
- Oak Hill — below the 10,000 floor. Residential-only municipality under the floor.
- Nashville-Davidson — about 690,000 residents across 525 square miles. A consolidated city-county. Normalizing it against towns of 10,000 to 180,000 produces arithmetic, not information — and breaking a city into sub-areas and measuring them against each other is the comparison Fair Housing treats most severely. Its neighbourhoods are not municipalities and are not measured.
For Will, Mindy and John
Saying this out loud
One rule carries this whole card: say the number, not the placing. Everything else follows from it.
Not: a sentence that names a winner, in any wording — including “has the most.”
“Brentwood permitted about eight new units for every thousand it already has, averaged over the last three years. Gallatin’s figure is about sixty-nine. Which of those is the number you care about?”
Two figures with their unit, and a question. No superlative, no ordering, no verdict — and it ends with the client talking.
- ·Name the tradeoff in the same breath. A town in the middle of construction and a town whose housing is mostly already built are two different facts, not a better and a worse.
- ·Never add bands across categories, out loud, for anyone, including as a joke. The units are not commensurable; the sum means nothing and sounds like a verdict.
- ·Never repeat an ordering off a category page as a recommendation. The page lets a reader sort one column of one unit; that does not license us to read the result back to them.
- ·If the question is “which one is the best?” — answer “that depends on which of these you want most of, and here is how they trade off.” Never answer the question as asked.
- ·If the question is which town has the most of something — the question our own category pages invite — give two or three towns’ numbers, in order of nothing, and let the client decide which matters to them.
- ·Never volunteer, and never confirm, anything this index does not measure. Point to the methodology page, which names every exclusion and where to go instead.
- ·Never volunteer, and never confirm, anything about schools, crime, or who lives somewhere — not on a call, not in a text, not “off the record.” Point to the official sources on /school-and-crime-data and stop there.
Version and changelog
v1.2 · 2026-09-02
Rounding fixed to happen once, at recording, so the printed figure is the figure that produced the band. A cell now stamps the version of the edge table that banded it rather than the version of this document, so correcting the methodology does not re-band anything. Labels may no longer claim “neither end is better” where that would not be credible. The one surviving ordering — a sortable single column on a category page — is named and governed by four rules. Categories resting on an observed range must disclose, from their first published day, that their edges get one re-set after all eighteen towns are measured. Every Census input must use the latest released vintage. Launch shape changed to category pages first. No band edge moved, so no band moved.
v1.1 · 2026-09-02
Rank bands replaced by absolute value bands: a 10 became a statement about a place rather than a placing among eighteen towns, ties became honest, and a blank in one town stopped moving every other town. The two-ended label became required on every category. Three categories were cut (annual events, live music, walkable centre) and one was narrowed. Machine-readable surfaces stopped emitting 1–10 integers entirely.
v1.0 · 2026-08-30
Initial design: twenty-one categories, eighteen towns, rank-band scoring, tiered re-measurement cadence, renderer-enforced expiry, published exclusions, risk register.
Questions
- How is the 1–10 band calculated?
- Each category publishes nine edges in its own unit, set in advance and identical for every town. For a most-first category the band is one plus the number of edges the measurement reaches or clears; for a shortest-first category it is ten minus the number of edges the measurement exceeds. A measurement sitting exactly on an edge belongs to the higher band. The measurement is rounded once, when it is recorded, and the rounded figure is both what prints and what is compared against the edges — so anyone can recompute our band from the number on the page and the edges on this one, and must get our answer.
- Why is there no overall score for a town?
- Because a single number summarizing a town is the artifact that places one town above another, and a licensed real estate agent publishing that is steering under the Fair Housing Act. It is a permanent design decision: there is no total field in the data model, no average, no weighted index and no leaderboard, and a build guard fails the deploy on the vocabulary. The eighteen measurements are also in different units and cannot meaningfully be added.
- Why don't you rate schools?
- School ratings are the most widely used proxy for familial status and race in American real estate, so an agent scoring districts is steering with an extra step. We will name the district or the assigned school for a specific address, which is a factual answer, and we measure nothing about quality. The official sources are linked at /school-and-crime-data: the Tennessee Department of Education report card, the federal school directory, and each district's own zone lookup. No private rating company is named or linked in this index.
- Why isn't Nashville in the index?
- Nashville-Davidson is a consolidated city-county of roughly 690,000 people spread across 525 square miles. Normalizing it against municipalities of 10,000 to 180,000 produces arithmetic rather than information, and breaking a city into sub-areas and measuring them against each other is the comparison Fair Housing treats most severely. The index covers eighteen separately incorporated municipalities within a 55-minute off-peak drive of downtown Nashville, each with at least 10,000 residents.
- Which categories are published today?
- Two of the eighteen: residential building permits per 1,000 existing housing units, and jobs located in the town per 100 employed residents. Both are derived end to end from federal bulk files, so both are complete across all eighteen towns on day one. The other sixteen are defined, listed on this page with the reason they are not yet published, and render as blanks until they are measured. A town page publishes only once a town has bands in at least nine of the eighteen categories — until then a town's numbers live on the category pages, where they are complete.
The measurements
Working out where you fit
These measures trade off against each other by design. If you would rather answer eight questions than read eighteen columns, the quiz walks the same tradeoffs and shows its reasoning.