How we measure property managers.

Every operator in our coverage markets is evaluated against two eligibility tests before a scorecard is produced. The tests are designed to filter single-rental owners and one-off listings while admitting operators with meaningful market presence.

A property manager qualifies if both of the following are true:

1. At least 30 listings observed in the trailing 12 months (anchored to our data refresh date for the market).
2. At least three distinct addresses or at least one community where we observe thirty or more distinct units in the trailing 12 months.

The two-pronged second test admits both scattered-site operators (who hit the diversity threshold through breadth) and single-asset multifamily operators (who hit it through depth at a single community).

Three operator states.

Clearing the two tests above is not the whole story. An operator can qualify on size and still have stopped appearing in our listing data — and those two facts deserve different treatment. Every operator we observe therefore sits in exactly one of three states.

The recency window is 60 days, measured from the market's data refresh date. It is judged at the operator-name level, aggregating the newest event across the id fragments an operator churns through over time (see §07), so an operator that is still listing is never called dormant on the strength of one stale fragment.

What dormant does and does not claim. It is a statement about our listing record, not about the business. We can see that an operator stopped posting listings with us on a particular date; we cannot see why. A portfolio may have changed hands, syndication may have moved elsewhere, or the operator may simply be between lease-ups. So we report the observed fact — the date of the last listing we saw — and stop there.

Why they are held out of the baselines. A dormant operator's twelve-month window ended whenever their listings did, so ranking it against operators whose window ends today compares two different things. Dormant operators are still scored against the ranked cohort — their metrics and stars are computed on the same scale — but they are not membersof it, and they are absent from every median, percentile, and eligible-operator count. That separation is what makes the state purely additive: no ranked operator's numbers move because some other operator went quiet.

Earlier versions of this methodology deleted these operators outright. That silently discarded real market presence — in one market, an operator carrying more than four thousand listings in the trailing year simply vanished from the answer to “who operates at scale here?” Keeping them visible and clearly labeled is more useful than pretending we never saw them.

Earlier methodology versions labeled this window “T6M” on the market headline tile, which was a labeling drift; the actual eligibility filter has always been T12 in production. v0.6.3 (Patch 2) corrects the surfaced label so it matches the underlying computation. No operator gains or loses eligibility from the relabel.

Category exclusions.

Two filters remove non-operators before any operator is formed. First, listings whose source company type is a data-platform artifact — property-management marketing software or listing-syndication services — are dropped at the row level; they never become an operator, a scorecard, a search result, or a market count. Second, a small curated denylistcatches source-misclassified artifacts that slip past the type filter — for example, a listing-syndication platform mislabeled as an “Owner” in the source data that would otherwise top a market. The denylist is deliberately narrow, reviewed as a code change, and surfaced read-only in our admin tooling. These are the exclusion rules the earlier methodology promised to document once they existed.

Broker vs. property manager.

We type each operator as a property manager or a brokerfrom the source company-type signal — a majority vote across the operator's listings, with the parent company's type taking precedence. Brokers are tracked and scored, but within their own cohort: a property manager is only ever ranked against other property managers, and a broker only against other brokers (the rent-performance baseline is likewise split), and brokers are hidden from the default ranked lists behind a “show brokers” toggle.

Because some franchise operators — notably Real Property Management offices — hold brokerage licenses and are tagged “Brokerage” at the source despite operating as property managers, a small curated override reassigns those specific operators to the property-manager cohort. The override is keyed on the operator's name, so it survives the source re-issuing new internal ids.

The URU is the four-level hierarchy Dwellsy assigns: building or community → address → unit → room/bed. All metrics in this scorecard operate on the unit level (the third tier), aggregated up to the operator's full portfolio.

URU resolution happens upstream of this scorecard in Dwellsy's core data infrastructure. The scorecard consumes resolved URUs as inputs — it does not derive them.

Operator typemeasures how the operator's portfolio is organized — whether their units are concentrated in multi-unit communities (multifamily buildings, build-to-rent communities, condo developments) or distributed across individually-managed single-family rentals. v0.6.2 refines the v0.6.1 three-way split into a four-way axis by splitting MF/BTR by median community size.

We group the operator's portfolio by community and count distinct units the operator manages at each community. A community is concentrated if the operator manages 10 or more units there. From that:

• SFR (Scattered) — concentrated share < 30%.
• Small MF/BTR — concentrated share ≥ 70% AND median concentrated community size 10–49 units.
• Large MF/BTR — concentrated share ≥ 70% AND median concentrated community size ≥ 50 units.
• Hybrid — concentrated share between 30% and 70% (no scale split).

Unit-type override (applied first).Concentrated share alone conflates “scattered” with “single-family.” An operator running apartments spread across many small buildings can look Scattered by concentration even though they are plainly a multifamily operator. To correct this we compute each operator's house share — observed house-type units divided by house-plus-apartment units — and when it is ≤ 10% (essentially all apartments) we classify the operator as MF/BTR regardless of concentrated share. The Small-vs-Large split then follows the same median concentrated community-size rule (≥ 50 units → Large, otherwise Small). This override runs before the concentrated-share bands above; only operators that are not apartment-dominant are classified by concentrated share. About a third of eligible operators carry the override.

Why the Small vs Large MF/BTR split. MF/BTR community size is a structural distinction. A 200-unit Class A operator has a different risk profile and different operating profile than an operator running 8-unit walk-up small MF. The v0.6.1 five-cell taxonomy collapsed these into one MF/BTR bucket; v0.6.2 makes the distinction visible.

Scale(Institutional vs Independent) measures the operator's footprint. An operator is Institutional if they manage 500 or more distinct units across all Operator IQ coverage markets in the trailing 12 months, Independent otherwise. The 500-unit threshold is a judgment call; in practice it cleanly separates names that operate at scale requiring institutional capital structures from established local and regional operators.

Scale classification considers an operator's observed presence across all Operator IQ coverage markets, not just the market in which a given scorecard is published. Operators are Institutional if their combined trailing-12-month observed units across all our covered markets meet or exceed 500. This rule lets us recognize national operators whose footprint in any single market falls below the threshold but whose cross-market scale is substantial. The Hybrid bucket does not carry a scale split — a Hybrid operator is simply Hybrid regardless of cross-market urus.

Figure 1. The seven-cell taxonomy. The type axis (rows) splits operators into SFR, Small MF/BTR, and Large MF/BTR by the apartment-dominant override, concentrated share, and median community size. The scale axis (columns) splits each type into Independent and Institutional by cross-market urus. Hybrid carries no scale split — it is its own classification. Cell colors match the quadrant badges used elsewhere on the scorecard.

The taxonomy is structural, not evaluative. Each cell contains operators of varying quality. The classification answers “what kind of operator is this?” — the rest of the scorecard answers “how well do they operate?”

Why this measure is structural to operator type. Single-family operators cannot meaningfully cherry-pick which inventory they show on Dwellsy. Every property is unique. A renter searching for a three-bedroom house in a specific neighborhood is looking for that specific home with its specific layout, yard, and location — the operator cannot substitute Property B for Property A. To capture any rental, the SFR operator must list it. The cherry-picking risk is structurally low.

Multifamily and BTR operators sit on undifferentiated inventory in a leasing office. A community with 20 vacant two-bedroom units can list five and route walk-in prospects to the rest. The cherry-picking option is structurally available, and some operators use it — historically as a strategy to control which units appear in third-party search results.

Community Visibility measures whether this is happening.

Scope. We compute Community Visibility for operators who meet three conditions: at least one community where they manage 30 or more units, at least 50% of their inventory in concentrated communities, and at least 12 monthsof listing history at those communities. Operators who don't meet all three conditions don't have this section on their scorecard — for them, the question is either unanswerable (Scattered operators, where there's no honest denominator) or not yet measurable (operators below the tenure gate).

Formula.For each qualifying community, we compute the operator's expected listing volume in the trailing 12 months based on the community's true unit count (the structural community-size field from Dwellsy's core data, present in every listing row) and a default annual turnover rate of 20%. We compare that expectation to the operator's actual listing count.

The 20% turnover assumption matches the empirical cross-market norm across Chattanooga, Jacksonville, and Nashville under v0.6.1 — and aligns with the U.S. national rental-household mobility rate.

The ratio answers:

“Of the units that should have plausibly turned over and been listable in T12, how many did this operator actually list?”

Three-state taxonomy (v0.6.1, unchanged in v0.6.2). v0.6 simplified the Community Visibility output to three states. The legacy fourth state (“above expected — comprehensive coverage”) was retired because it implied a comparative judgment the data couldn't support; a visibility ratio above 1.0× simply means the operator is listing comprehensively at higher-than-default turnover, which is a positive signal but doesn't warrant a separate color-coded tier.

Reported states:

• Partial visibility (ratio <0.5×) — observed listings represent less than half of expected. Notably reduced visibility relative to community structure.
• Likely partial visibility (0.5–0.8×) — most but not all expected listings present. Possible normal turnover variation, possible selective listing.
• Comprehensive visibility (≥0.8×) — listings cover the substantial majority of expected turnover. Within expected range for a fully-transparent operator.

Ratios materially above 1.0× are meaningful signal — they identify operators visibly more transparent than the cohort norm, which is a credibility-positive signal. Institutional Class A MF communities typically turn over faster than the cohort average; a 2.1× visibility ratio for an operator like UDR reads as the operator genuinely listing comprehensively.

Days on Market (DOM).

DOM measures lease-up speed — how efficiently an operator moves a vacant unit off the market. It is the joint-largest component of the composite (30%) because it captures pricing strategy, marketing reach, and lease-up execution in a single clean signal.

For each trailing-12-month listing where the listing was deactivated on or after it was created, per-listing DOM is the number of days the listing stayed live.

The operator's DOM is the median across its house and apartment listings. We use median rather than mean because listing-duration distributions are right-skewed. Lower DOM means faster lease-up: the metric is inverted for star assignment, so an operator in the top quartile of its cohort (fastest lease-up) earns a gold star.

Tenant Retention.

Longer tenure reflects multiple compounding operator behaviors — tenant screening, property condition, responsiveness, fair renewal pricing — and is one of the cleanest behavioral signals in the scorecard.

We measure retention with a Kaplan-Meier survival estimateof how long units hold a tenant. For every unit an operator has listed two or more times, each occupied interval — from a listing's deactivation to the same unit's next listing creation — is a completed tenancy, provided it lasts at least 3 months (shorter re-posts are re-listing noise and are dropped). A unit whose most recent listing has closed but not yet re-listed contributes a right-censored observation: the tenant may still be in place, so we know only that tenure is at least the time since deactivation.

Pooling every unit's events and censored observations, the Kaplan-Meier product-limit estimator yields S(t) — the probability a tenancy lasts at least t months. The ranked metric is S(18), the share of tenancies that reach 18 months, shown as a percentage and phrased “about X% of tenancies reach 1.5 years.” Higher is stickier, so a top-quartile operator earns a gold star; we also report the full curve at 12, 18, and 24 months. Because censoring is handled natively by the estimator, an operator with a short observation window is no longer biased downward the way a raw gap-median would be — the censoring correction the earlier methodology deferred is now the metric itself.

We bucket trailing-six-quarters listings by quarter, compute median rent within each bedroom bucket (1-bedroom, 2-bedroom, 3-bedroom-and-up), and average across buckets weighted by the operator's overall bedroom mix. The result is the mix-adjusted median rent per quarter.

The trajectory chart shows the last six quarters. The headline YoY change compares the two most recent trailing-four-quarter windows: it is the mean mix-adjusted rent across the most recent four quarters divided by the mean across the prior four quarters, minus one. Averaging four quarters on each side smooths quarter-to-quarter noise; the figure is computed only when each four-quarter window has at least two non-null quarters.

Formula.We compute the operator's mix-adjusted YoY rent change (from §06) and subtract the MSA cohort median YoY change over the same period. Operators whose rents grew faster than the cohort median are positive on Rent Performance. Operators who lagged the cohort are negative.

The delta shown on a scorecard uses the market-wide property-manager median as its baseline. For ranking — the percentile, star, and composite contribution — each operator is compared only within its own operator type: property managers against the PM median, brokers against the broker median, never pooled.

This isolates operator pricing capability from inherited portfolio quality. Every operator in the cohort is compared to the same peer-group baseline during the same period. Class A operators are not rewarded for managing high-rent inventory; they are rewarded only when they push rents faster than other Class A operators (who would be reflected in the cohort median). Similarly, Class C operators aren't penalized for low rent levels — only for failing to push rents at peer rates.

Confounders we disclose.The metric is meaningful but noisier than DOM or Tenancy. We control for bedroom mix but not for square footage, neighborhood, building age, or amenity differences within an operator's portfolio. Three real noise sources:

• Submarket exposure (operators concentrated in gentrifying neighborhoods see faster growth regardless of skill).
• Mix shift within the trailing window (an operator adding higher-rent properties mid-window shows artificial growth).
• Capital events (operators who renovated mid-window push rents through investment, not pure leasing skill).

We weight Rent Performance at 10% of the composite ranking to reflect these confounders. A future version (v0.7) will refine the metric to compare only units that appear in both periods — eliminating the mix-shift confound and likely justifying a heavier weight at that point.

Market rent growth aggregate.

Each market displays a median year-over-year rent growth figure computed across the ranked operators in that market. The figure represents the typical ranked operator’s portfolio rent trajectory over the trailing 12 months. The benchmark line (“vs national”) compares the market value against the median across all ranked operators across all coverage markets — a single national number that does not vary by market.

Computation.Each operator’s value is its own mix-adjusted YoY rent change (the four-quarter-over- four-quarter figure from §06) — notthe cohort-relative delta. The market aggregate is the equal-weighted median of those per-operator values across the market’s eligible property managers(brokers are excluded). Operator-equal weighting keeps the metric interpretable — “the typical property manager in this market” — rather than letting a few large portfolios dominate; a unit-weighted alternative is a possible future refinement.

The market-vs-national delta is surfaced as a tile benchmark line (green if > +0.2 pp, orange if < −0.2 pp, neutral within the band). The national reference is a single value computed across every covered market; where a snapshot does not yet carry it, the tile shows the market value without a national comparison. Submarket-level rent growth is not computed — listing-level geographic aggregation with minimum-N controls remains a future candidate; under a submarket filter the headline tile retains the MSA-wide value with an explicit scope annotation.

State-level aggregates.

State-level aggregates pool operators across all MSAs in a state. The counts — active operators and operators eligible for ranking — deduplicate by canonicalOperatorId, so a multi-market operator counts once per state rather than once per MSA. The state-level medians (DOM, rent growth) still pool one value per operator per MSA, so an operator that appears in several in-state MSAs contributes once for each; pushing canonical dedup down to the medians is a future candidate.

The state landing pages at /property-managers/[state] surface four operator-weighted tiles — active operators (sum across MSAs), eligible for ranking (sum across MSAs), median DOM T12 (operator-weighted median across the pooled in-state ranked operators), and rent growth T12 (operator-weighted median of pmYoyChangeacross the same pool). Median DOM and rent growth carry a “vs national” benchmark line where the national reference is the operator-weighted median across every ranked operator in every covered MSA — the same single national number that Patch 3 already computes for market-level rent growth, and its DOM analogue computed at runtime.

Single-MSA states get the same UX as multi-MSA states; the state page renders one MSA card. As coverage expands and adds new MSAs in already-covered states, the state page auto-updates without any data-layer changes: state membership is derived from each market’s state field.

Share trajectory.

Share trajectoryshows how this operator’s share of ranked-cohort listing activity has changed year-over-year. The metric is computed across continuing operators with substantial presence in both periods (at least 30 listings in each), normalized to share of the cohort’s total listings so that proportional pipeline expansion across all operators produces a 0% trajectory. Real movement indicates relative gain or loss of market position.

Continuing cohort. Operators with at least 30 listings in BOTH the trailing 12 months (T12) AND the prior 12-month window (T24-T12). Operators outside the cohort fall into one of two display categories: Newly tracked(≥30 listings T12 but <30 in the prior window) or New operator (zero prior listings). These operators see a context pill in place of a comparison number; their data is excluded from the cohort median.

Why share rather than absolute? An earlier version of the metric computed absolute year-over-year listing-count change. A pressure test surfaced three biases that made the absolute version unusable: pipeline-coverage expansion (every operator appeared to grow even if they did nothing), thin-baseline noise (operators with 1 listing in the prior period produced absurd growth percentages), and survivor bias (operators that shrank to zero between periods were systematically excluded from the median). The share-based reframe addresses the first two biases directly and partially addresses the third. The pressure-test results, post-revision, show plausible directional signal consistent with known market dynamics — Phoenix at +10.07% (established operators consolidating), Memphis at −9.89% (SFR aggregators entering aggressively), Clarksville at −15.81% (heaviest fragmentation in the v0.6.3 footprint).

Why no star treatment? Share trajectory is a contextmetric, not a performance one. A higher share isn’t reliably better: longer tenancies → fewer relistings → lower share (good operationally, lower share); improving operationally drops the share via the same mechanism. M&A activity, portfolio composition shifts, and new entrants all move share without reflecting operator health. Star treatment requires a metric where “higher = better” is reliably true; share trajectory fails that test. The scorecard shows the metric with cohort + national context and methodology disclosure so readers can form their own judgment.

Residual caveats. Coverage bias is only neutralized if pipeline improvements affected all continuing operators uniformly — non-uniform improvement (e.g., a new ingestion source biased toward aggregators) would still distort. Survivor bias is now partly addressed: an operator that stops listing is classified dormant rather than deleted, so it stays visible with the date of its last observed listing (see §01). Bias persists for operators that shrank toward zero without crossing the recency window. Listing-level re-listing methodology affects numerator and denominator alike but is a counting artifact worth acknowledging. The metric is shown for context and is not used in ranking or composite scoring.

Operator identity.

Two problems complicate “who is this operator?”: the same operator can appear in several markets, and the source data periodically re-issues an operator new internal ids, splitting one operator into several records within a single market. We resolve both into a single canonical operator entity, which powers the cross-market profile at /operators/[canonical-slug]and dedup’d state-level counts.

Cross-market identity is ID-based.Operators appearing in multiple markets are linked primarily by Dwellsy’s own parent-company id (parentCompanyId), taken per operator as the modal parent id across its listings. At merge time, operators sharing a parent id are grouped into one canonical entity with a shared id slugged from the parent-company name. This id-based grouping is authoritative— it overrides any name-based assignment. It unites a parent’s differently-branded entities and keeps same-named-but- differently-owned operators apart, which is exactly the roll-up / parent-entity mapping earlier methodology had deferred.

Curated name mapping is the fallback.Operators that carry no parent id fall back to a human-curated cross-market mapping held in versioned decision files. These are reviewed for same-name false positives — generic names that collide across markets but represent unrelated companies (e.g. two distinct “Trinity Management Company” operators) are excluded and stay separate. Curated groupings are protected: the id linker never overrides them. A conservative name normalization — lowercase, strip legal suffixes and punctuation, collapse whitespace — is used to propose candidate matches for that review and to normalize parent-company names so corrected casing survives the merge.

Within-market fragment merge.

When the source re-issues an operator new internal ids, one operator fragments into several no-parent records in the same market. Two mechanisms re-pool them. An exact-tier auto-merge groups no-parent records whose names are identical after stripping legal suffixes and punctuation — but only when the shared name is distinctive(at least two tokens, at least one not a generic word like “property” or “management”), so purely-generic or single-token names never auto-merge. A curated merge list handles the remainder, and curated decisions win over auto-merges on any conflict.

Guards.Placeholder names (“Company Name Not Provided” and similar) and any pair on an explicit do-not-merge denylist are never merged; structural invariants — every survivor is a real member, no record spans two distinct names, no slug collisions — are asserted on every pipeline run, and a sign-off report is emitted for review. Below-threshold id-bearing fragments are surfaced only to an internal merge tool, never to the seed, the operator table, or search.

State-level count dedup. State-level operator counts deduplicate by canonicalOperatorId— a multi-market operator that appears in Nashville, Memphis, and Clarksville counts once in Tennessee’s state-level total.

Concession activity.

For every operator in coverage, the classifier scans T12 listing descriptions for stereotyped concession language — “one month free”, “move-in special”, “no deposit”, percent-off promotions, and similar patterns — and computes the share of T12 listings that mention at least one. The result is surfaced on the scorecard as the operator’s concession rate, with a listing-weighted market concession rate as cohort context.

Regex-based, v1 catches stereotyped language. Detection is pattern-matching, not semantic. The v1 dictionary covers about a dozen pattern families (free month, percent off, dollar off, no/reduced deposit, move-in special, explicit concession, rent reduction, lease special, limited- time offer, waived fee, free rent). Indirect or paraphrased language (“ask us about specials” without naming the special) will be missed. A v2 LLM-grader pass is on the v0.7 backlog.

Context, not ranked.Concession activity does not feed the composite ranking and does not award stars. It’s presented as a present-tense signal of demand or supply stress at the operator level — high participation can mean any of: aggressive lease-up, soft submarket, large institutional discounting program. Read alongside DOM, rent growth, and share trajectory rather than in isolation.

Cohort comparison. The market reference is a listing-weighted concession rate — total concession-mentioning T12 listings divided by total T12 listings across the ranked cohort (operators with no T12 listings are excluded). Listing-weighting is used instead of a per-operator median because the per-operator distribution is heavily zero-inflated, so its median is often just 0. Operators more than 20 percentage points above the market rate receive an orange accent (elevated concession activity vs the cohort); operators more than 20 pp below get a green accent (low concession activity).

Five subscores, each on a 0–100 scale, are computed from trailing-12-month listings. Each richness subscore saturates near the p90 of the cross-market operator distribution, so the top decile earns 100 and the rest spread across the range:

• Completeness — percentage of listings that are fully populated on all three core fields: a non-empty description, at least one photo, and at least one amenity. The test is all-or-nothing per listing — a listing missing any one of the three does not count toward completeness.
• Amenities — the mean number of amenities per listing, scaled so that an average of 18 reaches 100: min(100, 100 × mean_amenities ÷ 18).
• Description — a 0.5 / 0.5 blend of text length and content richness, assessed over listings that have a description: 0.5 × min(100, 100 × mean_distinct_words ÷ 195) + 0.5 × 100 × min(1, mean_content_categories ÷ 6). Length uses distinct words (robust to whitespace or boilerplate padding); content richness counts how many of four descriptive areas — amenities, location, transit and parking — the prose touches, so an informative listing outscores a long but repetitive one. Rules are scored separately, below. An operator needs at least five non-blank descriptions to be assessed on the non-blank subset; below that, blanks count (their absence is already reflected in Completeness).
• Photos — the median number of photos per listing, scaled so that a median of 30 reaches 100: min(100, 100 × median_photos ÷ 30).
• Stated rules — whether the listing tells a renter the terms: pet policy, fees and deposits, and lease requirements. Scored on how many of those three areas a typical description covers, with two of three reaching 100: 100 × min(1, mean_policy_areas ÷ 2). Assessed on the same non-blank basis as Description — a blank description is already penalised by Completeness, and counting it again here would charge an operator twice for one omission. These three were previously folded into content richness as three of seven interchangeable areas, which let an operator saturate that subscore on amenities and location while telling a renter nothing about the terms. They are also the least-stated part of a listing in practice, which is why they now carry their own weight.

The reported Marketing Discipline score is a weighted blend of the five subscores, not a simple average:

Operators with consistently well-prepared, informative listings score in the 80s and 90s. Operators with sparse data, missing photos, or threadbare descriptions score lower.

Scored against a fixed bar, not a cohort.

Marketing Discipline is the one metric whose star is awarded on an absolute threshold rather than a rank within a peer cohort: gold at 80 and above, silver at 70 and above, everywhere and for every operator.

Every other metric has to be market-relative, because it measures something local — days on market, rent and retention all move with conditions in the metro. Listing completeness does not. A complete listing is complete in Bozeman and in Los Angeles, so a fixed bar is the honest comparison, and it is the one an operator can act on: the gap to 80 is a specific list of things to add, whereas a percentile only says to outrank whoever happens to be nearby.

Earlier versions ranked this metric within the same cohorts used for the others, which produced the perverse result that an operator scoring 60 could earn gold for leading a weak cohort while one scoring 82 earned silver in a strong one — while the scorecard printed the raw score beside it. The bars above were set against the national distribution: 19.1% of ranked operators clear 80 and a further 23.1% clear 70, close to the shape the previous quartile ranking produced.

The composite is a single weighted score computed from the metrics above. It is versioned and load-bearing internally, but it is never surfaced on a scorecard as a score or an ordinal rank — per our standing rule, scorecards show only stars, values against a cohort benchmark, and positions. The composite’s only visible effect is as the final tie-break in market-list ordering, which leads with gold- then silver-star counts and falls back to the composite only to break ties within an equal-star bucket. The weights still set that ordering and drive the PDF export, so we document them here.

Weights for operators with Community Visibility computed:

Weights for operators without Community Visibility (the section is suppressed for Scattered and Hybrid operators below the visibility gate, and for MF/BTR operators under the 12-month tenure threshold):

The 15% normally allocated to Community Visibility redistributes proportionally to the other four components. Both schemes sum to 100%, so composite scores remain comparable across the full cohort.

Marketing Discipline (15%) and Community Visibility (15%) are secondary but meaningful signals. Marketing Discipline reflects listing-side rigor; Community Visibility reflects transparency. Both are real quality differentiators, both are harder to game than they look, and both deserve weight without dominating.

Rent Performance (10%) is included as a pricing-skill signal but weighted lower than the cleaner metrics due to its documented confounders.

What we do not weight. Rent level. Portfolio quality. National scale beyond the MSA. These are descriptive characteristics, not performance signals. We surface them as context but do not let them drive operator rank.

Star system.

In place of a rank, each metric earns a binary star. Per-metric stars are what the scorecard actually leads with — they appear in the header star summary, on each operating-metric card, and in the market-list row chips.

• 🌟 Gold star — top quartile of the applicable cohort (≥75th percentile).
• ⭐ Silver star — second quartile (50th–75th percentile) — above-median position within cohort.
• No star— below the 50th percentile. The cohort qualifier still renders (“Present in cohort”) but no star icon. This reinforces operator dignity — top performers earn stars; others simply have no star.

Cohort hierarchy. Star assignment requires choosing which cohort to compare against. v0.6.2 pre-computes three percentile ranks per metric per PM and selects the applicable cohort via a fallback waterfall:

1. Primary cohort — same MSA + same 7-cell quadrant. Used if N ≥ 10.
2. Fallback cohort — same MSA + same operator type (SFR / MF/BTR / Hybrid), any scale. Used if primary N < 10 and fallback N ≥ 10.
3. MSA cohort — all eligible operators in the same MSA. Used as the final fallback.

The cohort label displayed in the scorecard (e.g., “Gold star · Chattanooga SFR Independent cohort”) reflects whichever level was actually selected.

Broker / property-manager partition. Every cohort — and the rent-performance baseline — is partitioned by operator type, so property managers are compared only against other property managers and brokers only against other brokers. The two operator types are never pooled at any level of the waterfall.

Lending signals.

Earlier versions surfaced a dedicated block of four underwriting-oriented signals. That standalone block has been retired on the web scorecard, and one signal — the old Vacancy Signal — has been removed entirely. The three survivors are now folded into the scorecard’s Scale & Fit section as position context rather than a separately scored section:

• Operator Stability— the operator’s observation history (years visible) shown against a cohort median. Context only; no star.
• Geographic Concentration — top-3 city share of observed urus, with a cohort median. Descriptive — no star; concentration is neither inherently favorable nor unfavorable.
• Pricing Tier— the operator’s latest mix-adjusted median rent positioned within the MSA rent distribution: Premium (≥75th pct) / Mid-market (25–75th) / Value (<25th). A positional label, not evaluative.

For anyone who wants the original at-a-glance panel, the PDF exportstill renders a dedicated three-signal “Lending Signals” page (Operator Stability, Geographic Concentration, Pricing Tier). None of these feed the composite.

Things we measure cleanly. Lease-up speed, tenant retention, listing data completeness, multifamily/BTR transparency, basic portfolio classification.

Things we measure with caveats.Rent Performance carries known confounders (submarket exposure, mix shift, capital events). Ordering within thin-data buckets may favor small-sample outliers; per-metric qualification gates suppress the least-supported values, but we don't yet apply a graded confidence discount.

Things we don't yet measure.

• Operator transparency for Scattered (SFR) operators. The cherry-picking question is unanswerable for SFR operators in the listings data alone — there is no external denominator we can construct. SFR Credibility is deferred to v1.x, pending claim-flow portfolio attestation.
• National scale beyond our covered markets. Operators with substantial portfolios in markets we don't yet cover may classify as Independent under our methodology even when they operate at institutional scale nationally. Resolution path: expanded market coverage and operator portfolio attestation via the claim flow.
• Granular unit quality (square footage, amenities, year built, condition) beyond bedroom count.
• Submarket exposure within an MSA.

Things this scorecard cannot tell you. Whether the operator will renew their lease with you. Whether a specific unit is well-maintained. Whether the operator is currently for sale or in a transition. Whether market-level conditions are favorable.

Observation precision.

Every figure on a scorecard is qualified as observed, not total portfolio. The seed surfaces three distinguishable unit-count fields per PM so templates can phrase precisely:

• urusT12 — distinct units observed listing in the trailing 12 months. The smallest, most-conservative number.
• observedCommunities — count of concentrated communities where we observe the operator listing.
• observedCommunityTotalUnits— sum of the top-down PM-managed unit counts across those observed communities. A proxy for portfolio scale at those locations — not a claim about the operator's full portfolio.

Templates phrase explicitly: “managing 8 observed large multifamily communities in the Nashville MSA — communities totaling approximately 2,400 units, with 1,069 distinct units observed listing in trailing 12 months.” We never claim “manages 1,069 units” or “operates 2,400 units” — both would imply we know the operator's full portfolio.

Estimated managed units.Because observed urus undercount an operator’s book — a unit only surfaces when it lists, on turnover — we also publish a portfolio-size estimate that scales observed urus back up by unit-type turnover (house urus × 3.3 + apartment urus × 2.6). We report it as a band — <50 through 1,600+ — and never as a point figure. Calibration against operators who told us their own count showed the estimate runs materially low for apartment-heavy books, and that what remains after every available signal is coverage: units that never list with us at all. No multiplier recovers those, so the estimate is a floor rather than a census, and a band is the most we can honestly claim. It is never fed into the composite, and is documented in full on the portfolio-estimator methodology page.

Operator-dignity language gate.

Every auto-generated string — executive summaries, distinguishing characteristics, map narratives — passes through a dignity-language validator at seed time. Forbidden tokens include weak, poor, strong, excellent, underperforming, manages X, operates X. Acceptable replacements use quartile language and observation qualifiers: “Gold star · Lease-up Performance, top quartile in cohort” rather than “Strong leasing performance.” “5 communities observed in our coverage” rather than “Their portfolio of 5 communities.” The system measures; it does not editorialize.

Deferred to future versions.

The following improvements are tracked for future releases:

• Same-unit-controlled Rent Performance — compares only units present in both periods, eliminating the mix-shift confound; likely justifies a heavier composite weight.
• Minimum-N confidence multiplier on the composite — a graded discount for thin-data operators, beyond the current per-metric qualification gates and rationale text.
• SFR Credibility instrument — deferred until claim-flow portfolio attestation provides external scope data for scattered operators.
• Submarket-aware peer cohorts and submarket-level rent growth — activate when listing-level geography with minimum-N controls is added.
• Persistent eligibility per window — a stability component (consistent eligibility across refreshes) not yet computed.
• Canonical dedup on state-level medians — state counts already dedup by canonical identity; the medians still pool one value per operator per MSA.
• Operator dispute / appeal process — as scorecards reach operators, a defined correction path.

Material changes — new metrics, re-weightings, threshold shifts — bump the version. Cosmetic changes do not. Prior versions remain accessible, and every scorecard carries the version it was computed under so historical scorecards can be interpreted in their original frame.

Recent versions:

Data is refreshed monthly. The current snapshot reflects listing activity through Aug 20, 2026.