Policy

Glossary

Every headline number and column header in a 2ndSight report, defined in the terms an operator or a broker actually uses. Each entry says what the figure means, what it is computed from and, the part that decides whether it gets read correctly, what it does not say.

The score

The verdict layer: the numbers most often quoted out of the report. The two baselines here are the pair most often mistaken for each other.

Concept fit

The verdict, the large percentage

What it means: The model's estimated probability that this concept, at this address, is still trading 4 years after opening.
Computed from: A calibrated survival model fitted on Chicago's licence record since 2002, reading five drivers: the citywide record of this cuisine-and-format, the operator's prior openings, how often restaurants in this zip turn over, how crowded the zip already is with this cuisine, and how that cuisine specifically has fared in it.
What it does not say: It is not a grade out of 100 and not a prediction about your business. Most scores sit in a narrow band: 8 in 10 across the city fall between 52 and 70, so 57 is an ordinary result rather than a poor one. It cannot see your menu, your operating skill, your lease terms or your capital. It describes how comparable attempts have historically ended, and nothing more. Its location signal is measured at zip level, so two addresses in one zip score identically for the same concept and operator. What the address changes is the rest of this report.

What this number means →

Average concept

The verdict, under the score and in the first scorecard row

What it means: What the model predicts for an average concept in an average location, currently 57%. This is the line to read your own score against.
Computed from: The model's own prediction with every driver set to its average value (prediction-at-mean). It is the anchor the scorecard's signed points are measured from, which is why those points sum to your score exactly.
What it does not say: It is not the same number as the base rate, and the two are not interchangeable. This is the model's prediction for an average input; the base rate is the observed outcome across a real group. They sit a few points apart in a fixed direction, so arithmetic that swaps one for the other will miss the score.

Base rate

The verdict, and the cohort panel

What it means: The share of a real, named group of past attempts that were still trading at the horizon. An observed historical outcome, not a model output.
Computed from: Counting the attempts in the cohort that reached the horizon still licensed, over the attempts whose outcome is actually observable. Attempts too recent to have a known ending are excluded rather than counted as survivors.
What it does not say: It is not the model's baseline and it is not a forecast. It describes what happened to a group in the past, carrying no adjustment for anything specific to your concept or address.

Driver / scorecard points

The verdict, the signed rows beside the score

What it means: How far each of the five inputs moved your score away from the average concept, in percentage points, with the measurement behind it.
Computed from: Each driver's contribution measured in log-odds against the average concept, then expressed as points. The rows add up to your score minus the average, and there is no hidden term or rounding fudge beyond a single point.
What it does not say: A driver is not a cause and not a lever. Operator experience is the clearest case: it is the strongest single driver and it is a correlation. Opening more restaurants does not, by itself, make the next one survive.

Concept rank

The verdict ("#3 of 27") and the alternatives

What it means: Where the concept you submitted places among every concept this address can be scored for, best first.
Computed from: Scoring each cuisine-and-format combination with enough comparable history at this address, then ordering them.
What it does not say: A rank is not a margin. The top and bottom of a 27-concept slate can sit a few points apart, so third place does not mean the two above it are meaningfully better. Read the scores, not just the positions.

Cohort rank

The verdict, beside the cohort figures

What it means: Where this concept's citywide cohort places among all cohorts that have a real denominator behind them.
Computed from: Ranking cohorts by their observed survival, having first dropped every cohort too small to support the comparison.
What it does not say: It ranks the cuisine-and-format citywide, not this address. A strong cohort rank at a weak address is a common and entirely coherent result.

The record

What the underlying data is, and where it stops. These decide whether a number is honest, and the two most commonly misread items in the report are here.

Attempt

Throughout: the unit everything else is counted in

What it means: One restaurant's run at one address, from the day it was licensed to the day the licence lapsed or the record ends. The unit the model is trained on and every count is expressed in.
Computed from: Chicago's Retail Food Establishment licence record since 2002, rebuilt into continuous runs so a renewed licence is one attempt rather than several, and a new operator at the same address is a new attempt.
What it does not say: An attempt is not a business and not a brand. One company running three locations is three attempts; a restaurant that renewed its licence eight times is still one.

Cohort

The verdict, and the closure panel

What it means: The group of past attempts your concept is compared against: the same cuisine and service format, citywide, across the years where the outcome is known.
Computed from: Filtering the licence record to attempts matching the submitted cuisine and format, then keeping only those old enough for their outcome at the horizon to be observable.
What it does not say: A cohort is not a set of direct competitors and not a list of businesses near you. It is a historical comparison group drawn citywide, and it may contain nothing within miles of the address.

Survival rate

The cohort figures, at 2, 4 and 6 years

What it means: The share of the cohort still trading that many years after opening.
Computed from: Counting attempts in the cohort that reached that anniversary still licensed, over the attempts that could have been observed reaching it.
What it does not say: The complement is not a closure rate. A 61% four-year survival does not mean 39% closed. Part of the remainder is attempts whose outcome is not yet observable. Closures take roughly three years to surface in the licence record, so any recent-period closure figure would be an undercount, which is why none is published.

Denominator / comps

Printed beside every cohort figure

What it means: How many past attempts the figure above it is actually computed from: the number that tells you how much weight the percentage can carry.
Computed from: A count of the attempts in the cohort with an observable outcome at that horizon.
What it does not say: A percentage over a small denominator is not a weaker version of the same claim, it is a different kind of claim. Read the count before the percentage, and treat single-digit denominators as directional at best.

Median years open

The cohort panel

What it means: The point by which half the cohort had closed: a plainer summary of longevity than a survival percentage at a fixed anniversary.
Computed from: The median run length across the cohort's attempts with a known ending.
What it does not say: It is not an average lifespan and not a forecast of how long you would last. Because it uses attempts with known endings, it describes the closures more faithfully than the survivors.

Censored / not yet observable

Notes under the cohort and closure panels

What it means: An attempt whose ending has not happened yet, or has happened but is not yet visible in the record. Counted in neither the survivors nor the closures.
Computed from: The licence record starts in 2002 and closures surface in it roughly three years late, so an attempt needs 4 years plus that lag before its outcome can be stated. Everything more recent is held out rather than assumed.
What it does not say: Censored is not the same as survived, and treating it that way is the single most common way to overstate a survival figure. It is also why the most recent years are missing from panels that would otherwise look more complete.

Widened comp set

The alternatives, marked on individual rows

What it means: A flag saying the exact cuisine-and-format had too little history, so the comparison was broadened (to the cuisine across formats, or the format across cuisines) to reach a usable denominator.
Computed from: A fallback applied only when the exact match falls below the minimum denominator, and disclosed on every row where it fired.
What it does not say: A widened row is not directly comparable to an unwidened one. It answers a broader question than the one you asked, and its score is correspondingly less specific to your concept.

The economics

Every figure in the economics section is a modelled range built from comparables. None of it is quoted to you by a landlord, and none of it has seen your P&L.

Revenue band

The economics, ramp and steady state

What it means: A low–mid–high range of plausible annual revenue for this concept at this address. Ramp covers years 1–2, steady state year 3 onward.
Computed from: A base annual volume for this format and price tier, corrected against the payroll records of real Chicago restaurants, then scaled by the ownership ramp, the cuisine, the neighbourhood tier, the size of the space, and how comparable restaurants nearby are actually trading.
What it does not say: It is not a forecast of your revenue, and it is not a confidence interval. We checked it against 1,422 Chicago restaurants whose sales we estimated from their federal payroll-loan records: the range contained that figure four times in ten. Read it as where the number tends to land, not as a bound it stays inside. It also assumes the doors open, the concept is executed competently, and the space trades like its comparables.

Rent band ($/SF/yr)

The economics, and wherever rent is quoted

What it means: The range of asking rents per square foot per year for comparable retail space in this neighbourhood tier.
Computed from: Asking rents collected monthly from commercial listing sources, banded by neighbourhood tier.
What it does not say: Asking rent is not achieved rent, and a band is not an offer. It excludes concessions, free-rent periods, tenant improvement allowances and percentage-rent clauses, all of which move the real number materially. Verify against the actual lease.

Occupancy cost

The economics, as an annual figure and a share of revenue

What it means: What the space costs per year, and what share of forecast revenue that consumes: the ratio a lease either fits inside or does not.
Computed from: The rent band applied to the square footage, expressed against the revenue band. Where you supplied a quoted rate, that rate is used and shown against the band.
What it does not say: It is rent-driven only. It excludes utilities, common area maintenance, property taxes passed through and insurance, so the true occupancy load on a triple-net lease is higher than the figure shown.

EBITDA / margin

The economics, the scenario rows

What it means: Modelled earnings before interest, tax, depreciation and amortisation, and that figure as a share of revenue, under each scenario.
Computed from: The revenue band less modelled cost of goods, labour, rent and other operating costs at industry-comparable ratios.
What it does not say: These are category ratios, not your books. They cannot see your supplier pricing, your wage structure, your owner draw, or your debt service, the last of which is excluded by definition.

Payback period

The economics, the scenario rows

What it means: How many years of modelled EBITDA it would take to recover the entry cost.
Computed from: Entry cost divided by the scenario's annual EBITDA, with no discounting.
What it does not say: It is not a return, an IRR or a valuation. It ignores the time value of money and assumes steady-state earnings arrive and hold. A blank payback means the scenario does not recover the entry cost at all.

Entry cost

The economics

What it means: The modelled capital required to open: build-out, equipment, licences and opening costs.
Computed from: Per-square-foot build-out comparables for the format, applied to the space's size, unless you supplied a figure.
What it does not say: It is not a quote and not a budget. Kitchen condition, venting, grease interception and ADA work are the items that most often break this estimate, and none of them appear in any public record.

The ground

From the ground to the movement: what is physically around the address and, the part to read carefully, the proxies standing in for things nobody publishes directly.

Neighbourhood tier

The economics and the ground

What it means: A grouping of Chicago neighbourhoods by retail rent level, used to select the rent band and the revenue comparables.
Computed from: Banding neighbourhoods by their observed asking-rent distribution.
What it does not say: A tier is a rent grouping, not a quality or desirability judgement, and not a verdict on the neighbourhood. Tier boundaries are thresholds, so two addresses either side of one are more alike than their labels suggest.

Trade area

The patrons

What it means: The residential population around the address whose demographics are reported: population, income, household composition and age mix.
Computed from: US Census American Community Survey figures for the block groups around the address.
What it does not say: It describes who LIVES there, not who eats there. At a downtown or transit-adjacent address the daytime and visitor populations can dominate trade, and neither appears in these figures at all.

Demand band (High / Medium / Low)

The competition and the movement, on competitor rows

What it means: How much attention a venue pulls relative to others, banded into thirds.
Computed from: Google rating multiplied by review count, ranked within the local comparison set, or against the citywide distribution when the local set is too small to support three bands.
What it does not say: It is a visibility proxy, not measured demand. It is not foot traffic, covers or revenue, and it systematically favours venues that have been open longer and have had more time to accumulate reviews.

Competitive density

The competition

What it means: How many comparable restaurants operate within a radius of the address, and how crowded that makes it for this cuisine.
Computed from: Counting currently-licensed restaurants within the radius, filtered to genuine restaurants and then to the relevant cuisine.
What it does not say: Density is not automatically bad. Clustering raises both competition and draw, and the model reads those as separate drivers rather than assuming one dominates.

Daypart

The clock

What it means: When restaurants around this address are open (breakfast, lunch, dinner and late night) as a profile of the area's trading pattern.
Computed from: Published opening hours for enriched venues in the surrounding area.
What it does not say: Open is not busy. These are stated hours, not traffic, so a daypart the area is open for is not evidence that the area trades in it.

Walkability index

The patrons

What it means: The EPA National Walkability Index for the block group containing the address, on a 1–20 scale, with the band it falls in.
Computed from: The EPA's published index joined to the address's block group. The distance from the address to that block group's centroid is printed so a loose match is visible.
What it does not say: It is a block-group measure, not a measure of your doorstep. A high index does not mean footfall passes this particular unit, and the match distance is the number to check before relying on it.

The limits

Stated here rather than buried, because these three constrain how precisely every other number can be read.

Address precision

Noted where the report resolves the address

What it means: How precisely the address could be located: to the address itself, to the block, or only to the ZIP code.
Computed from: Geocoding the submitted address against the licence record and the civic datasets, then reporting the finest level that matched.
What it does not say: Where the read resolves only to ZIP, the score is a ZIP-level quantity: two different addresses in that ZIP score identically. That is a real limit on resolution, not a rounding artefact, and it is stated rather than hidden.

Restaurant (vs licence holder)

Every count of businesses in the report

What it means: A genuine restaurant, as opposed to any holder of Chicago's Retail Food Establishment licence.
Computed from: The licence record filtered to exclude the categories that hold the same licence without being restaurants: groceries, corner shops, pharmacies, hotels and school cafeterias.
What it does not say: The licence record is not a restaurant list, and the gap is not small: roughly 45% of licence holders are not restaurants. Any figure elsewhere that counts licence holders is counting a different and much larger population.

Drift

Why the model carries a trained-through date

What it means: The world moving away from the data the model was fitted on. The scoring model is a frozen artifact trained through a fixed date; drift is what accumulates between that date and the day you read the report.
Computed from: Re-scoring a fixed reference cohort against the current licence record and comparing the result to a recorded baseline. Because the cohort is held constant, any movement is the underlying population changing rather than the code changing.
What it does not say: Drift is not an error rate and it is not a correction applied to your score. It says whether the population the model describes still resembles the one it learned from: a check on whether the model is still the right instrument, not an adjustment to the number it produced. Nor is it the same as code drift, which is a copy of a rule disagreeing with the original and is caught separately.

If a number is not on this page

That is a gap worth telling us about, not a term you were expected to already know. The list is built from what the report actually renders, 28 entries at present, and a figure on screen with no entry here is a defect in this page.

Questions about this page? Email hello@use2ndsight.com. We answer within one business day.

Terms Privacy Refunds Data sources Security Methodology