What was compared

The analysis asks how four declared amenities occur together among a narrowly
defined set of Airbnb listings. It uses three publicly available Inside Airbnb
city files selected before viewing the results: Austin, London and Bangkok. The
cities are purposive examples of different host markets, not a sample representing
three countries or the world.

A host with a comparable property can use these results as local supply context
when deciding which genuine features to explain or photograph. The study does not
estimate whether adding an amenity, changing a caption or buying advertising would
increase bookings. No price, review, availability, occupancy, revenue, host-profile
or location-coordinate analysis was performed.

Sources and dates

The download directories are dated September 17, 2026 for Austin and London, and
September 22 for Bangkok. Those are source-batch dates. The actual last_scraped
ranges in the files are September 19–20 for Austin, September 18–23 for London,
and September 23–26 for Bangkok. Use the capture ranges when describing timing.
The source manifest pins the exact compressed bytes with SHA256 hashes.

The official dictionary's visible detailed-listings v4.3 sheet supplies the field
definitions. The downloaded workbook also contains a hidden v4.7 tab; it is not
presented as the current visible dictionary. The workbook's SHA256 is recorded in
the frozen plan. The source publisher's stated assumptions remain relevant: this
is a snapshot of publicly presented listings, not a booking census or property
inspection.

Listing unit and cohort

The unit is a unique listing ID within its city file. It is not necessarily a
unique physical property or host. Nonempty IDs must consist of ASCII decimal
digits with at least one nonzero digit. IDs are never converted to floats or
integers; leading-zero strings remain distinct. This avoids loss of precision for
large identifiers.

Before selecting the cohort, all retained analytical fields are compared as exact
source strings for each ID. Identical duplicates retain one record. A conflicting
analytical record or a malformed nonempty ID forces a hold. Missing IDs are counted
and excluded. The flow separately reports original conflicting rows and distinct
conflicting records, so duplicate accounting can be reconciled. A malformed CSV
row is also a source-integrity failure rather than a silently usable observation.

Eligible records have room_type exactly "Entire home/apt", bedrooms numerically
one, and accommodates numerically two. Finite Decimal parsing is used: a value
such as 1.0 matches one. Empty, malformed and nonfinite numeric values are invalid;
other finite values outside the target are excluded. Capacity describes the
maximum advertised guests, not observed party size. No filter based on availability
or review activity is applied. The cohort is not described as active or bookable.

Amenity definitions

Only the exact tokens in token-mapping.json count as declared presence. The mapping
was frozen after a pooled token-only schema check and before viewing frequencies,
city associations or combinations. No substring matching or post-result additions
are used.

- Wi-Fi: the enumerated Wifi/Fast wifi service labels, including Pocket wifi and
  the zero-speed label in the primary definition. This measures declared service,
  not verified speed, reliability or working availability.
- Dedicated workspace: the exact category label. It does not certify a quiet room,
  a good chair, ergonomics or a physically inspected setup.
- Kitchen: the exact Kitchen category. Kitchenette, outdoor kitchens, appliances
  and cooking supplies alone do not qualify. "Kitchen" does not certify a fully
  equipped kitchen.
- Washer access: the enumerated washer labels, including free/paid and in-unit/
  in-building variants. Access need not be private, free or inside the unit.
  A dishwasher, laundry service, dryer or nearby laundromat does not qualify.

A blank or missing amenities field, empty array, malformed JSON, non-array value,
non-string element or empty-string element makes all four features unknown for
that record. Such records stay in the eligible count but not the analyzable base.
For a valid nonempty array, a feature not matched means "not recorded in this
captured amenity list", not proven physical absence.

Tables and denominators

All 16 four-bit patterns appear, including zero-count patterns. Bit order is Wi-Fi,
workspace, kitchen, washer. A 1 means the feature is recorded and a 0 means it is
not recorded in an otherwise analyzable list. Unknown lists never become 0000.

Marginal and all-four joint shares use all analyzable eligible listings in that
city/variant as denominator. A leave-one-feature-out conditional share asks:
among listings declaring the other three features, how many also declare this
one? Its numerator is the all-four count and its denominator is shown separately.
All four conditionals and all three unordered city pairs are retained: 12 signed
comparisons per variant, with the first named city minus the second. Percentages
are not rounded before calculations. Exact fractions govern gates; percent and
percentage-point fields are display conveniences.

The primary analyzable counts are 1,273 in Austin, 15,086 in London and 11,257 in
Bangkok. The corresponding unknown counts are 1, 20 and 12 among eligible totals
of 1,274, 15,106 and 11,269. The complete cohort flow is provided, not just these
final denominators.

Fixed checks and sensitivities

The entire table is repeated for property_type exactly "Entire rental unit".
This reduces broad property-type mixing; it does not standardize buildings,
neighbourhoods, hosts, quality or other market differences.

A second full-table sensitivity removes Pocket wifi and Wifi – 0 Mbps from the
Wi-Fi allowlist, unless another qualifying Wi-Fi token is also recorded. Its
analyzable base and the other feature definitions are unchanged. It tests a label
classification choice, not physical connection quality.

Unknown-amenity stress bounds allocate every unknown eligible record to the
extreme compatible state for each metric. For k positives among n analyzable
records and u unknown records, marginal/joint bounds are k/(n+u) to (k+u)/(n+u).
For k all-four records among b declaring the other three, conditional bounds are
k/(b+u) to (k+u)/(b+u). Cross-city lower bounds subtract the second city's upper
bound from the first city's lower bound; upper bounds do the reverse. These are
identification stress bounds, not sampling confidence intervals. Individual
extremes need not be jointly attainable across every displayed metric at once.

A source comparison is not estimable. "Previous scrape" is the only nonempty
source label, and blanks remain unknown. No neighbourhood-search category appears.
Missing labels are not imputed and no alternative date subgroup is substituted.
The post-unblinding clarification is separate from the immutable frozen plan.

Qualification rules

The numerical gate requires all three primary cities to have at least 100
analyzable listings, at most 5% unknown amenities among eligible records, no
conflicting or malformed nonempty IDs, resolved feature mappings, and all 12
primary conditional bases of at least 50.

At least one prespecified conditional city contrast must have an absolute primary
difference of at least 15 percentage points. That same feature/city contrast must
retain its direction and at least 10 points under the unknown-data bounds,
entire-rental-unit table and narrower-Wi-Fi table. Both relevant conditional bases
must reach 50 in each computed sensitivity. The unavailable source comparison is
neither passed nor failed. Thresholds assess editorial usefulness, not statistical
significance; no p-values, survey weights or sampling confidence intervals are
claimed for this nonrandom captured frame.

All counts, patterns, contrasts, composition tables, bounds and gate decisions
were independently reproduced from the same source hashes before publication.
Four of the twelve primary contrasts met the fixed numerical requirements.
A numerical pass alone does not establish a good article: the narrative still
requires separate factual, editorial and accessibility review. All three cities
and all four features remain visible, including small or flat comparisons.
