Cambodia: Duplicate Listings and Housing Supply Counts

A portal counter is useful for describing what the portal is publishing, but it is not a census of physical apartments. In the 29 September 2026 snapshot, the national realestate.com.kh sale filter for condos showed 2,247 cards. That tells us how many records the site placed inside that filter at the time of the check. It does not establish that 2,247 different homes were simultaneously available across Cambodia.

Several ordinary publishing patterns can create extra cards. The same owner may instruct more than one agent, a unit may appear on several websites, or an advert may be reposted after an earlier version has gone stale. A physical apartment can also have separate sale and rental adverts. Those are different records for channel analysis, but only one home if the question is how many physical units exist. The calculation on this page is sale-only, so rental records are deliberately excluded.

There is another category that should be removed before any market-level duplicate analysis: technical copies. Language variants, legacy URLs and alternative views of the same database record can produce multiple web addresses while retaining one stable listing identifier. Counting those as multi-agent duplication would exaggerate the effect. In the audit below, such technical variants are normalized away before the raw-card count begins.

Time Square 5 provides a clean example of genuine cross-platform repetition. IPS Cambodia and Khmer24 both carry Property Code 20993 with the same $250,000 asking price, 40th floor, 90 sqm and three-bedroom configuration. The shared code is the decisive anchor: two pages can be treated as one apartment without relying on a vague resemblance. What that example proves is that duplication exists; it does not justify merging every similar listing in the same building.

The opposite shortcut is equally risky. An advert disappearing from search does not prove a sale, because it may have expired, been removed, replaced or reposted. Headline listing totals therefore should not be converted into transaction counts, housing stock or unique competing supply without another layer of identity checking. The useful number depends on the question being asked and on how confidently individual cards can be tied to physical units.

The numerical result uses a fixed identity-check set assembled on 29 September 2026 at 05:48 Cambodia time (UTC+7). The national realestate.com.kh condo-sale search was recorded separately at 2,247 cards to show the scale of the portal counter, but it is not used to estimate a duplicate percentage. We did not adjudicate every card in that national result. The deeper check is confined to sale listings in BKK1, where the pages expose enough unit-level detail for meaningful matching.

The audit set contains exactly nine cards. Five are sale units visible on the opened IPS Cambodia Time Square 5 project page at the time of the check: Property Codes 16371, 17373, 21126, 16541 and 20485. Two more are the IPS Cambodia and Khmer24 pages for Time Square 5 Property Code 20993. The final pair is from J Tower 2: realestate.com.kh listing ID 254426 and Khmer24 advert 12911091. Both describe a two-bedroom, 91 sqm apartment on the 20th floor and carry a notably similar package of fittings and furnishings, while the accessible versions show different asking prices — $250,000 and $240,000.

The 20993 pair is a hard match. A shared property code plus matching floor, size, bedroom count and price is strong enough to collapse two web pages into one physical unit. Starting from nine cards, that confirmed cluster contributes one excess card, leaving eight units after confirmed merges. The confirmed excess is therefore about 11.1% of this deliberately selected nine-card set. It is arithmetic for this audit only, not a duplicate-rate estimate for BKK1, Phnom Penh or Cambodia.

The J Tower 2 pair is different. Its descriptive overlap is strong, including unusual furnishing details, and a different asking price does not prove that the homes are different: multiple agents can market one physical unit on different terms. Yet the accessible pages do not expose a common unit number or stable cross-platform property code. That makes the match probable rather than proven. If those two adverts refer to one apartment, the set contains seven distinct units; if they are separate apartments, it contains eight. The defensible result is consequently a 7–8 range rather than a tidier but less supportable single number.

This was designed as a dense identity test, not a random market sample. It is useful because it contains both an exact cross-platform match and a difficult near-match, which lets the reader see where certainty ends. It does not independently verify that every advertised unit was still physically available from the owner at the capture time. Technical language variants or legacy URLs sharing the same stable record identifier are removed before the nine-card starting count, so they cannot inflate the market-duplicate result.

These figures apply only to the fixed nine-card BKK1 sale set checked on 29 September 2026. The 2,247-card national portal count is not deduplicated here.

The strongest match is one that carries a unit-specific identifier across sources. Time Square 5 Property Code 20993 appears on both IPS Cambodia and Khmer24 with the same floor, area, bedroom count and asking price. The shared code removes most of the ambiguity: this is not a judgment based on a similar sofa or a copied headline. For the calculation, those two pages are treated as one confirmed physical apartment.

A genuine unit number can play the same role, provided it identifies an individual home, not a floor-plan type. Below that level, a rare combination of attributes may still be persuasive: building, exact floor, precise area, orientation or view, distinctive fit-out and photographs that cannot reasonably be explained by shared project marketing. The more standardized the development, the less weight should be placed on any single attribute.

The J Tower 2 pair sits in the middle. Both adverts point to a two-bedroom unit on the 20th floor at 91 sqm, and their descriptions overlap on several unusual furnishing details. The accessible versions carry different asks — $250,000 on realestate.com.kh and $240,000 on Khmer24 — which does not rule out a duplicate because agents can market the same home on different terms. That is still far stronger evidence than matching on area alone, so the pair is classified as probable. Yet the available pages do not expose the same unit number or a common stable property code, which is why the count is not reduced automatically.

Images are particularly easy to overrate. Developer renders, show-unit photography, common-area shots and standard room images can appear across many legitimate listings. Similar wording is also only supporting evidence because property descriptions are routinely copied or syndicated. The decisive question is whether the evidence identifies one physical apartment, not merely the same project or unit type.

This threshold prevents two opposite distortions. Merging on weak evidence can erase a real competing home; refusing to merge an exact ID match lets one apartment influence the analysis twice. A useful cleaned set therefore keeps confirmed matches separate from probable or unresolved ones instead of pretending that every card can be classified with equal certainty.

J Tower 2 is a useful counterexample because similarity is common inside a standardized tower. A recent indexed version of the realestate.com.kh project inventory showed 19 sale cards, many of them two-bedroom apartments of 65 sqm or 91 sqm. Repeated layouts up the building, similar fit-outs and recycled project descriptions are normal. A matching project name and floor area therefore cannot identify a single physical unit.

The visible inventory also shows why attributes need to be combined, not used one at a time. The 65 sqm format appears across floors 10, 12, 15, 23, 26 and 29, while 91 sqm units appear on numerous other levels. Area alone is plainly not a unique key. If two cards also share a floor and price, the duplicate hypothesis becomes stronger, but without a unit number or another hard anchor it still should not be promoted automatically to a confirmed match.

Stock imagery creates the same problem. A kitchen render, pool photograph, lobby shot or standard bedroom image identifies the development far better than it identifies a specific front door. Even photographs from real apartments can repeat when multiple units were furnished to the same package. Images become meaningful identity evidence only when distinctive features line up with other unit-level facts.

Small area differences deserve caution in both directions. One source may quote net internal space while another uses a gross figure; rounding or an old data entry can also create a mismatch. Yet near-identical areas are not proof of sameness either. Size needs to be read alongside floor, orientation, layout, identifiers and photographs, not used as an automatic merge rule.

A clean-looking database is not the goal if it is achieved by deleting uncertainty. In this audit, the J Tower 2 pair remains probable because the evidence stops short of a unique shared identifier. Keeping a 7–8 range is more informative than forcing the result to seven simply because a single number would look neater.

The first practical effect is on price analysis. If one apartment appears two or three times and every card is treated as an independent comparable, that unit receives extra weight in a median, a range or a comp set. When all duplicates carry the same price, the weighting is distorted; when agents advertise the same home at different prices, the apparent spread can widen as well. Establishing how many physical units sit behind the cards should come before drawing conclusions from their asking prices.

The second effect is building-level competition. A buyer looking at a tower with dozens of adverts may assume there are dozens of distinct sellers competing for attention. Removing hard duplicates can narrow that picture, although even a cleaned card count is not a live inventory guarantee. An advert may remain online after its status or terms have changed. The cleaned number is a better description of visible competing units, but it still belongs to a particular capture date.

The third issue is market scale. Nine carefully checked cards do not become national statistics merely because the identity work is rigorous. This set is intentionally concentrated in BKK1 and selected for pages that expose enough detail to test whether two records point to the same home. Its roughly 11.1% confirmed excess cannot be applied to the 2,247-card national portal result and cannot support a statement about Cambodia’s true inventory. A market-wide estimate would require a much broader, predefined sampling design and consistent adjudication across the covered population.

The 7–8 range is therefore more useful than a falsely exact figure. Eight is the count after merging only the pair supported by a hard shared identifier. Seven is the lower bound if the strongly similar J Tower 2 pair ultimately proves to be one unit. The gap is not model error; it is a visible record of what the public pages cannot resolve with confidence.

The page’s conclusion is deliberately narrow. Portal cards and physical apartments are different units of measurement, and good cleaning should preserve uncertain cases instead of hiding them. For a specific building or comparable set, count confirmed physical units, keep probable matches separate, and resist turning a small identity audit into a market-wide supply statistic.