The Art and Craft of Fraudulent App Promotion in Google Play

May combine two claims. The quote only partly states this claim.

The paper found rapid review bursts in both qualitative and quantitative investigations, consistent with assumptions in prior fraud-detection work, and interview participants revealed adaptive rate control as an avoidance technique.

“We found ASO workers who post fake reviews in rapid bursts in both our qualitative and quantitative investigations.This is consistent with assumptions made in previous fraud detection work, e.g., [26, 28, 32, 36-38, 40, 42, 43, 51, 52, 59-61, 63, 65, 92, 95, 96, 100].”

• cites [26] sourceSupported
• cites [28]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [32] sourceNot checked — no full text of the source
• cites [36] sourceNot checked — no full text of the source
• cites [37] sourceNot checked — no full text of the source
• cites [38]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [40] sourceNot checked — no full text of the source
• cites [42]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [43]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [51] sourceNot checked — no full text of the source

May combine two claims.

Several participants admitted to reusing common linguistic patterns and copying reviews across similar products, and the quantitative study confirmed this.

“several participants admitted to reuse common linguistic patterns and copy reviews across similar products.”

• Also stated in: § 5.10 Review Writing · boosted

  “We also confirmed this finding in our quantitative study.”

• cites [38]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [42]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [51] sourceNot checked — no full text of the source
• cites [52]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [60]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [61]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [63]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [64]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [65]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [74] sourcePartly supported

  In A. Data Collection (page unavailable), the paper reports observing duplicate reviews with similar content for one or two stores, which supports the review-reuse aspect. It does not report participants admitting to that behavior or to reusing linguistic patterns, nor does the cited passage establish the claimed quantitative confirmation. (LLM, conf=0.84)

  Experimental — accuracy not yet measured

• AssumptionRests on an assumption: The participants' self-reported strategies and behaviors are accurate.

May combine two claims.

Most participants control a few hundred accounts, but a few control or have access to several thousands; one participant (P13) claimed to be part of a team of 13 workers controlling 80,000 accounts.

“Most participants control a few hundred accounts, however, a few control or have access to several thousands: P13 claimed to be part of a team of 13 workers who control 80,000 accounts.”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

May combine two claims.

In the quantitative dataset, one participant (F18) revealed 83 accounts, and another (F35) reviewed 927 unique apps from 42 accounts.

“one participant (F18) was able to reveal 83 accounts that he controls, and F35 has reviewed 927 unique apps from his 42 accounts.”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.
• AssumptionRests on an assumption: The 1,164 accounts and 21,767 reviews collected from ASO workers are actually fraudulent.

May combine two claims.

Two participants said they also post from emulators running on laptops; one participant (P13) claimed to have 13 laptops using BlueStacks and 13 smartphones.

“2 participants said that they also post from emulators running in laptops, e.g., P13 claims to have 13 laptops and use the BlueStacks emulator [5] to install and review apps, and also 13 smartphones.”

• cites [5]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

One participant (P7) claimed to have 3,000 accounts and to run CPI/CPA campaigns to incentivize people to install apps.

“P7 did not specify the number of organic accounts that they can access, but stated”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

Probably not a substantial claim: may not pass S3 (load-bearing).

Most participants claimed to purchase their devices on the regular market.

“Most participants claimed to purchase their devices on the regular market.”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

Probably not a substantial claim: may not pass S3 (load-bearing).

One participant (P18) claimed to run a mobile repair shop and use devices he receives for repair to write reviews.

“P18 said (1)”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

Probably not a substantial claim: may not pass S3 (load-bearing).

One participant (P4) claimed to be part of a company with around 150 people who organize 15,000 organic ASO workers through virtual groups.

“P4 claimed to be part of a big company with around 150 people in their team, who organize 15,000 organic ASO workers through virtual (WhatsApp, Facebook) groups.”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

Probably not a substantial claim: may not pass S3 (load-bearing). The quote only partly states this claim.

One participant (P5) claimed to have a Facebook group combining 60 real users to write reviews.

“P5 said”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

May combine two claims.

Five of the 18 interview participants claimed to work alone, while the other 13 claimed to have a team with at least 10 members.

“5 participants claimed to work alone.The other 13 participants claimed to have a team with at least 10 members.”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

May combine two claims.

The quantitative dataset contains 1,164 Google Play accounts controlled by 39 ASO workers and 21,767 fake reviews posted from those accounts for 6,362 unique apps.

“The data includes 1,164 Google Play accounts that the 39 ASO workers revealed to control, and 21,767 fake reviews posted from these accounts for 6,362 unique apps.”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.
• AssumptionRests on an assumption: The 1,164 accounts and 21,767 reviews collected from ASO workers are actually fraudulent.

May combine two claims.

Some participants have an almost 1-to-1 account-to-device mapping (P8, P18), others have up to 7 accounts per device (P2, P3, P7, P10, P15), and others have significantly more accounts than devices (P1: 15,000:300; P13: 80,000:30).

“P8 and P18 have an almost 1-to-1 account-to-device mapping.Participants such as P2, P3, P7, P10 and P15, have a small but manyto-one mapping, e.g., up to 7 accounts per device.Others, such as P1 and P13, claim to have significantly more accounts than devices (e.g., 15,000:300 and 80,000:30 respectively).”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

May combine two claims.

Participants claimed to have worked on between 150 and 4,000 apps in total, and between 6 and 50-60 apps in the past month, with a mean of 34.11 and a standard deviation of 18.37 for the past month.

“they claimed to have worked on between 150 and 4,000 apps in total, and between 6 and 50-60 apps in the past month (M = 34.11,SD = 18.37).”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

Probably not a substantial claim: may not pass S3 (load-bearing).

Four participants (P1, P10, P11, P13) claimed to own only low-end, cheap devices.

“4 participants (P1, P10, P11, P13) claimed to own only lowend, cheap devices.”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

Probably not a substantial claim: may not pass S3 (load-bearing).

Other participants (P7, P15, P16) claimed to own a mix of low, medium and high-end devices, dominated by low-end devices.

“Others (P7, P15, P16) claimed to own a mix of low, medium and high-end devices, dominated by low-end devices.”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

The quote only partly states this claim.

Nine of the 18 interview participants claimed to organize or be part of online teams of organic users who use personal accounts to post fake reviews.

“9 participants claimed to organize or be part of online teams of”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

May combine two claims.

Nine participants claimed to post fraud from mobile devices, and 11 claimed this also happens from mobile devices of organic ASO workers they control.

“9 participants claimed to post fraud from mobile devices; 11 participants claimed this also happens from the mobile devices of organic ASO workers that they control.”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

Probably not a substantial claim: may not pass S3 (load-bearing).

All interview participants claimed to own or have access to multiple mobile devices.

“All the interview participants claimed to own or have access to multiple mobile devices.”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

May combine two claims.

Fifteen of the 18 interview participants claimed a hierarchical structure in their organizations; 11 described specific roles including job managers, team admins, and new account creators.

“15 participants claimed a hierarchical structure of their organizations (Figure 2).11 of them described specific roles in their organizations, that include job managers, who interface with the developers and manage work from the marketplace, team admins, who organize, distribute tasks, and verify the work of review posters, and new account creators.”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

May combine two claims.

Many ASO workers work in unstructured teams and share their earnings.

“many of them work in a more unstructured team and the whole team share their earnings.”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

May combine two claims.

Profit-sharing arrangements among ASO teams include monthly salaries, even splits, preferential cuts for job managers (10-25%) and team leads (10-50%), and flat rates of $0.40 per review.

“One participant claimed to pay team members a monthly salary, while another one claimed an even split among members.Three of them mentioned preferential cuts for the job manager (10-25%) and team lead (10-50%) and equal split of the rest among the actual review posters.Two participants claimed a flat rate for the review posters ($0.40 per review).”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

Probably not a substantial claim: may not pass S3 (load-bearing).

Two participants claimed to work in teams with a flat organization.

“However, 2 participants claimed to work in teams with a flat organization.”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

May combine two claims. The quote only partly states this claim.

Ten of the eighteen interview participants used Facebook and/or WhatsApp to form online teams and used the corresponding messenger apps for team communication.

“10 interview participants (P3, P5, P6, P8, P9, P11, P12, P14, P17, P18) said that they used Facebook and/or Whatsapp to create online teams.”

Probably not a substantial claim: may not pass S3 (load-bearing).

One participant (P12) skips jobs when app-device compatibility is an issue.

“P12 skips the job.”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Two participants (P9, P15) said they provide as many reviews as possible from compatible devices and contact the developer.

“P9 and P15 said that they provide as many reviews as they can from their compatible devices, and contact the developer to explain the problem.”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

May combine two claims. The quote only partly states this claim.

Fourteen participants worked on recently launched apps, with declared numbers ranging from 1-2 jobs in the past month to 20-40 jobs.

“14 participants said that they have worked on recently launched apps, and either the hiring developer mentions that the app was recently launched, or that they infer this information based on the app status when posting their first review.”

May combine two claims.

Most of these reviews were written from old low-end devices (45.98%), old mid-end (31.41%), and fresh low-end (15.31%) devices.

“Further, most of these reviews were written from old low-end devices (45.98%), old mid-end (31.41%) and fresh low-end (15.31%) devices.”

• AssumptionRests on an assumption: The 1,164 accounts and 21,767 reviews collected from ASO workers are actually fraudulent.
• AssumptionRests on an assumption: The device model inference via the Google Play bug accurately reflects the devices used to post the reviews.

May combine two claims. The quote only partly states this claim.

Participants used mixed accounts-per-device strategies: some logged in to one account at a time, some logged in to multiple accounts (2-5) per device, and one had a 1-to-1 mapping of 30 devices to 30 accounts.

“P10, P11, P13, P18 said that they only log in to one account at a time, on any device that they control.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims. The quote only partly states this claim.

Fourteen participants waited, opened, or used the app before reviewing it, and all participants admitted to performing retention installs.

“14 participants claimed to wait, open, or even use the app before reviewing it.”

May combine two claims.

In Figure 6, F3, F7, F11, and F31 use old devices; most others (F1, F2, F13, F20, etc.) use both newly released and old devices.

“F3, F7, F11, and F31 use old devices.Most others (F1, F2, F13, F20, etc), use both newly released and old devices.”

• AssumptionRests on an assumption: The 1,164 accounts and 21,767 reviews collected from ASO workers are actually fraudulent.
• AssumptionRests on an assumption: The device model inference via the Google Play bug accurately reflects the devices used to post the reviews.

May combine two claims. The quote only partly states this claim.

Four participants said an app can be reviewed without prior installation from a device where the account is logged in, and the paper tested this and verified it works.

“We have tested this claim and verified that it works as suggested.”

May combine two claims. The quote only partly states this claim.

Sixteen interview participants saw jobs specifying daily review counts, most suggested 2-3 reviews per day, some recommended up to 30-40 per day, and several said the recommended rate depends on the app's existing review count.

“16 interview participants claimed to have seen jobs (1-45 in the past month) that specify how many reviews per day the workers should post.”

May combine two claims. The quote only partly states this claim.

Some participants used VPNs while others explicitly did not, and one used VPN or proxy only when the job specification required it, such as a USA proxy to install from the USA.

“P1, P3, P5, P13, P15 admitted to use VPNs, while the other 10 explicitly claimed to not use them.”

May combine two claims. The quote only partly states this claim.

Some interview participants selected devices and accounts sequentially in a round-robin manner, while seven of eighteen claimed random choice of accounts and devices, and one chose accounts based on filter-avoidance success rate.

“P5, P7, P10, P13, P18 select the devices in a sequential, round-robin manner, while P5, P7, P13, P15, P16, P18, select the accounts sequentially.”

May combine two claims.

Most participants kept the app for a few days after reviewing it, with durations ranging from 1 day to 2 months.

“Most participants said that they keep the app for a few days after reviewing it: 1 day (P1, P5 and P15), 2-3 days (P4, P5, P8, P10, P13, P14), 1-2 weeks (P17), and 7 days -2 months (P2).”

May combine two claims.

Fourteen participants exhibited less pronounced lockstep behaviors, and one worker used more than 50% of his 17 accounts in only two of 121 apps.

“However, 14 participants exhibit less pronounced lockstep behaviors, e.g., F36 (Figure 10 right).Out of 121 apps reviewed, in only two apps, F36 used more than 50% of the 17 accounts he revealed.”

May combine two claims.

All interview participants were aware that their fake reviews could be detected and deleted, and all reported that Google had deleted some of their reviews.

“All interview participants are aware of their fake reviews being detected and deleted.All of them have reported that Google deleted some of their reviews.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Two participants preferred email to validate user accounts, one said Google may force phone-number validation, only one claimed virtual phone numbers were accepted, and all others said they used real phone numbers.

“P2 and P3 said that they prefer to use e-mail to validate user accounts.P3 also said that Google may force them to use phone numbers.Only P16 claimed that”

May combine two claims.

Several participants reported limits on phone-number reuse: one number could be used for 3-5 accounts but not immediately, and at least 3 months had to pass between two verifications using the same number.

“Several participants reported limitations on phone number reuse.For instance, P3 and P8 said that one number could be used for 3-5 accounts but not immediately, while P1 said that”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

One participant (P7) claimed to have a dedicated team to manage devices, charging a device every 2-3 days, keeping devices on during office time and switching them off after 11pm-midnight.

“He claimed that they have a dedicated team to manage all the devices, and charge a device every 2-3 days.Further, he claimed that they keep the devices on during office time, and switch them off after 11pm-midnight.”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

May combine two claims. The quote only partly states this claim.

Six participants claimed to store devices on a table, easily accessible; one (P1) stored them in a separate room; one (P7) kept high-end phones in a locker after use.

“6 participants claimed to store the devices on a table, easily accessible.P1 claimed to store the devices in a separate room.P7 said that”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

May combine two claims. The quote only partly states this claim.

All participants except one saw ASO jobs requiring a campaign duration, with twelve reporting longest campaigns of 1-6 months and six reporting 7-18 months.

“All the participants except P4 said that they have seen ASO jobs that require a duration for the review posting campaign.”

May combine two claims. The quote only partly states this claim.

Twelve participants saw jobs requiring only old accounts, some participants worked on jobs using only fresh accounts, and some regularly used a mix of old and new accounts.

“12 participants claimed to have seen jobs that required only the use of old accounts.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims. The quote only partly states this claim.

Some participants avoided reviewing other apps to evade detection, while online team members of other participants did review other apps they normally used, and one participant used real users and instructed them to use other popular apps for authenticity.

“P2, P3, P5, P7, P10, P13, P15 and P16 said that they do not review other apps to avoid detection.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims. The quote only partly states this claim.

Interview participants attributed review deletions to device re-use, improper VPN use, improper app use, extended account use, and misfires of Google's fraud detection.

“Participants reported diverse reasons for deletion:”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims. The quote only partly states this claim.

One participant reported that bots can post reviews, ratings, and installs when the client permits, while all other participants reported writing reviews manually without scripts.

“All of the other participants said that they write their reviews manually, and do not use any script for this purpose.”

The quote only partly states this claim.

Two participants (F25, F32) used Plan Ceibal tablets (price 0 EUR) to write 159 reviews for 137 apps.

“Participants F25 and F32 used this device model to write 159 reviews for 137 apps.”

• AssumptionRests on an assumption: The 1,164 accounts and 21,767 reviews collected from ASO workers are actually fraudulent.
• AssumptionRests on an assumption: The device model inference via the Google Play bug accurately reflects the devices used to post the reviews.

May combine two claims.

All eighteen participants claimed to have been re-hired for apps they previously promoted, with total times mean 186.1, SD 190.7, min 15, max 600.

“All 18 participants claimed to have been re-hired for apps that they previously promoted (total times M = 186.1,SD = 190.7,Min = 15, Max = 600).”

May combine two claims.

Most interview participants reported that deleted reviews were under 5% of all reviews they posted, four reported 10-20% deletion, and one participant said deletion ranged from 2% to 30% depending on the app.

“Although most of them have reported deletion as a small or negligible percentage (under 5%) of all the reviews they posted, four of our interview participants have said that 10-20% of their reviews were deleted.P6 said that the review deletion percentage depends on the app and ranges from 2% to 30%.”

The quote only partly states this claim.

One ASO worker showed nearly perfect lockstep behavior, using the same set of 15 accounts for almost all 40 apps in the same order, with some variation such as reverse order for one app.

“ASO worker F7 (left) shows a nearly perfect lockstep behavior with the same set of 15 accounts used for almost all the 40 apps, and in the same order.”

May combine two claims.

Some ASO workers were active for more than 1 year for an app, with some active up to more than 4 years.

“We observe workers who have posted reviews for certain apps, for more than 1 year, and up to more than 4 years.”

Probably not a substantial claim: may not pass S3 (load-bearing).

One participant (P11) claimed to have bought his 45 devices from the black market at a very low price.

“P11 said, about his claimed 45 devices, that”

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

May combine two claims.

Two participants said their teams used virtual devices running on laptops, while the others claimed to use mobile devices or real users with mobile devices.

“P10 and P13 said that their teams use virtual devices running in laptops.The others claimed to use mobile devices or have access to real users equipped with mobile devices.”

May combine two claims.

Participants with many devices described streamlined management, while those with fewer described workarounds for cost and compatibility, such as outsourcing.

“Participants with many devices reported streamlined solutions to manage them, while those with fewer devices reported ways around cost limitations and compatibility issues, e.g., further outsourcing jobs.”

May combine two claims.

Twelve participants posted reviews from at least 20 different device models; one (F35) used at least 84 distinct device models.

“We found that 12 participants posted reviews from at least 20 different device models; F35 used at least 84 distinct device models.”

• AssumptionRests on an assumption: The 1,164 accounts and 21,767 reviews collected from ASO workers are actually fraudulent.
• AssumptionRests on an assumption: The device model inference via the Google Play bug accurately reflects the devices used to post the reviews.

May combine two claims.

Six of eighteen interview participants claimed lockstep-indicative behaviors, and twenty-five of thirty-nine quantitative participants exhibited lockstep behaviors, some using accounts in the same order across multiple apps.

“6 out of 18 interview participants claimed lockstepindicative behaviors; 25 of the 39 quantitative study participants exhibit lockstep behaviors, some even using their accounts in the same order to review multiple apps.”

May combine two claims.

The unknown device category includes reviews posted from Google Play's website interface and certain types of emulators.

“We confirmed that the”

• AssumptionRests on an assumption: The device model inference via the Google Play bug accurately reflects the devices used to post the reviews.

May combine two claims. The quote only partly states this claim.

Participants bypassed SIM-card limitations by using friends and family phone numbers, buying many SIM cards, borrowing SIM cards, using phone-number verification services, and purchasing already validated accounts.

“P3, P10, P11 and P17 use friends and family: P3 said that”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims. The quote only partly states this claim.

One participant described flushing a virtual device and changing its MAC address before login, using the account-device pair for a few days, then leaving the account unused for 1-1.5 months because Google does not check the new login from the same MAC after that interval; another reset the device with cccleaner after 3 days.

“P10 said that before logging in to an account, they flush the virtual device and change its MAC address.”

May combine two claims. The quote only partly states this claim.

ASO workers claim access to many devices, either owned or obtained through organic-fraud communities, and this claim is partly confirmed by the gold standard fraud data.

“We found ASO workers who claim to have access to large number of devices, either owned, or accessed through their communities of organic fraud.”

May combine two claims.

Of the 9,942 presumed fake reviews from known device models, 61.3% were posted from low-end, 38.2% from mid-end, and 0.5% from high-end devices; 77.39% were from old models and 19.66% from new models.

“We found that 61.3% of reviews were posted from low-end, 38.2% from mid-end, and 0.5% from high-end devices, while 77.39% are from old and 19.66% from new models.”

• AssumptionRests on an assumption: The 1,164 accounts and 21,767 reviews collected from ASO workers are actually fraudulent.
• AssumptionRests on an assumption: The device model inference via the Google Play bug accurately reflects the devices used to post the reviews.

Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing), S4 (based). May combine two claims. The quote only partly states this claim.

Six participants upvoted reviews written by their team from other accounts, one participant downvoted negative reviews to trigger Google's filtering and remove them, and one provided upvote and downvote services to move positive reviews up and negative down.

“6 of the 18 participants (P5, P7, P10, P14, P15, P16) said that they upvote reviews written by their team from other accounts.”

May combine two claims.

The paper also reports claims from eight of eighteen participants and evidence from fourteen of thirty-nine participants of random account and device choice.

“However, we also report claims (8 of 18 participants) and evidence (14 of 39 participants) of random account and device choice.”

May combine two claims.

Over 585 apps with at least 10 fraudulent reviews, the distribution of review timing was left-skewed, with 50% of reviews posted less than 3 months after app launch, and the median and third quartile of first reviews from the 39 participants' accounts were 113 and 344 days after release.

“The distribution is left-skewed, with 50% of the reviews being posted after less than 3 months after app launch.However, we observe cases where the first reviews from any of the accounts of our 39 participants, are posted long after the app was released: the median and 3rd quartile are 113 and 344 days respectively.”

May combine two claims.

One participant said Google keeps only one review per device if multiple reviews come from one device, and the paper tested and verified that reviewing an app from different accounts on the same device is possible.

“We have tested this claim and verified that it works as suggested.”

May combine two claims. The quote only partly states this claim.

Quantitative analysis found some ASO workers who post many reviews per day in bursts, others who post fewer reviews but still in bursts, and some who post few reviews spaced through the day with mean inter-review times of 8-9 hours.

“We observe several participants, e.g., F7, F9, F10, F16, F27, F28, F31, who tend to post more reviews per day, an do so in bursts.”

May combine two claims.

Five interview participants reported accessing their accounts regularly, twelve reported doing so manually, and three used scripts or automatic login systems to keep accounts alive and detect inaccessible ones.

“5 interview participants (P1, P11, P13, P15, P16) said they access their accounts regularly.12 participants said that they access them manually.However, 3 participants (P13, P15, P16) said they use scripts and automatic login systems to”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims. The quote only partly states this claim.

Some participants worked on jobs requiring accounts created to post one review and then abandoned, with costs of $8 and $10, and participants said this was because Google does not filter singleton-account reviews due to its fraud detector needing more information to build account reputation.

“P1, P2, P7, P10, P13, and P15 said they worked on jobs where they had to create accounts just to post one review and then to abandon them.”

May combine two claims.

Twelve participants said they use screenshots of their reviews as proof of work, and five said they send usernames of accounts used to post reviews.

“12 participants said that they use screenshots of their reviews.”

• Also stated in: § 5.12 Proof of Work · plain

  “5 participants said that they send the usernames of accounts that they used to post reviews.”

• AssumptionRests on an assumption: The participants' self-reported strategies and behaviors are accurate.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

P10 said they use no profile picture because pictures define demographics and buyers do not want this now.

“P10 however said that”

• AssumptionRests on an assumption: The participants' self-reported strategies and behaviors are accurate.

May combine two claims. The quote only partly states this claim.

Eight participants denied plagiarism and self-plagiarism, while six admitted to some form of plagiarism, with reasons including copying only when buyers mention the source, short similar reviews, using a review dataset across apps, and modifying reviews from similar apps.

“8 participants (P1, P3, P5, P12, P13, P14, P15, P18) denied plagiarism and self-plagiarism.P2, P4, P6, P9, P11 and P17 however admitted to some form of plagiarism.”

May combine two claims.

36 of 47 accounts controlled by F34 are closed, and 29 of 35 accounts of F31 are closed.

“Notably, 36 out of the 47 accounts controlled by F34 are closed, as are 29 out of 35 accounts of F31.”

• AssumptionRests on an assumption: The accounts that the 39 workers claimed to control were in fact controlled by those workers.

May combine two claims.

The co-review graphs of 16 workers have an average weight of at least 10, up to 78.61 for F13.

“Further, the co-review graphs of 16 workers have an average weight of at least 10, up to 78.61 for F13.”

• AssumptionRests on an assumption: High co-review overlap between accounts is a valid indicator that those accounts share the same controller.

May combine two claims.

Most interviewed participants said that developers provide the text of reviews, but participants can also write their own reviews.

“Most interviewed participants said that the text of the reviews is provided by the developers, but also they can write their own reviews.”

• AssumptionRests on an assumption: The participants' self-reported strategies and behaviors are accurate.

May combine two claims.

The mean number of jobs assigned was 3.48, seven workers were hired by the same developer more than 10 times, and one developer hired 6 workers to each promote at least 10 apps.

“We observe that the mean number of jobs assigned is 3.48, and 7 workers have been hired by the same developer more than 10 times.We found one developer that hired 6 workers to each promote at least 10 apps.”

May combine two claims.

In the same data, 1.67% of reviews were 3-star and 2.81% were 1- or 2-star.

“Only 1.67% were 3-star and 2.81% were 1 or 2 star reviews.”

• AssumptionRests on an assumption: The accounts that the 39 workers claimed to control were in fact controlled by those workers.

May combine two claims.

The repeated reviews are short, generic, and app-agnostic.

“We note that these reviews are short, generic, and app-agnostic.”

May combine two claims.

Three participants said they do not maintain any specific rating ratio, while P4 and P12 post only 5-star reviews.

“3 participants (P6, P11, P14) said that they do not maintain any specific ratio, while P4 and P12 post only 5-star reviews.”

• AssumptionRests on an assumption: The participants' self-reported strategies and behaviors are accurate.

May combine two claims.

Fourteen accounts revealed by F13 form a clique, and on average any two accounts reviewed 78 apps in common.

“14 accounts revealed by F13 form a clique, and on average, any two accounts reviewed 78 apps in common.”

• AssumptionRests on an assumption: The accounts that the 39 workers claimed to control were in fact controlled by those workers.

May combine two claims.

Some participants admitted purchasing new accounts; P1 claimed more than 10,000 and P13 claimed 47,000.

“5 participants (P1, P5, P13, P17) admitted to purchase new accounts.”

• Also stated in: § 5.13 Account Creation · plain

  “P1 claimed to have purchased more than 10,000 accounts, while P13 claimed to have purchased 47,000 accounts.”

• AssumptionRests on an assumption: The participants' self-reported strategies and behaviors are accurate.

May combine two claims.

Two participants always wrote their own reviews, while the other participants both received or requested review text from the developer and wrote their own reviews.

“2 participants (P3, P4) said that they always write their own reviews.The other participants said that they both receive or request the review text from the developer, and they also write their own reviews.”

May combine two claims.

Manual investigation of accounts revealed by the 39 workers found multiple instances of repeated profile photos, mostly of glamorous people, and simple patterns in account names.

“we manually investigated the accounts revealed by the 39 workers, and found multiple instances of repeated profile photos, mostly of glamorous people, and simple patterns in the account names.”

May combine two claims.

Most participants claimed to write short reviews, and this is reflected in the gold standard fraud data.

“Further, most participants claimed to write short reviews, which is also reflected in our gold standard fraud data.”

• AssumptionRests on an assumption: The participants' self-reported strategies and behaviors are accurate.

May combine two claims. The quote only partly states this claim.

Eleven of eighteen participants typed reviews directly on devices, while others used a mix of typing and cut-and-pasting, and some organized remote teams and did not know their workers' typing actions.

“11 of the 18 participants said that they type the reviews directly from their devices.”

May combine two claims.

Some participants create new accounts when they lack enough accounts for a job, when Google deletes their accounts, or when a job requests more reviews than they can provide.

“P2 and P9 claimed to create new accounts when they don't have enough accounts for a job, especially when the job requests accounts from a specific geographic region.”

• Also stated in: § 5.13 Account Creation · plain

  “P5 and P18 create new accounts when Google deletes some of their accounts.”

• Also stated in: § 5.13 Account Creation · plain

  “P15 create new accounts when the job requests more reviews than they can provide.”

• AssumptionRests on an assumption: The participants' self-reported strategies and behaviors are accurate.

Probably not a substantial claim: may not pass S3 (load-bearing).

P4, P7, and P14 said that using organic ASO workers ensures they use real user names.

“P4, P7 and P14 said that their use of organic ASO workers, ensures that they use real user names.”

• AssumptionRests on an assumption: The participants' self-reported strategies and behaviors are accurate.

May combine two claims. The quote only partly states this claim.

Eleven participants claimed their reviews exceeded 10 words (10-40), two admitted their reviews were short (3-5 words), and one motivated short reviews because Google may match patterns and common short messages are easy and less likely to be complained about.

“11 participant claimed that their reviews exceed 10 words (10-40).P3 and P4 admitted that their reviews are short (3-5 words).”

May combine two claims.

The study identified profile photo plagiarism behaviors and a claimed developer-driven trend to avoid profile photos.

“We identified profile photo plagiarism behaviors, but also a claimed developer-driven trend to avoid profile photos.”

May combine two claims.

The two workers with the most distant review-length CDFs wrote 542 and 771 reviews, with P(length <= 10 | F7) = 0.88 and P(length <= 10 | F26) = 0.06, and the overall fake-review CDF was closer to F7 with P(length <= 10) = 0.63.

“P(Len th ≤ 10|F 7) = 0.88 P(Len th ≤ 10|F 26) = 0.06.The overall fake review word count CDF is closer to F7, with the overall P(Len th ≤ 10) = 0.63.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Work verifications can take place at the team level.

“Work verifications can take place at the team level.”

• AssumptionRests on an assumption: The participants' self-reported strategies and behaviors are accurate.

Probably not a substantial claim: may not pass S3 (load-bearing).

One participant argued that overly long reviews look like paid reviews because real users lack time to post a paragraph.

“P6 argued that”

May combine two claims.

The median and third quartile of time between app launch and first reviews from the 39 participants' accounts are 113 and 344 days, respectively.

“the median and 3rd quartile are 113 and 344 days respectively.”

May combine two claims.

The distribution of time between app launch and fraudulent reviews is left-skewed, with 50% of reviews posted less than 3 months after launch.

“The distribution is left-skewed, with 50% of the reviews being posted after less than 3 months after app launch.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Of 560 contacted workers, 72 replied, 25 qualified, and 18 participated.

“out of the 560 contacted workers, 72 replied to our invitation, 25 qualified, and 18 agreed to finally participate.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Crowdsourcing sites such as Upwork, Fiverr, Zeerk, and Peopleperhour are reliable sources of ASO activities.

“We have used crowdsourcing sites such as Upwork, Fiverr, Zeerk, and Peopleperhour for years, and have found them to be reliable sources of ASO activities.”

Classified as probably not a substantial claim. May combine two claims.

Most ASO workers suggest they use apps before reviewing them and keep them installed after review for a while to mimic genuine behaviors.

“Most ASO workers suggest that they use apps before reviewing them, and keep them installed after review for a while, to mimic genuine behaviors.”

• AssumptionRests on an assumption: The participants' self-reported strategies and behaviors are accurate.

May combine two claims.

The qualitative and quantitative studies provide evidence that developers rehire ASO workers not only for the same app but also for other apps they develop.

“Our qualitative and quantitative studies ( § 5.9) provide evidence that developers rehire ASO workers not only for the same app, but also for other apps that they develop.”

May combine two claims.

Participants are part of teams that are physically co-located or online, hierarchical or flat, and sockpuppet-account-based or organic.

“Participants are part of teams that are either (1) physically co-located or online, (2) hierarchical or flat, and (3) sockpuppet account based or organic.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

11,934 unique device models were used to post over 198 million reviews in Google Play.

“11,934 unique device models were used to post over 198 million reviews in Google Play.”

May combine two claims.

Most workers have mixed impact, with many targeted apps seeing up to a 5-star increase and a few seeing up to a 2-star drop.

“Most workers however have mixed impact, with many of their targeted apps seeing up to 5 star increase in average rating during their active interval, and a few others seeing up to a 2 star drop.”

May combine two claims.

F9, F10, F14, F16, F17, and F27 have only unknown devices; F35 used at least 84 distinct device models.

“F9, F10, F14, F16, F17, and F27 have only unknown devices; F35 used at least 84 distinct device models.”

May combine two claims.

The paper presents results from the first structured interview study of 18 ASO workers recruited from 5 sites, plus a quantitative investigation with 39 other ASO workers from the same sites.

“In this paper we present results from the first structured interview study of 18 ASO workers we recruited from 5 sites, concerning their fraud posting work in Google Play, and also a quantitative investigation with data that we collected from 39 other ASO workers recruited from the same sites.”

May combine two claims.

Most devices used to post reviews are old and low-end (45.98%), mid-end (31.41%), or fresh and low-end (15.31%); high-end and free devices have also been used.

“Most devices are old and low-end (45.98%) or mid-end (31.41%), or fresh and low-end (15.31%).High-end and even free devices have been used!”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Participants claimed expertise in fake reviews and ratings on Google Maps, Apple Store, Amazon, Facebook, and Twitter, fake installs in Apple App Store, fake likes and followers in Facebook and Instagram, and influential tweets in Twitter.

“The participants in our studies also claimed expertise in fake reviews and ratings in Google Maps, Apple Store, Amazon, Facebook and Twitter, fake installs in Apple App Store, fake likes and followers in Facebook and Instagram, and influential tweets in Twitter.”

• AssumptionRests on an assumption: The participants' self-reported strategies and behaviors are accurate.

May combine two claims.

Eight workers have very short inactive intervals and are more intensively active for the apps they target.

“8 workers have very short inactive intervals, thus are more intensively active for the apps that they target.”

May combine two claims.

The mean number of jobs assigned by the same developer to the same worker is 3.48, and 7 workers were hired by the same developer more than 10 times.

“We observe that the mean number of jobs assigned is 3.48, and 7 workers have been hired by the same developer more than 10 times.”

May combine two claims.

The median and third quartile of the time between app launch and the first reviews from accounts of the 39 participants are 113 and 344 days, respectively.

“the median and 3rd quartile are 113 and 344 days respectively.”

May combine two claims.

For the 585 apps that received at least 10 fraudulent reviews, the distribution of time between app launch and fraudulent review posting is left-skewed, with half of the reviews posted less than three months after launch.

“The distribution is left-skewed, with 50% of the reviews being posted after less than 3 months after app launch.”

May combine two claims.

Twelve workers have complete co-review graphs, with density equal to 1.

“12 workers have complete graphs (den-sity=1).”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The study used GDPR-recommended pseudonymisation and deleted all device-to-identity links after data collection.

“We used GDPR [70] recommended pseudonymisation for data processing and statistics, and other generally accepted good practices for privacy preservation.After data collection, we have deleted all device-to-identity links and only generated statistics that allowed us to validate our assumptions.”

• cites [70] sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

The study procedure was approved by the institutional review board of a major North American university.

“The whole study procedure was scrutinized and approved by the institutional review board of a major North American university (IRB-18-0077@FIU).”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Google accepted the authors' device model identification finding and invited them to join its hall of fame.

“Google has accepted our finding and has invited us to join their hall of fame.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Two researchers independently coded the interview data, achieving an inter-coder agreement of 84.61%.

“Two members of our team independently coded the data.The inter-coder agreement was 84.61%.”

May combine two claims.

Fraudulent behaviors, including fake reviews and sockpuppet accounts, are prevalent on popular online peer-opinion sites and negatively affect users.

“Popular online services that provide millions of users with access to products, news, social relationships and peer-opinions, are besieged by fraudulent behaviors, that skew public opinion and bias product reputation and popularity [25,35,45,54,75,80,91].”

• cites [25]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [35]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [45] sourceNot checked — no full text of the source
• cites [54]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [75]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [80]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [91]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

The quote only partly states this claim.

The study confirms that fraud is perpetrated by experts who control many sockpuppet user accounts.

“Our study thus confirms observations made by existing work, that fraud is perpetrated by experts who control either (1) many sockpuppet user accounts, e.g., [28, 37, recruited online [13,17].”

• cites [28]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [37] sourceNot checked — no full text of the source
• cites [13]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [17] sourceNot checked — no full text of the source

• AssumptionRests on an assumption: The ASO workers' self-reported claims about their activities, accounts, devices, and organizations are truthful and accurate.

May combine two claims.

Rate control implies longer campaigns because workers need more time to post their review quota, and this is supported by interview statements and quantitative evidence.

“Rate control implies longer campaigns, as workers need more time to post their review quota.This is further supported by statements made by several interview participants and by evidence we extract from the quantitative investigation.”

May combine two claims. The quote only partly states this claim.

The paper's interviews confirmed prior observations that ASO workers adjust behavior to avoid detection, including using VPNs and mobile device emulators on PCs, write genuine reviews for unhired products, upvote their own reviews, and use singleton accounts; the paper also reports new behaviors of downvoting negative reviews and unexpected benefits of singleton accounts.

“Further, we identified new black hat ASO behaviors, that include downvoting negative reviews to promote their filtering by Google, and the unexpected benefits of using singleton accounts.”

May combine two claims.

The review-without-install vulnerability breaks Google's intended security design and facilitates fake reviews by reducing the resources ASO workers need.

“This vulnerability breaks Google's intended security design [20] and facilitates the creation of fake reviews by reducing the amount of resources needed from the ASO worker.”

• cites [20]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Eight workers had very short inactive intervals and were more intensively active for targeted apps, while workers such as F3, F24, F32, and F33 had longer inactive intervals, suggesting rehiring.

“8 workers have very short inactive intervals, thus are more intensively active for the apps that they target.However, ASO workers such as F3, F24, F32 and F33, have longer inactive intervals, suggesting rehiring.”

May combine two claims.

Qualitative and quantitative studies provide evidence that ASO workers tend to be hired early after app launch, or even before launch, to control review sentiment, and that developers rehire some workers later when honest feedback reduces product rating.

“Our qualitative and quantitative studies provide evidence confirming observations and assumptions made in previous work, that (1) ASO workers tend to be hired early after app launch, or even before launch, to control review sentiment, see e.g., [38,53,61,63,64,96,100] and (2) developers rehire some of these workers at later times, when honest feedback reduces the product rating [53].”

• cites [38]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [53]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [61]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [63]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [64]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [96]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites [100]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

Multiple survey answers were corroborated with quantitative measurements and manual verification by the authors.

“We have corroborated multiple survey answers with quantitative measurements, and also manual verification by the authors.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The device-model vulnerability could be used by ASO workers to blend in by mimicking the distribution of devices observed in Google Play.

“However, this vulnerability could also be used by ASO workers to blend in with normal users, by mimicking the distribution of devices observed in Google Play.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The study does not claim that its findings apply to other sites or other types of ASO work.

“Thus, we do not claim that our findings apply to other sites or other types of ASO work.”

May combine two claims.

The study did not perform a complete exploration of the ASO worker universe and cannot claim participants are a representative sample.

“We have not performed a complete exploration of the ASO worker universe, and cannot claim that our participants are a representative sample.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The study does not know all accounts controlled by the 39 ASO workers.

“First, we do not know all the accounts controlled by the 39 ASO workers.”

Classified as probably not a substantial claim.

The presented findings should be understood as situated information and not as generalized facts.

“The presented findings are hence needed to be understood as situated information and not as generalized facts.”

May combine two claims.

Different strategies have different impact on ASO workers' ability to avoid detection and on the ratings of apps they target.

“We conclude that different strategies have different impact on the ability of ASO workers to avoid detection and impact the ratings of apps that they target.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The quantitative study was performed only on a subset of accounts controlled by 39 participants.

“Our quantitative study is also performed only on a subset of the accounts controlled by 39 participants.”

May combine two claims.

The paper reports Google Play vulnerabilities and new findings about ASO worker capabilities, behaviors, and detection avoidance strategies.

“We report Google Play vulnerabilities, and new findings about the capabilities, behaviors and detection avoidance strategies claimed and exhibited by ASO workers.”

May combine two claims.

The computed average rating of an app is imperfect because the study lacks access to ratings posted without reviews and may not correctly model Google's algorithm.

“Finally, our computed average rating of an app is imperfect, since (1) we do not have access to ratings posted without reviews, and (2) may not correctly model Google's algorithm, that e.g., may assign weights to ratings based on perceived usefulness, fraudulence or recency [72].”

• cites [72] sourceNot checked — no full text of the source

Classified as probably not a substantial claim.

The study did not ask participants about their strategies on other platforms.

“However, we did not ask participants, questions about their strategies in other platforms.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Other workers will likely have both fewer and more capabilities than the study participants.

“Thus, other workers will likely have both fewer and more capabilities than the participants in our studies.”

May combine two claims.

The study reveals previously unknown ASO strategies, insights into previously proposed defenses, and Google defense vulnerabilities.

“we reveal previously unknown ASO strategies, provide insights into previously proposed defenses that may be effective against them, and report Google defense vulnerabilities.”

May combine two claims.

Remote attestation inspired solutions without device TPMs will not be secure, and defeating such solutions will require significant investment from ASO workers.

“While remote attestation inspired solutions (e.g., [46]) will not be secure without device TPMs, defeating such solutions will require significant investment from ASO workers.”

• cites [46]Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Probably not a substantial claim: may not pass S3 (load-bearing).

The study does not claim to have contacted most active ASO workers.

“However, we do not claim that we were able to contact most of the active ASO workers.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

The cellular provider used during account validation can be used to detect inconsistencies with the claimed profile of the user account.

“The cellular provider used during account validation can also be used to detect inconsistencies with the claimed profile (e.g., location) of the user account.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

The opaque nature of commercial fraud detection systems prevents establishing costs and scalability of recommendations or determining if they are already implemented.

“The opaque nature of commercial fraud detection systems prevents us from establishing the costs and scalability of implementing the proposed recommendations, or from determining if they are already implemented.”

Classified as probably not a substantial claim. May combine two claims.

The study is limited by difficulty recruiting participants and sensitivity of the data.

“Taken together, our study is limited by the difficulty to recruit participants and the sensitivity of the data.”

Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

The findings provide new suggestions for future research.

“the presented findings, even with all their limitations, provide new suggestions for future research.”

May combine two claims. The quote only partly states this claim.

Organic ASO workers are likely to use their devices continuously, like normal users, while sockpuppet accounts are more likely to experience inactive interludes due to their workflow.

“Organic ASO workers are likely to use their devices continuously, like the normal users that they almost are.”

• Also stated in: § 6 DISCUSSION AND RECOMMENDATIONS · hedged

  “Sockpuppet accounts are more likely to experience inactive interludes given the dynamic of their workflow ( § 5.5).”

Probably not a substantial claim: may not pass S3 (load-bearing).

The computed impact is oblivious to simultaneous campaigns by other workers on the same apps.

“Third, the impact that we computed, is oblivious to simultaneous campaigns being conducted by other workers on the same apps.”