{"id":"30ce088a-eb85-4e57-bbab-1609ba6f84d3","arxiv_id":"2611.13125","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"MiniZoo is an evolving dataset of 16,273 WeChat mini-app packages with metadata, supporting reproducible research on mini-app software quality, security, and privacy.","lead":"MiniZoo is a new dataset of 16,273 WeChat mini-app packages with metadata, designed to grow over time. It provides a shared resource for research on mini-app software quality, security, and privacy, similar to AndroZoo for Android apps.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"No significant objection identified: the dataset claim is internally consistent, and the weakest point—discovery bias—is explicitly acknowledged and does not undermine the stated contribution.","rationale":"The paper is a dataset-description paper; its main claims are existence, scale, organization, and planned evolution. I checked the arithmetic and definitions: 13,009 apps with 1,149 multi-version apps can yield 16,273 distinct packages under plausible per-app version counts, and the SHA-256/AppID distinction is coherent. The collection pipeline is described in enough detail to be reproducible, with crawler code released. The authors' Section VI limitations accurately cover coverage, temporal gaps, maintenance, and platform scope; no unacknowledged internal contradiction surfaced. The reader's weakest assumption—keyword-discovery representativeness—is real but explicitly disclaimed, so it does not undermine the dataset's stated research value. The only load-bearing unverified element is external availability of the live archive, which is a deployment check rather than a scientific flaw. The verdict therefore remains unchanged.","tokens_in":10209,"tokens_out":4833,"duration_ms":67322,"concrete_test":"Obtain a test AndroZoo API key, download the current minizoo.tar.gz archive, and run an independent consistency check: count rows in minizoo.csv, verify that all SHA-256 hashes are unique and match the provided .wxapkg files, confirm that each AppID in minizoo_metadata.jsonl appears in minizoo.csv, and recompute the total package volume from the size field to compare with Table I. If these checks pass, the central availability claim is confirmed; any mismatch would pinpoint the exact record needing correction.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is that MiniZoo provides a usable, growing corpus of WeChat mini-app packages with metadata, distributed through AndroZoo. The least secure condition is the representativeness of the keyword-based discovery queue (Section III), since 688 queries cannot cover the whole WeChat ecosystem. However, the authors explicitly disclaim exhaustive coverage (b104) and frame the goal as collecting and preserving as many mini apps as possible rather than guaranteeing completeness (b105), so the acknowledged coverage bias does not contradict the stated contribution. The crawler design also includes concrete safeguards: AppID checks, WeChat checksum verification, re-inspection after acquisition, and AppID–SHA-256 deduplication (Section III). The only remaining trust anchor is external—whether the live archive at androzoo.uni.lu/minizoo actually matches the reported counts and metadata schema—which is a verification issue, not a flaw in the paper's argument. I therefore find no load-bearing technical concern.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"MiniZoo is a resource paper: it presents an evolving, publicly distributed corpus of WeChat mini-app packages, hosted and managed through the AndroZoo infrastructure. As of the reported snapshot (9 November 2026), the corpus comprises 13,009 mini apps and 16,273 distinct package contents (SHA-256), totaling 30.33 GiB, of which 1,149 apps (8.83%) have multiple recorded versions. The paper's main technical content is the description of the collection pipeline (Section III): a keyword-based discovery module (688 retained queries, with the query space expanded by app names and overlapping Chinese substrings), an initial crawler that verifies the opened app's AppID against the requested one, and a version crawler that revisits previously collected apps; both crawlers validate packages against WeChat checksums, re-inspect the app after acquisition, and deduplicate on (AppID, SHA-256) pairs. The data organization (Section II) consists of a CSV index, a JSON Lines metadata file, and the .wxapkg packages, distributed as a nightly-updated tar.gz archive under AndroZoo access conditions (Section IV). The paper closes with research opportunities (Section V) and an explicit account of coverage, temporal, maintenance, and platform-scope limitations (Section VI).","tokens_in":10319,"tokens_out":14215,"duration_ms":105207,"significance":"The paper addresses a real gap: prior mini-app collections are either very small, not publicly shared, or static snapshots, and none is tied to the AndroZoo distribution infrastructure that the community already uses. The strengths are concrete rather than rhetorical. First, the integrity checks in Section III (AppID verification before and after acquisition, WeChat checksum validation, re-inspection, and (AppID, SHA-256) deduplication) give reusers a falsifiable basis for trusting the packages. Second, the weakest methodological point, the representativeness of keyword-based discovery, is explicitly acknowledged in Section VI, where the authors disclaim exhaustive coverage and frame the goal as preservation of as many mini apps as possible; that acknowledged bias does not contradict the stated contribution, so I do not treat it as a load-bearing objection. Third, both the crawler code (GitHub) and the dataset (androzoo.uni.lu/minizoo) are public, making the reported counts checkable. Fourth, the platform-agnostic metadata schema is a credible foundation for the stated extension to Alipay, Baidu, and Telegram. The paper makes no claim stronger than its methodology supports.","major_comments":[],"minor_comments":[{"comment":"The sentence 'These results are then combined with existing catalogue and collection records to build the discovery catalogue' leaves the origin, size, and provenance of the 'existing catalogue and collection records' unspecified; since the paper presents provenance as part of its contribution, please state what these records are, where they come from, and how much of the 13,009 mini apps each discovery source contributed.","section":"Section III (Mini App Discovery)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"Please state explicitly whether the 16,273 'distinct package contents (SHA-256)' count individual .wxapkg files including subpackages; if subpackages are counted as separate entries, clarify how readers should map this figure to app-version snapshots, and consider reporting the number of distinct (AppID, version code) pairs.","section":"Section II, Table I","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The collection environment is reported precisely (en-US locale, Europe/Luxembourg timezone, WeChat 8.0.72, Android 16), but the possible effect of language and region on WeChat search results is not discussed in the limitations; a sentence in Section VI on this source of selection bias would help reusers assess representativeness.","section":"Section III vs. Section VI","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The JSON Lines metadata is described only through representative fields; listing the exact field names and value formats, or pointing to the schema within the archive, would make the metadata description self-contained.","section":"Section II (minizoo_metadata.jsonl)","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"minor_revision"},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"Colleague,\n\nQuick take: this is a solid, honest dataset paper, and the only genuinely new thing in it is the resource itself — a publicly hosted, versioned, growing WeChat mini-app corpus, distributed through AndroZoo. That resource is worth having; the paper that describes it is competent but not deep.\n\nWhat it does well: the authors separate mini-app identity (AppID) from package content (SHA-256), which is exactly the right schema for reuse and for any longitudinal work. The crawler section is concrete: discovery by keyword search (688 queries, Sec. III), AppID verification before acquisition, WeChat checksum verification, re-inspection after copying, and AppID–SHA-256 deduplication that keeps different apps separate even when content hashes collide. Those are real safeguards, and the Web-facing description (access via AndroZoo API key, nightly tar.gz refresh) is enough for someone to actually use the dataset tomorrow. The authors also state their limitations plainly — coverage is not exhaustive [b104–b105], version histories have temporal gaps [b106–b107], maintenance cost is acknowledged [b108–b109], and only WeChat is covered [b110]. I did not find a load-bearing technical flaw. The reader's weakest-assumption point (keyword discovery bias) is real but explicitly acknowledged, and it does not contradict the stated contribution of \"as many as possible, not exhaustive.\"\n\nWhere it is soft, in proportion: the novelty is genuinely modest — the crawler is adapted from MiniCrawler, the distribution reuses AndroZoo's infrastructure, and the paper does not report any measurement or validation beyond package counts and size distribution. The claim that the metadata supports privacy/security research is plausible but unverified; the paper does not show, for example, that the 1,149 multi-version apps actually exhibit meaningful evolution, or that the corpus reproduces any prior result. The only real external trust anchor is the live site (androzoo.uni.lu/minizoo) matching the reported counts; that is a verification issue, not an argument flaw. One small oddity: the retained discovery log lists 688 queries and the corpus is 13k apps — fine, but the paper gives no sense of how many queries failed or how discovery will scale past the backlog.\n\nWho this is for: researchers working on WeChat mini-app security, privacy, or quality who need a shared body of packages with version information. That audience will reach for the dataset, not the paper. It deserves peer review — a serious editor should send it out because the community needs a citable artifact description and the data is verifiable. My recommendation: accept with the usual request to tighten the validation story in the research-opportunities section.\n\nCheers.","headline":"A sincere, well-scoped dataset paper for the mini-app community: MiniZoo delivers 16k WeChat mini-app packages with version history, clear crawling methodology, and an honest limitations section.","tokens_in":10708,"tokens_out":1376,"duration_ms":13544},"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek/deepseek-v4-flash","headline":"MiniZoo gives the research community a shared, continuously grown corpus of WeChat mini apps: 13,009 apps, 16,273 packages, and preserved version histories.","keywords":["mini apps","WeChat mini programs","software repository","dataset","software evolution","security analysis","privacy analysis","AndroZoo"],"falsifier":"Take an independently built list of WeChat mini apps, for instance the AppIDs studied in prior measurement work, and count how many of them appear in MiniZoo's keyword-derived catalogue; a low recall would show the corpus is skewed toward whatever the 688 seed queries happened to surface. A second check targets the longitudinal promise: revisit a sample of the 1,149 multi-version apps and see whether intermediate releases are systematically missing, which would make the recorded version histories sparser than the dataset's framing suggests.","tokens_in":11603,"feed_emoji":"📦","tokens_out":12599,"duration_ms":92249,"temperature":0.7,"pith_summary":"MiniZoo sets out to give the mini-app research community what AndroZoo gave Android research: a shared, continuously maintained collection rather than scattered one-off snapshots. The initial release holds 13,009 WeChat mini apps, 16,273 distinct package contents, and 30.33 GiB of data, including 1,149 apps captured in more than one version. Its central organizational idea is to separate the mini app's stable identity, the AppID, from the hash of a specific package file, so the same app can accrue a version history. The collection is rebuilt nightly and distributed under AndroZoo's access conditions, with plans to add further host ecosystems. The value is that security, privacy, and quality studies of mini apps have been hard to compare because each research team collected its own small or partial dataset; a common evolving corpus makes independent evaluation and longitudinal analysis feasible.","feed_headline":"MiniZoo releases 16,273 WeChat mini-app packages for research","feed_subtitle":"Nightly-updated with version histories, it lets security studies share one corpus instead of one-off collections.","key_machinery":"The load-bearing abstraction is the identity split: AppID names the mini app, the way an Android package name names an app; SHA-256 names a specific package archive, the way an APK hash does; and the pair of the two is the corpus's deduplication key. That split is what lets the dataset grow by versions instead of by raw file count and is the premise behind the metadata schema and the version crawler. Around it sits the three-module collection workflow — discovery seeded by 688 keyword queries, a crawler for newly found apps, and a revisiting version crawler — which is the mechanism that keeps the corpus evolving between nightly archive rebuilds.","core_discovery":"The paper's central claim is that a shared, continuously maintained corpus of mini apps is not just desirable but feasible, and that MiniZoo is that corpus. Its foundational distinction is between a mini app's stable identity, the WeChat AppID, and a package's content identity, its SHA-256 hash: the same AppID maps to many packages over time, and that mapping is what makes version history and longitudinal study possible. On this basis the paper reports 13,009 mini apps with at least one associated package, 16,273 distinct package contents, a total volume of 30.33 GiB, and 1,149 apps (8.83 percent) with multiple recorded versions as of November 9, 2026. Collection runs through three modules — keyword-based discovery, a crawler for new apps, and a version crawler that revisits known apps — whose outputs are consolidated and deduplicated on the AppID-hash pair. The authors argue the resource fills a concrete gap: existing mini-app collections are tiny fixed snapshots, or large corpora whose owners share only selected samples, so tool evaluations and measurements cannot be compared or repeated. Hosting through AndroZoo supplies a maintenance model, access conditions, and a distribution channel already established in the Android research community, with a nightly-updated archive and planned selective downloads.","pith_inferences":["If MiniZoo follows AndroZoo's growth trajectory, a testable consequence is that published mini-app studies will begin citing it as their data source, which would make independent replication of security and privacy measurements routine.","The AppID-hash schema is claimed to be platform-agnostic; a concrete test is whether Alipay, Baidu, or Telegram packages fit the existing CSV and JSON Lines structure without schema changes.","The authors acknowledge gaps in version histories; a follow-up they do not propose is publishing per-app revisit schedules so longitudinal studies can model the irregular sampling explicitly.","The discovery log's coverage is unmeasured, so an external benchmark comparing MiniZoo's AppID set against known reference lists would let researchers quantify the corpus's biases before drawing ecosystem-wide conclusions."],"forward_implications":["TaintMini, WeMinT, and any other mini-app analysis tool can be run on identical packages from one corpus, making published security and privacy measurements directly comparable rather than dependent on each team's private collection.","The 1,149 apps with multiple recorded versions already provide a base for studying how package size, API usage, and dependencies change; continued collection widens that base.","Because the archive is rebuilt nightly, researchers always obtain a fresh, dated snapshot of the ecosystem rather than a frozen release.","Planned selective downloads by SHA-256 or AppID will let researchers retrieve only the packages they need, following AndroZoo's established download model.","Extending dedicated crawlers to Alipay, Baidu, and Telegram would produce cross-ecosystem comparisons under the same metadata schema."],"supporting_citations":[{"why":"Supplies the hosting, access-conditions, and continuous-growth model MiniZoo adopts, and the evidence that a shared corpus can sustain a research community.","marker":"[23]"},{"why":"Supplies the maintenance experience behind AndroZoo that MiniZoo draws on to keep a large evolving corpus operating.","marker":"[24]"},{"why":"Supplies the discovery technique, keyword searches with expanding query queues, that the Mini App Discovery module is inspired by.","marker":"[13]"},{"why":"Supplies the dynamic-instrumentation toolkit both crawlers use to locate and inspect running mini apps inside WeChat.","marker":"[25]"},{"why":"Names the concrete use case: TaintMini's static taint analysis can be evaluated against identical MiniZoo packages.","marker":"[7]"},{"why":"Contrast case for the gap being filled: a community collection reported as very large but shared only as selected samples.","marker":"[21]"},{"why":"Contrast case for the scale problem: public mini-app artifacts can contain as few as ten apps.","marker":"[22]"}],"fun_headline_variants":["MiniZoo opens 16,273 mini-app packages with version histories","Shared mini-app corpus holds 16,273 packages across 13,009 apps","MiniZoo tracks AppID-to-hash versions for 16,273 mini-app packages","AndroZoo-hosted MiniZoo offers 16,273 mini-app packages for reuse","MiniZoo dataset maps 13,009 mini apps to 16,273 package versions"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The resource is only as representative as its discovery method: MiniZoo's usefulness to mini-app research generally assumes that keyword searches on WeChat surface a sufficiently diverse and representative slice of the ecosystem, and the authors themselves state that they cannot enumerate all mini apps to measure what share they captured.","fun_headline_variants_meta":{"raw":{"variants":["MiniZoo opens 16,273 mini-app packages with version histories","Shared mini-app corpus holds 16,273 packages across 13,009 apps","MiniZoo tracks AppID-to-hash versions for 16,273 mini-app packages","AndroZoo-hosted MiniZoo offers 16,273 mini-app packages for reuse","MiniZoo dataset maps 13,009 mini apps to 16,273 package versions"]},"model":"deepseek/deepseek-v4-flash","effort":"low","cost_usd":0.000365,"raw_usage":{"cost":0.000365183,"is_byok":false,"cost_details":{"upstream_inference_cost":0.000365183,"upstream_inference_prompt_cost":9.8735e-05,"upstream_inference_completions_cost":0.000266448},"total_tokens":2549,"prompt_tokens":1085,"completion_tokens":1464,"prompt_tokens_details":{"audio_tokens":0,"video_tokens":0,"cached_tokens":0,"cache_write_tokens":0},"completion_tokens_details":{"audio_tokens":0,"image_tokens":0,"reasoning_tokens":1367}},"tokens_in":1085,"tokens_out":1464,"cost_source":"openrouter_usage","duration_ms":10327,"image_pages":[],"temperature":1.0,"billed_attempts":1,"reasoning_tokens":1367,"transport_provider":"openrouter","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-11-11T18:00:02.007409+00:00","model_set":{"reader":"deepseek/deepseek-v4-flash"},"pipeline_version":"pith-pipeline@v1.4.1-daily-deepseek-or-medium","input_kind":null,"input_chars":null,"input_total_chars":null,"passes_missing":null,"published_standing":null,"published_confidence":null,"objection_audit":null,"run_record":null,"basis":{"kind":"not_recorded","source":null,"chars_given":null,"chars_total":null},"receipt":null,"falsifier":"Take an independently built list of WeChat mini apps, for instance the AppIDs studied in prior measurement work, and count how many of them appear in MiniZoo's keyword-derived catalogue; a low recall would show the corpus is skewed toward whatever the 688 seed queries happened to surface. A second check targets the longitudinal promise: revisit a sample of the 1,149 multi-version apps and see whether intermediate releases are systematically missing, which would make the recorded version histories sparser than the dataset's framing suggests.","supporting_citations":[{"cited_title":"AndroZoo: collecting millions of Android apps for the research community,","cited_arxiv_id":null,"evidence_quote":"Supplies the hosting, access-conditions, and continuous-growth model MiniZoo adopts, and the evidence that a shared corpus can sustain a research community."},{"cited_title":"AndroZoo: A Retrospective with a Glimpse into the Future,","cited_arxiv_id":null,"evidence_quote":"Supplies the maintenance experience behind AndroZoo that MiniZoo draws on to keep a large evolving corpus operating."},{"cited_title":"A Mea- surement Study of Wechat Mini-Apps,","cited_arxiv_id":null,"evidence_quote":"Supplies the discovery technique, keyword searches with expanding query queues, that the Mini App Discovery module is inspired by."},{"cited_title":"Frida: Dynamic Instrumentation Toolkit,","cited_arxiv_id":null,"evidence_quote":"Supplies the dynamic-instrumentation toolkit both crawlers use to locate and inspect running mini apps inside WeChat."},{"cited_title":"TaintMini: Detecting Flow of Sensitive Data in Mini-Programs with Static Taint Analysis,","cited_arxiv_id":null,"evidence_quote":"Names the concrete use case: TaintMini's static taint analysis can be evaluated against identical MiniZoo packages."},{"cited_title":"Welcome to the MiniSec Store,","cited_arxiv_id":null,"evidence_quote":"Contrast case for the gap being filled: a community collection reported as very large but shared only as selected samples."},{"cited_title":"Better Safe than Sorry: Uncovering the Insecure Resource Management in App-in- App Cloud Services,","cited_arxiv_id":null,"evidence_quote":"Contrast case for the scale problem: public mini-app artifacts can contain as few as ten apps."}],"lane_accuracy":{"status":"not yet measured","lane":{"model":"deepseek/deepseek-v4-flash","input_kind":null},"url":"https://pith.science/accuracy"},"review_version":1}