0 · The catalog (raw lake)
countLoad the file indexexact · ~570M428M objects · 3.48 PiB
countKind census (what the lake is)exact · ~570Mcode < 0.5% of bytes
1 · Routing by kind
routeSplit code from mediaexact→ apps / archives / loose / media
2 · Decompiled apps (route=binary)
countCount distinct APK containersexact + 2% dedup (n=45,057)2.26M uploads → ~2.13M apps
countjadx decompile → Javaexact · ~5.1B4.30B .java (gross)
dedupExact content-hash dedup3 blind slices4.30B → 1.15B distinct (~27%)
classifyLayer-1: mark obfuscated by pathexact · name-based527M files → unattributable
filterLayer-1: recognized SDK / generatedexact · name-based163M files → removed
filterLayer-1: shared libraryexact · recurrence rule57M files → removed
classifyApp-authored core (upper bound)exact · name-based403M files · ~33B lines
gateApp-level authored gatesampled → TVD 0.00682.1M → 1.6M apps
verifyLayer-2 body gradingn=293 apps / 1,620 files~93% of name-clean app code is body-clean
filterModded-app family filterexact · full pop~50–60k mod variants
3 · Source archives (route=archive)
countCount distinct extracted archivesexact727,617 archives
filterSource filter (≥1 real source file)exact727k → 320k source archives
dedupCross-channel re-upload dedupexact320k → ~263k distinct
gateRepo-level gate (lenient)strip by pattern · SWH n=670320k → ~300k source repos
classifyLanguage classificationexactJS 93k · PHP/Py 56k · C/C++ 43k · …
filterVendored / library / SDK-dump strip (lenient)exact · patternsstrip 100%-known garbage only
verifyPer-language authored split (exact)exact · full pop/langC/C++ 33% · Java 85% · …
verifyNovelty check — Software Heritage (definitive)n=670 archives / 9,152 files~76% of repos contain novel code
measureFile-count & depth distributionsn=72,417 (10%)structure shape of the tier
4 · Loose files (non-archive)
countLoose text poolexact · 5.5M5.51M loose uploads
classifyDeterministic extension classifyexact · 5.5Mtxt 41% · xml 19% · html 10%
verifyContent spotcheck (read the bytes)n=1,023 files read1,023 files read · per-ext purity
measureReweight to poolexact × measured purity~496k code files · 9%
5 · Lines (LoC)
measureLine calibration (not bytes÷35)2.08M counted linessource density re-measured
measureDistinct vs kept (delivered) frames10% sample~10B unique / ~38–43B delivered
6 · The 0.2% download sample
filterSelect candidatesuniform hash samplecityHash % K sample
filterFetch complete repos + completeness filtermechanicalfull repos (code + docs)
measureSignals: minified + noveltyper-repominified_frac · AF · PUB
filterFilter + score + capmechanical → ~500 repos~500 curated repos
7 · Consistency & verification
verifySankey internal consistencyexact · arithmeticleft = right = 2.425M
verifyCross-figure reconciliationexact · full-page auditone number per quantity
verifyStatistical soundness6 sizes (1/2000–1/50)design effect ≈ 1
all steps
Select a step to see its numbers, rule, thresholds, and confidence.