Skip to content

Deterministic machine analysis · ANALYSIS REPORT

China's network filtering has no honest single rate

Multiple scoped instruments report observable network interference, while their incompatible panels, protocols and denominators rule out one national percentage.

Current gate result: Published because citations are complete, all 3 republished measurements are fresh and rights-eligible, and 2 lineage-independent groups meet the configured minimum of 2.

DETERMINISTIC MACHINE ANALYSIS · NO HUMAN INTERVIEW. This report uses checked-in aggregate evidence only. It cannot add testimony, observe private conduct or treat missing reporting as confirmation.

Report type
AnalysisReport
automated_evidence_analysis
Updated
Published
12 Aug 2026 · 16:51 UTC
Revision receipt
machinev-5e1357c794ace1a5f3d423cc
Immutable revision JSON

Analysis with citations attached

The paragraph is the published unit; the ledger beneath it shows the exact evidence IDs attached to each sentence and the de-duplicated lineage union for the block.

Claim block 1 · network-denominators

OONI anomaly index reports 58.3 percent across 272,935 completed measurements, OONI in-path middlebox index reports 1.37 percent across 9,283 completed middlebox tests, and Inside-China fixed DNS panel reports 1.0 ratio across 5 qualifying answered measurement domains. The three cited values do not share a population, protocol, test family, denominator, or even a common unit, while measurements with the same independence-group identifier count only once. The cited measurements record observable interference within their named methods, but they cannot be combined into one defensible national filtering percentage.

  1. OONI anomaly index reports 58.3 percent across 272,935 completed measurements, OONI in-path middlebox index reports 1.37 percent across 9,283 completed middlebox tests, and Inside-China fixed DNS panel reports 1.0 ratio across 5 qualifying answered measurement domains.

    Sentence 1 citations: evidence-ooni-gfw, evidence-in-path-interference, evidence-inside-view

  2. The three cited values do not share a population, protocol, test family, denominator, or even a common unit, while measurements with the same independence-group identifier count only once.

    Sentence 2 citations: evidence-ooni-gfw, evidence-in-path-interference, evidence-inside-view

  3. The cited measurements record observable interference within their named methods, but they cannot be combined into one defensible national filtering percentage.

    Sentence 3 citations: evidence-ooni-gfw, evidence-inside-view

Block evidence union: evidence-ooni-gfw, evidence-in-path-interference, evidence-inside-view
Independent lineage groups: pipeline:palimpsest-inside-view, publisher:ooni

Claim block 2 · network-independence

After lineage de-duplication, the cited record contains two independent groups: pipeline:palimpsest-inside-view and publisher:ooni. The in-path instrument remains useful as a distinct technical view, but it contributes no additional independent group because it reuses OONI measurements.

  1. After lineage de-duplication, the cited record contains two independent groups: pipeline:palimpsest-inside-view and publisher:ooni.

    Sentence 1 citations: evidence-ooni-gfw, evidence-inside-view

  2. The in-path instrument remains useful as a distinct technical view, but it contributes no additional independent group because it reuses OONI measurements.

    Sentence 2 citations: evidence-ooni-gfw, evidence-in-path-interference

Block evidence union: evidence-ooni-gfw, evidence-inside-view, evidence-in-path-interference
Independent lineage groups: pipeline:palimpsest-inside-view, publisher:ooni

When the evidence arrived—and which sources travel together

Evidence chronology and source lineageOrdering exposes which observations predate the report and which receipts share a lineage. It is descriptive, not a causal sequence.
  1. OONI anomaly indexsupport · publisher:ooniInspect receipt
  2. OONI in-path middlebox indexsupport · publisher:ooniInspect receipt
  3. Inside-China fixed DNS panelsupport · pipeline:palimpsest-inside-viewInspect receipt

Receipts, values, rights and interpretation limits

Each public archive is a redacted aggregate capsule bound to the original input hash. Raw readings, IP-valued answers, contact data and person-level records are not copied into the archive.

Evidence receipts cited sentence by sentence in this report
RelationEvidence / lineageValue or summaryObservedInterpretation limit
supportRESEARCH_DATASET OONI anomaly indexevidence-ooni-gfw · publisher:ooniooni-gfw-latest.json · sha256 90accfb51b97976a774e922f9874ab3af2d4fe9a81b934354dfa714981bf1009 58.3 percent; denominator: 272935 completed measurements 24 Aug 2026 · 02:00 UTC OONI's anomaly index uses completed measurements across several test families; the separate total-measurement count also includes failed attempts. An anomaly is not a confirmed block and the probe/test mix is not China's population.Open redacted evidence capsule (addressed by original input hash)
Attribution required for redistribution.Redistribution: ATTRIBUTION_REQUIRED · Reuse: derived_only · Training: prohibitedProvider(s): OONI (provider terms)Upstream lineage: publisher:ooniLicense / provider terms: CC BY-NC-SA 4.0 source dataLocal public source receipt · Upstream sourceSource statement: OONI aggregation API (api.ooni.io), probe_cc=CN
supportRESEARCH_DATASET OONI in-path middlebox indexevidence-in-path-interference · publisher:ooniin-path-interference-latest.json · sha256 b2eba327467c6e780d015440bd7d26a5c5f1814fdb4efade486125003e323592 1.37 percent; denominator: 9283 completed middlebox tests 24 Aug 2026 · 02:07 UTC This is derived from OONI measurements and shares the publisher:ooni lineage. Its completed-test denominator and test family differ from the OONI reachability index.Open redacted evidence capsule (addressed by original input hash)
Attribution required for redistribution.Redistribution: ATTRIBUTION_REQUIRED · Reuse: derived_only · Training: prohibitedProvider(s): OONI (provider terms)Upstream lineage: publisher:ooniLicense / provider terms: CC BY-NC-SA 4.0 source dataLocal public source receipt · Upstream sourceSource statement: OONI aggregation API (api.ooni.io), probe_cc=CN
supportRESEARCH_DATASET Inside-China fixed DNS panelevidence-inside-view · pipeline:palimpsest-inside-viewinside-view-latest.json · sha256 9fe7b2157fc21a8b26214153b614b0c01fe9f8b5e9839f45e753aa6546831291 1.0 ratio; denominator: 5 qualifying answered measurement domains 24 Aug 2026 · 02:10 UTC The ratio is for a fixed, deliberately censorship-sensitive domain panel and consented cloud probes; it is not a representative domain or user sample.Open redacted evidence capsule (addressed by original input hash)
Attribution required for redistribution.Redistribution: ATTRIBUTION_REQUIRED · Reuse: derived_only · Training: prohibitedProvider(s): Globalping (provider terms), Team Cymru (provider terms)Upstream lineage: publisher:globalping, publisher:team-cymruLicense / provider terms: MIT for Palimpsest output; provider terms applyLocal public source receipt · Upstream sourceSource statement: Globalping (api.globalping.io) volunteer probes inside mainland China

Countercases, limitations and falsifiers

These records are part of the report, not caveats hidden after the conclusion.

Countercases

  • Some OONI anomalies can arise from endpoint or network failure rather than deliberate filtering.
    countercase id
    countercase-benign-anomaly
    disposition
    Retained; the report claims observable interference, not that every anomaly is a confirmed block.

    Evidence: evidence-ooni-gfw

  • A fixed panel chosen for censorship sensitivity can saturate even if most domains remain reachable.
    countercase id
    countercase-panel-selection
    disposition
    Retained; the panel ratio is never generalized to all domains or users.

    Evidence: evidence-inside-view

Limitations

  • The instruments observe different test populations with different denominators and units.
    consequence
    No weighted average or single national percentage is calculated.
    limitation id
    limitation-incompatible-denominators
  • Probe, resolver, domain and remote-vantage coverage is not a probability sample of China's people or networks.
    consequence
    Findings are bounded to the named instruments and collection windows.
    limitation id
    limitation-vantage-coverage
  • OONI reachability and in-path metrics reuse the same publisher lineage.
    consequence
    They count as one independent group at the publication gate.
    limitation id
    limitation-shared-lineage
  • Rights-ineligible candidates excluded from numeric republication: censored-planet (LINK_ONLY/metadata_only).
    consequence
    An available source is not republished or counted when its current contract permits metadata-only reuse.
    limitation id
    limitation-rights-gated-candidate

Falsifiers

  • The observable-interference finding would weaken if independently operated China and control vantages converged within instrument error across repeated windows.
    falsifier id
    falsifier-control-convergence
    status
    not-triggered

    Evidence: evidence-ooni-gfw, evidence-inside-view

  • A national-rate estimate would become testable only after a representative sampling frame and common outcome definition exist.
    falsifier id
    falsifier-common-denominator
    status
    evidence-needed

    Evidence: evidence-ooni-gfw, evidence-in-path-interference, evidence-inside-view

Hypotheses tested

  • Network interference affecting China-facing tests is observable across independently produced measurement methods.
    disposition
    supported
    falsifier ids
    ["falsifier-control-convergence"]
    hypothesis id
    hypothesis-observable-interference

    Evidence: evidence-ooni-gfw, evidence-inside-view

  • The 3 headline values estimate one common national filtering rate.
    disposition
    rejected
    falsifier ids
    ["falsifier-common-denominator"]
    hypothesis id
    hypothesis-single-national-rate

    Evidence: evidence-ooni-gfw, evidence-in-path-interference, evidence-inside-view

Why this became AnalysisReport

The report type is the result of declared deterministic checks. Passing a gate does not imply human reporting occurred.

citation coverage
1.0
evaluated at
2026-08-24T02:38:03Z
failed gate ids
[]
independent group ids
["pipeline:palimpsest-inside-view", "publisher:ooni"]
minimum independent groups
2
observed independent groups
2
publishable
yes
status
passed
Deterministic evaluation and publication-gate checks
GateResultObserved / requiredReceipt detail
Every analytical sentence has exact evidence IDssentence-citationsyes5 / 55 of 5 analytical sentences have non-empty citation arrays.
Every republished numeric value passes redistribution and reuse gatesnumeric-reuse-rightsyes3 / 3Rights-ineligible candidates excluded from numeric republication: censored-planet (LINK_ONLY/metadata_only).
Independent evidence groups after lineage de-duplicationindependent-groupsyes2 / 2OONI and the in-path instrument count once because they share publisher:ooni lineage.
Cited measurements are current under declared source deadlinesfresh-evidenceyes3 / 3Freshness comes from the validated evidence mesh, not the machine desk's wall clock.
Countercases, limitations and falsifiers are explicitadversarial-reviewyes3 / 3The report includes all three adversarial-review surfaces.

Methodology

  1. Hash the exact four public input files and validate their declared schemas.
    reproducible
    yes
    step id
    method-pin-inputs
  2. Resolve predeclared network values and denominators against JSON pointers in their exact raw artifact bytes.
    reproducible
    yes
    step id
    method-bind-evidence
  3. Republish and count only values whose mesh resource permits derived or full-text reuse with open or attribution-required redistribution.
    reproducible
    yes
    step id
    method-gate-rights
  4. Resolve independence groups through the evidence mesh before counting corroboration.
    reproducible
    yes
    step id
    method-deduplicate-lineage
  5. Require complete sentence citations, fresh evidence, independent groups and adversarial review.
    reproducible
    yes
    step id
    method-gate-publication

Current head and preserved revisions

The mutable report JSON points at the current case; this revision stays addressable even after a correction.

Correction status
none
Last corrected
No material correction recorded
Policy
A changed input receipt creates a new immutable revision; material errors are named in this history.
  1. Initial deterministic revision for this exact cited input set.
    Revision
    machinev-539389b2bf97e09ff5b287b4
    Change type
    initial-publication
    Published
    12 Aug 2026 · 16:51 UTC
  2. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-f22f305f307866121353565d
    Change type
    data-refresh
    Published
    12 Aug 2026 · 19:56 UTC
  3. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-06d6d791c63f7ea229f7ef84
    Change type
    data-refresh
    Published
    13 Aug 2026 · 05:52 UTC
  4. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-30c6872df9cd8528d793b36a
    Change type
    data-refresh
    Published
    13 Aug 2026 · 08:12 UTC
  5. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-82e860199998691e31af4a5c
    Change type
    data-refresh
    Published
    13 Aug 2026 · 14:16 UTC
  6. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-64f016a25eafdb288d0bf588
    Change type
    data-refresh
    Published
    13 Aug 2026 · 19:46 UTC
  7. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-4922c7c9419e091569d6ceba
    Change type
    data-refresh
    Published
    14 Aug 2026 · 03:57 UTC
  8. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-ce72808af0048937fc31b859
    Change type
    data-refresh
    Published
    14 Aug 2026 · 10:39 UTC
  9. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-5cef31fa32eca8d7910626af
    Change type
    data-refresh
    Published
    14 Aug 2026 · 14:03 UTC
  10. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-ba64b8b909dcc36622f0f2cb
    Change type
    data-refresh
    Published
    14 Aug 2026 · 19:42 UTC
  11. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-8412c57fd4f37c224e42ca14
    Change type
    data-refresh
    Published
    15 Aug 2026 · 02:08 UTC
  12. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-8003dbc0bd6e678cdb23cb13
    Change type
    data-refresh
    Published
    15 Aug 2026 · 07:19 UTC
  13. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-9eb09629e25e643bd095a631
    Change type
    data-refresh
    Published
    15 Aug 2026 · 13:04 UTC
  14. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-44601e9340d4588f15e28f48
    Change type
    data-refresh
    Published
    15 Aug 2026 · 13:07 UTC
  15. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-8c8cc87127b481248eaf5c4d
    Change type
    data-refresh
    Published
    15 Aug 2026 · 13:23 UTC
  16. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-582cc95ef8d64d5c2c3f9fbd
    Change type
    data-refresh
    Published
    15 Aug 2026 · 18:52 UTC
  17. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-298ffd57bbecb7d27688e3dd
    Change type
    data-refresh
    Published
    15 Aug 2026 · 18:54 UTC
  18. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-35446fd21ec033ee8b41d12e
    Change type
    data-refresh
    Published
    15 Aug 2026 · 19:15 UTC
  19. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-b28e6aa25c47f5612499076a
    Change type
    data-refresh
    Published
    16 Aug 2026 · 02:02 UTC
  20. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-81baf71220e567baf30e4592
    Change type
    data-refresh
    Published
    16 Aug 2026 · 02:05 UTC
  21. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-b4b943680eea4ceba65870b3
    Change type
    data-refresh
    Published
    16 Aug 2026 · 02:15 UTC
  22. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-946cdd79759fa7618249bd9d
    Change type
    data-refresh
    Published
    16 Aug 2026 · 07:19 UTC
  23. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-28b157779aeaa4371f691b72
    Change type
    data-refresh
    Published
    16 Aug 2026 · 13:07 UTC
  24. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-953d4773331b7a8f9faf1192
    Change type
    data-refresh
    Published
    16 Aug 2026 · 13:09 UTC
  25. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-0344e55baa57f0159fca4266
    Change type
    data-refresh
    Published
    16 Aug 2026 · 13:21 UTC
  26. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-0b8987ea3174dc6590a5dc22
    Change type
    data-refresh
    Published
    16 Aug 2026 · 19:07 UTC
  27. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-aa53cf7212a1cd43c583cc06
    Change type
    data-refresh
    Published
    17 Aug 2026 · 02:11 UTC
  28. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-ce515daffa912909c9ac6796
    Change type
    data-refresh
    Published
    17 Aug 2026 · 07:47 UTC
  29. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-b98fea7f9e5a5a3acf23dafa
    Change type
    data-refresh
    Published
    18 Aug 2026 · 11:54 UTC
  30. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-3285ef34c3a1ba3de8223950
    Change type
    data-refresh
    Published
    19 Aug 2026 · 07:32 UTC
  31. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-311dc403f9323c1522445054
    Change type
    data-refresh
    Published
    20 Aug 2026 · 03:58 UTC
  32. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-42199ebaa27f319f743616a8
    Change type
    data-refresh
    Published
    20 Aug 2026 · 07:34 UTC
  33. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-bf3887f602c2c2376f486699
    Change type
    data-refresh
    Published
    20 Aug 2026 · 14:52 UTC
  34. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-21011a599b2ad49c5c795c80
    Change type
    data-refresh
    Published
    22 Aug 2026 · 02:07 UTC
  35. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-9e4ce82aafbbbd83cf6bdf00
    Change type
    data-refresh
    Published
    22 Aug 2026 · 07:39 UTC
  36. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-d9761bab226a0c8288f550c7
    Change type
    data-refresh
    Published
    22 Aug 2026 · 13:29 UTC
  37. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-de60edeecde015f1f9b94390
    Change type
    data-refresh
    Published
    22 Aug 2026 · 19:02 UTC
  38. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-61826a1941b1b224a85e054d
    Change type
    data-refresh
    Published
    23 Aug 2026 · 05:22 UTC
  39. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-0510f900c5d24be817b894aa
    Change type
    data-refresh
    Published
    23 Aug 2026 · 08:29 UTC
  40. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-35924b9b522e81fe498a2dfb
    Change type
    data-refresh
    Published
    23 Aug 2026 · 18:03 UTC
  41. Deterministic refresh after a cited input, gate, or publication state changed.
    Revision
    machinev-5e1357c794ace1a5f3d423cc
    Change type
    data-refresh
    Published
    24 Aug 2026 · 02:38 UTC
Current report
report.json
Current revision
machinev-5e1357c794ace1a5f3d423cc
Source case
machine-source-80b1b871bed35fc29d24
Source revision
machine-sourcev-bd7a32196defb24a5247847f

What the machine was not allowed to use

  • analysis_mode
    value
    deterministic-machine-analysis
  • human_interviews
    value
    none
  • individual_allegations
    value
    none
  • inferred_motives
    value
    none
  • personal_data
    value
    none
  • prohibited_interpretations
    value
    ["Do not treat aggregate measurements as claims about identifiable people.", "Do not infer intent, motive, or responsibility from technical correlation.", "Do not describe this output as human reporting or a human interview."]