Skip to content

Research lead · OPEN INVESTIGATION

China's Network Filtering Has No Honest Single Rate

Testable question: Across three preregistered comparable rounds, do independent China network-measurement groups converge within twenty percentage points, or does measured interference remain selective and vantage-dependent?

A research lead comparing incompatible measurement denominators, independent vantages, and the counterevidence against treating selective filtering as a country-scale outage.

Current status: The research lead remains open while its public collection and falsification targets are incomplete.

RESEARCH LEAD · NOT A PUBLISHED INVESTIGATION. Draft claims remain under test and must not be read as findings.

Opened
Updated
Published
Not published
Version receipt
investigationv-aa5647b3f3e38f22c965c1a0
Immutable revision JSON

What is asserted—and what could overturn it

Claim wording is reproduced from the structured record. Confidence and review state are not probability scores.

Hypotheses under test

  • Comparable independent instruments converge closely enough to support one bounded national interference estimate.open · common-national-rate

    Linked falsification tests: three-round-convergence

  • Instrument scope and vantage continue to produce materially different rates even when collection windows are aligned.open · selective-vantage-dependent

    Linked falsification tests: three-round-convergence, three-round-clean-panel

  1. Claim under test · analytical finding · corroborated · draft

    The current structured round is consistent with selective and vantage-dependent interference, because independent instruments report materially different scoped measurements.

    Linked evidence receipts

    • ooni-anomaly-rate · OONI China anomaly index · support
    • censored-planet-rate · Censored Planet remote interference rate · support
    • inside-view-panel · Inside-China fixed-panel blocked share · support

    Linked counterevidence

    • no-country-scale-outage · IODA reports no current country-scale outage instrument firing, which argues against describing the scoped filtering measurements as a broad shutdown. · narrows_claim
    Claim limits.
    • The instruments sample different protocols, targets, probes, and completed-test populations. Their percentages may be compared as scoped signals, but not averaged into a national prevalence rate.
    • The present lead has not completed three preregistered comparable rounds. It cannot yet call the observed divergence persistent.
  2. Claim under test · methodological limitation · insufficient · draft

    These percentages cannot honestly be collapsed into a prevalence rate for China's population, households, or the whole internet.

    Linked evidence receipts

    • ooni-anomaly-rate · OONI China anomaly index · support
    • ooni-measurement-count · OONI accepted-measurement denominator · context
    • in-path-rate · OONI-derived completed-test middlebox index · context
    • censored-planet-rate · Censored Planet remote interference rate · support
    • inside-view-panel · Inside-China fixed-panel blocked share · support
    • inside-view-panel-size · Inside-China fixed-panel size · context
    • vantage-fusion-rate · Palimpsest derived vantage-fusion index · context

    Linked counterevidence

    • No counterevidence record is linked.
    Claim limits.
    • The instruments sample different protocols, targets, probes, and completed-test populations. Their percentages may be compared as scoped signals, but not averaged into a national prevalence rate.
    • Vantage fusion reuses upstream measurements. It is context, not another independent vote.
  3. Claim under test · artifact observation · insufficient · draft

    The current IODA record reports zero country-scale outage instruments firing; that narrows, but does not erase, the selective-interference evidence.

    Linked evidence receipts

    • ioda-instruments-firing · IODA country-scale outage instruments firing · counter

    Linked counterevidence

    • No counterevidence record is linked.
    Claim limits.
    • IODA is designed for country-scale connectivity events. It is counterevidence to a shutdown claim, not a clean bill of health for selective filtering.

Counterevidence and competing records

  • IODA reports no current country-scale outage instrument firing, which argues against describing the scoped filtering measurements as a broad shutdown.reviewed · narrows_claim · evidence ioda-instruments-firing

Falsification tests

  • A temporally overlapping multi-instrument IODA country event would falsify the current narrow statement that the round is not a broad outage.Status: untested

    Evidence needed: Aligned IODA, OONI, Censored Planet, and inside-view collection clocks.

  • Three preregistered comparable rounds in which independent normalized estimates converge within twenty percentage points would weaken the vantage-dependent hypothesis.Status: untested

    Evidence needed: Three aligned rounds with frozen panels, denominator definitions, and source-group labels.

  • Three rounds in which at least three independent source groups report no interference across the frozen test panel would falsify the current selective-interference finding for that panel.Status: untested

    Evidence needed: Three independent aggregate groups and a versioned fixed panel.

Inspect every receipt and interpretation limit

Scroll horizontally to inspect every evidence field.

Evidence receipts for this case file
Relation / classReceipt / upstream groupRecorded valueInterpretation limitProvenance / integrity
supportresearch OONI China anomaly indexooni-anomaly-rate · ooni-open-measurements 58.3Selector /signals/@id=ooni-gfw/payload/gfw_index The value is a share of accepted OONI measurements in its test mix, not a share of people, households, websites, or the internet. Source recordArtifact · · sha 6028bb16b60a · current
contextresearch OONI accepted-measurement denominatorooni-measurement-count · ooni-open-measurements 272935Selector /signals/@id=ooni-gfw/payload/n_completed_measurements Measurement count is not a probability sample of China's population or network endpoints. Source recordArtifact · · sha 6028bb16b60a · current
contextresearch OONI-derived completed-test middlebox indexin-path-rate · ooni-open-measurements 1.37Selector /signals/@id=in-path-interference/payload/middlebox_index This shares OONI provenance with the anomaly index and therefore is not independent corroboration; its denominator is completed tests only. Source recordArtifact · · sha 6028bb16b60a · current
supportresearch Censored Planet remote interference ratecensored-planet-rate · censored-planet-remote 4.27Selector /signals/@id=censored-planet/payload/cn_interference_rate_pct The remote side-channel method and denominator differ from OONI and cannot be compared as if both sampled the same population. Source recordArtifact · · sha 6028bb16b60a · current
supportresearch Inside-China fixed-panel blocked shareinside-view-panel · globalping-inside-view 1.0Selector /signals/@id=inside-view/payload/block_rate This is a deliberately selected panel and cannot estimate prevalence outside that panel. Source recordArtifact · · sha 6028bb16b60a · current
contextresearch Inside-China fixed-panel sizeinside-view-panel-size · globalping-inside-view 5Selector /signals/@id=inside-view/payload/n_censored_answered The panel size is small and deliberately scoped; it is not a national sample. Source recordArtifact · · sha 6028bb16b60a · current
counterresearch IODA country-scale outage instruments firingioda-instruments-firing · ioda-country-outages 0Selector /signals/@id=ioda-outages/payload/instruments_firing Absence of a broad routing or reachability outage does not imply absence of selective DNS, HTTP, or application filtering. Source recordArtifact · · sha 6028bb16b60a · current
contextderived Palimpsest derived vantage-fusion indexvantage-fusion-rate · palimpsest-derived-fusion 28.6Selector /signals/@id=vantage-fusion/payload/fused_index A weighted synthesis is not a new independent vantage and is not a population-prevalence estimate. Source recordArtifact · · sha 6028bb16b60a · current

Publication gate

A blocked gate keeps this work in the research-lead register regardless of how striking an individual measurement appears.

Scroll horizontally to inspect every publication check.

Structured publication-gate checks
CheckMinimumObservedResultDetail
Analytical claims presentanalytical-claims11PassedObserved 1; requires at least 1.
Independent groups per analytical claimindependent-groups23PassedObserved 3; requires at least 2.
Analytical claims with counterevidenceanalytical-counterevidence11PassedObserved 1; requires at least 1.
Analytical claims with explicit limitationsanalytical-limitations11PassedObserved 1; requires at least 1.
Analytical claims linked to hypothesesanalytical-hypotheses11PassedObserved 1; requires at least 1.
Counterevidence records reviewedcounterevidence-review11PassedObserved 1; requires at least 1.
Evidence receipts currentcurrent-evidence88PassedObserved 8; requires at least 8.
Evidence receipts integrity-verifiedverified-integrity88PassedObserved 8; requires at least 8.
Right-to-reply disposition completeright-to-reply11PassedObserved 1; requires at least 1.
Every linked falsification condition assessed with the claim survivingfalsification-assessed20Not passedObserved 0; requires at least 2.
Survived falsification conditions per analytical claimfalsification-per-claim10Not passedObserved 0; requires at least 1.
Claims editorially reviewedclaims-reviewed30Not passedObserved 0; requires at least 3.

Gate blocked. Publishable: false. Failed checks: falsification-assessed, falsification-per-claim, claims-reviewed.

Limitations and consequences

  • The instruments sample different protocols, targets, probes, and completed-test populations.Their percentages may be compared as scoped signals, but not averaged into a national prevalence rate.
  • The present lead has not completed three preregistered comparable rounds.It cannot yet call the observed divergence persistent.
  • Vantage fusion reuses upstream measurements.It is context, not another independent vote.
  • IODA is designed for country-scale connectivity events.It is counterevidence to a shutdown claim, not a clean bill of health for selective filtering.

Evidence still needed

Targets name aggregate public evidence to collect. They never identify people to target.

collecting · aggregate_only

bleedthrough-longitudinal

Does a controlled external vantage observe stable or changing DNS-injector response processes across repeated rounds?

Blocker: No validated longitudinal artifact has crossed the publication boundary; one round would not establish persistence.

Collected evidence

planned · aggregate_only

aligned-network-rounds

Do independent normalized measurements converge when their collection windows are aligned?

Blocker: Three frozen, comparable rounds are required.

Collected evidence

Methodology steps

  1. freeze-denominatorsRecord each instrument's target population, completed-test rule, panel, time window, and independence group before comparing values.Reproducible: true
  2. separate-vantagesCount shared OONI-derived products once and keep remote, inside-China, and country-outage vantages separate.Reproducible: true
  3. test-counterevidenceCheck country-scale outage telemetry before describing selective failures as broad connectivity loss.Reproducible: true

Correction, reply and safety state

Correction
none
No correction has been issued for this open research lead.
No correction timestamp
Correction policy
Right to reply
not_applicable
This open methods investigation makes no allegation about a person or institution.
0 institutions recorded
Safety
aggregate_only
Person-level data: false
Current structured record
case.json
investigation-cd196dfff93a0065d419

Right-to-reply boundary

No response is not evidence that a finding is true.

Correction boundary

Corrections append an immutable revision and preserve the previous public version.

Institutional reply register

  • No institution is recorded for reply.

Safety boundary. Public, aggregate evidence only; private contact details, volunteer identifiers and person-level records are excluded.

Prohibited interpretations

  • Do not describe the OONI anomaly index as a percentage of China's people, households, websites, or the internet.
  • Do not infer an actor, party, intent, motive, or causal mechanism from these aggregate measurements.
  • Do not describe one current round as persistent or nationally representative.

Allegations and motives

  • No allegation is made.
  • No motive is inferred.