Political data integrity is a records problem, not an opinion problem
A political number deserves belief when a stranger can check how it was produced. Integrity is that checkability — verified origin, a process fixed before data existed, versioned computation, and disclosure recorded at release. It is not a claim that the number is right.
The definition
A definition narrow enough to be testable
Political data integrity is the property that a published figure's origin, process, computation, and disclosure are recorded and inspectable by someone who does not trust the publisher. It is a property of records, not of intentions, brands, or ideology.
This definition is deliberately narrow. It excludes accuracy, which requires an external benchmark, and it excludes representativeness, which is a sampling property. A read with perfect integrity can still be wrong — and when it is, integrity is what lets a reader see why.
The components
Four components, plus the property that makes them durable
Each component answers one question, and each fails independently. Treating them as one undifferentiated promise of trustworthiness is how the category lost credibility in the first place.
What integrity is made of
- 01
Origin integrity
Each response is attributable to a uniquely verified, eligible participant.
Origin integrity asks a narrow question: did this response come from one real, eligible person, once. It says nothing about whether that person answered thoughtfully, and nothing about whether the assembled sample resembles the population.
- 02
Process integrity
The instrument and population definition were fixed before fielding, at a version.
A read that specifies its questions, screens, and analysis intent before data exists cannot be quietly reshaped afterwards to fit a preferred finding. Process integrity is the difference between a test and a search.
- 03
Computation integrity
Weighting and estimation ran under a registered, versioned method.
The figure names the procedure that produced it. Re-running the same specification at the same version is a procedure, not a negotiation with a vendor.
- 04
Disclosure integrity
Exclusions, uncertainty, and limitations were recorded at computation time.
Disclosure is not proof. It is the condition that makes judgment possible: a reader can see what was removed, how uncertain the estimate is, and what the read does not support.
- 05Beta
Tamper evidence
A later silent change to a released record becomes detectable from outside.
Commitments are published so an outside party can detect a quiet edit after release. This is currently on a test network, and it establishes detectability — never accuracy, and never certification.
Why the category needs it
The collection assumptions stopped holding
The professional infrastructure of public polling was built when most households answered an unknown caller. That has not been true for years, and the response-rate decline is the clearest published signal of it.
Measured in %. Source: Pew Research Center, response-rate trend reporting, 1997–2023..
| Period | % |
|---|---|
| 1997 | 36 |
| 2003 | 25 |
| 2012 | 9 |
| 2018 | 6 |
| 2023 | 1 |
Response rates fell by more than an order of magnitude, which shifts the burden of a published figure from collection to disclosure.
Telephone survey response rates fell from roughly 36% in 1997 to about 1% for typical RDD polling by 2023.
SourcePew Research Center, response-rate trend reporting, 1997–2023.
Does not showThat low response rates alone make a poll wrong. Weighting can compensate for some nonresponse; it cannot be checked from a published topline.
The 2016 and 2020 US presidential polling errors prompted formal task-force reviews of methodology and weighting practice.
SourceAAPOR ad hoc committee reports on 2016 and 2020 pre-election polling.
Does not showThat any specific pollster acted in bad faith. The reviews describe measurement and coverage problems, not fraud.
Public trust in institutional information sources has declined across two decades of tracking, including for polling and media.
SourceGallup and Pew Research Center confidence-in-institutions tracking, 2000s–2020s.
Does not showThat verifiable records restore trust. Records make checking possible; whether readers check is a separate question.
Figures are reported as published by their sources, with the period stated. We do not restate other organizations' internal practices, and we do not attribute error to bad faith.
The distinction that matters
Integrity is not accuracy, and neither is disclosure
The fastest way to discredit this category again would be to let verification imply correctness. These four dimensions answer different questions, and only two of them can be proven from a record.
What each dimension settles
Accuracy
Live· available nowIs the reported number close to the truth in the population?
Estimated, with uncertainty
Statistical inference. It carries error that no method removes.
What it settlesNothing on its own. Accuracy is assessed against an external benchmark, after the fact.
What it never settlesIntegrity does not deliver accuracy. A fully verifiable record can still report an estimate that misses.
Provenance
Live· available nowWhere did this figure come from, and under what procedure?
Proven from the record
An outsider can check this without trusting us.
What it settlesThat a specific specification, participant population, exclusion summary, and method version produced this figure.
What it never settlesWhether the procedure was the right one for the inferential target.
Statistical uncertainty
Live· available nowHow much would this estimate move on a repeat measurement?
Estimated, with uncertainty
Statistical inference. It carries error that no method removes.
What it settlesA stated interval under the stated model, published with the figure.
What it never settlesCoverage gaps, nonresponse bias, and question-wording effects, which sit outside the interval.
Integrity
Live· available nowCan an outsider check that the record was not altered or filtered to taste?
Proven from the record
An outsider can check this without trusting us.
What it settlesThat admission was verified, exclusions were recorded, computation was versioned, and disclosure existed at release.
What it never settlesWhether the finding is true. Integrity removes a failure class; it does not confer correctness.
Independent review
PlannedHas an outside body audited the methods themselves?
Disclosed, not proven
Stated openly so it can be judged — disclosure is not proof.
What it settlesNothing today. We state the absence rather than implying accreditation.
What it never settlesAnything, until an external reviewer publishes findings.
The boundaryVerification establishes that a record is what it claims to be. It does not establish that the estimate inside the record is true. Any page, deck, or article that collapses those two statements is overclaiming.
Proven means an outsider can check it without trusting us. Estimated means statistical inference with irreducible error. Disclosed means stated openly so it can be judged.
The standard, concretely
What a record with integrity contains
A useful standard is checkable part by part. Each part below states what a reader can inspect and what that part does not establish, so no single artifact can be mistaken for a guarantee.
Anatomy of a record
- 01
Specification and version
Live· available nowA reader can checkThe instrument, screens, population definition, and analysis intent as frozen before fielding.
Does not proveThat the specified design was well posed or the right test of the question.
- 02
Admission summary
Live· available nowA reader can checkThat every counted response came from a uniquely verified, eligible participant.
Does not proveThat the resulting sample is representative of the target population.
- 03
Exclusion summary
Live· available nowA reader can checkHow many responses were removed, and on what recorded basis.
Does not proveThat every manipulation attempt was detected, or that exclusion rules are judgment-free.
- 04
Method version and computation run
Live· available nowA reader can checkWhich registered weighting and estimation version produced the released figure.
Does not proveThat the weighting scheme suits your inferential target.
- 05
Uncertainty and limitations
Live· available nowA reader can checkThe interval and stated limits, recorded at computation time rather than added later.
Does not proveThat the interval covers non-sampling error.
- 06
Published commitment
Beta· Base SepoliaA reader can checkThat the released record has not been silently changed since publication.
Does not proveAccuracy, and it is not certification. Test network only.
Nothing here asserts accuracy. Every part is scoped to what it makes checkable.
How to audit it
When each piece of evidence comes into existence
Integrity is not a badge applied at publication. It accumulates at specific moments, and each moment has a different audience able to check it. Some of those checks are live today; two are not.
Step through the integrity lifecycle
Before any data
When the read is specified.
- Evidence created
- Instrument, screens, population definition, and analysis intent are recorded at a fixed version.
- Who can check it
- Any reader comparing the published analysis to the recorded pre-specification.
- What it does not settle
- Recorded on the platform. It is not registration on an external pre-registration registry.
Phase 1 of 8: Before any data
Live evidence versus planned transparencyPhases one through six are live today. Published commitments run on a public test network (Base Sepolia) and establish detectability only. Independent methodological review is planned and does not exist yet — treat our methods as documented, not accredited.
Applied to publishing
What a reader can check, by publishing model
This comparison is about what is available to an outside reader, not about anyone's internal rigor. Many conventional studies are carefully executed; the point is that a reader cannot verify that from the release.
Checkability from outside
| Dimension | Conventional release | PolitiMind |
|---|---|---|
| Who answered | Described in a methods note; the reader cannot check admission. | Verified unique admission, summarised with the release.How admission works |
| When the questions were fixed | Usually undisclosed relative to fielding. | Instrument and analysis intent recorded at a version before data exists.How the read is specified |
| What was removed | Typically an unexplained final N. | Exclusion summary published with the figure. |
| Which procedure produced the number | Weighting approach described in prose, versions rarely named. | Registered method version recorded with the computation run.What a record proves |
| Whether the record changed after release | Not detectable from outside.Practice varies: Test network. Detectability only — never accuracy or certification. | Published commitment makes silent edits detectable. Beta· Base Sepolia |
| Independent audit of methods | Varies; industry initiatives exist but coverage is uneven. | None today. We state the gap instead of implying accreditation. Planned |
Citation boundary
How to cite a PolitiMind figure honestly
If you are quoting one of our reads, the statements on the left are supported by the record. The statements on the right are not, and we will ask for a correction if they appear.
What the record supports
- This figure's specification, exclusions, method version, and uncertainty are recorded and inspectable.
- Every counted response came from a uniquely verified, eligible participant.
- The reported estimate can be recomputed from the published documentation at the stated method version.
- Participant identity is separated from response data and never accompanies a delivered dataset.
What it does not support
- This poll is accurate because it is verified.
- Blockchain anchoring proves the finding is true.
- PolitiMind's methods are independently accredited.
- Integrity removes sampling error, coverage gaps, or wording effects.
- Other pollsters manipulate their data.
For a specific figure, request the methodology statement tied to that read's method version rather than reusing language from an earlier release.
What this page claims
- Methodology is stated in plain language — sampling frame, weighting approach, margin of error, and known limitations travel with every result.
- Participation requires identity verification, so responses come from confirmed real people rather than anonymous traffic.
- Participant identity and response data are held separately, so buyers receive population-level intelligence and never participant identities.
- Every study runs behind integrity controls that detect duplicate, coordinated, and low-quality participation before results are published.
- Published results carry provenance: the specification, method version, and computation run behind a number are recorded and inspectable.
- BetaResult commitments are anchored to a public ledger so the record is tamper-evident after publication.
PolitiMind claims verified admission, pre-specified instruments, recorded exclusions, versioned computation, and disclosure at release. It does not claim accuracy, probability sampling from a national frame, electoral forecasting, or independent accreditation. Tamper-evident anchoring is beta on a public test network; external methodological review is planned.
Check it before you cite it
Participation requires identity verification, so responses come from confirmed real people rather than anonymous traffic.
How admission is verifiedPublished results carry provenance: the specification, method version, and computation run behind a number are recorded and inspectable.
See published resultsMethodology is stated in plain language — sampling frame, weighting approach, margin of error, and known limitations travel with every result.
Read the methodologyResult commitments are anchored to a public ledger so the record is tamper-evident after publication.
Step through the audit timelineNext step
Judge the standard against a figure you would have to defend
The test is not whether this page reads well. Take a political figure you would have to stand behind in print, and ask what a skeptical reader could verify from its release alone. That gap is the category problem, and it is the only thing this standard exists to close.
Continue the argument
Why Polling Is Broken
Name the structural failure they have felt but not articulated
Read itProof of Integrity
Inspect how a published number can be checked
Read itMethodology
Confirm PolitiMind understands sampling, weighting, and uncertainty
Read itPublic Results
Inspect a real published result and its methodology
Read it