Datasets:

apol
/

ai-election-manipulation-cases

 # Changelog
+## 0.4.0 — 2026-08-15
+- Defined the release unambiguously as a purposive, exploratory evidence index: not proof of agency transfer, not a prevalence sample, and not a causal estimate of election effects.
+- Distinguished 70 source records overall from the 56 distinct sources linked through 124 claim–evidence relations.
+- Added direct research entry points, a compact map of all 26 Hugging Face configurations, version-pinned loading recipes, and the related Agency Transfer collection.
+- Clarified that coded data are in English; Spanish is limited to the quick guide and selected metadata.
+- Separated internal structural/semantic QA from source verification, factual adjudication, sampling validity, and causal validation.
+- Documented source-preservation gaps, the external workbook runtime, mutable `main` versus a fixed `v0.4.0` revision, and the absence of a DOI.
+- Corrected stale DFRLab, Tortoise Media, and Senate Intelligence Committee URLs in the research ledgers and candidate register.
+- Added verified locator coverage to all 124 claim–source relations: 57 full-claim pinpoints, 47 partial/premise pinpoints, and 20 explicitly not verified because the cited object was inaccessible; removed unsupported edges and one unpreserved dynamic metric; replaced four blocked FCC landing pages with official FCC PDFs.
+- Marked research ledgers as transparent historical inputs rather than canonical v0.4.0 coded tables; superseded draft confidence annotations do not override generated data.
+- Aligned citation metadata with the canonical title, author, version, release date, and versioned Hugging Face revision.
 ## 0.3.1 — 2026-08-14
 - Corrected machine-readable date typing for empirical model studies in the unified catalog.

CITATION.cff CHANGED Viewed

@@ -1,15 +1,21 @@
 cff-version: 1.2.0
-message: "If you use this dataset, please cite it and the original sources for substantive claims."
 title: "AI, Elections and Agency Transfer Evidence Index"
 type: dataset
 authors:
   - given-names: "Miguel"
     family-names: "Guerrero"
     website: "https://miguelguerrero.eu"
-version: 0.3.1
-date-released: 2026-08-14
 license: CC-BY-4.0
-url: "https://huggingface.co/datasets/apol/ai-election-manipulation-cases"
 keywords:
   - elections
   - artificial intelligence

 cff-version: 1.2.0
+message: "Cite this versioned evidence index and the original sources for every substantive claim."
 title: "AI, Elections and Agency Transfer Evidence Index"
 type: dataset
+abstract: "A purposive, exploratory index of claims, evidence relations, mechanisms, selected observations, preparedness records, comparison cases, and model studies concerning AI and elections. It is not a prevalence sample, causal estimate, or empirical proof of agency transfer."
 authors:
   - given-names: "Miguel"
     family-names: "Guerrero"
     website: "https://miguelguerrero.eu"
+version: 0.4.0
+date-released: 2026-08-15
 license: CC-BY-4.0
+repository-code: "https://huggingface.co/datasets/apol/ai-election-manipulation-cases"
+url: "https://huggingface.co/datasets/apol/ai-election-manipulation-cases/tree/v0.4.0"
+identifiers:
+  - type: url
+    value: "https://huggingface.co/datasets/apol/ai-election-manipulation-cases/tree/v0.4.0"
+    description: "Versioned Hugging Face revision; record the exact commit hash used."
 keywords:
   - elections
   - artificial intelligence

CONTRIBUTING.md CHANGED Viewed

@@ -1,27 +1,47 @@
-# Contributing corrections and new cases
-This dataset accepts evidence corrections, source replacements, and proposed cases. A submission should identify the stable record ID, state one atomic correction or addition, link the strongest available source, and explain whether it changes occurrence, reach, attribution, behavioural effect, electoral effect, or institutional outcome.
-## Minimum evidence
 - Prefer an authoritative primary record plus independent corroboration.
-- Quote only the smallest passage necessary to locate the evidence; do not redistribute copyrighted articles or harmful media.
-- Separate creator, sponsor, distributor, platform, beneficiary, attributed state actor, and responding institution.
-- Report the denominator for every scale figure and add an `observations` row when the quantity is decision-relevant and suitable for comparison. Accounts, calls, completed calls, views, followers, impressions, recipients, and unique voters are not interchangeable.
-- Do not infer belief, turnout, votes, seats, or a result from reach alone.
-- Include counterevidence, denials, acquittals, reversed orders, and unresolved attribution.
 ## Promotion from `candidates`
-A candidate can enter the included corpus only after it has a stable `case_id`, one explicit `record_type`, claim-level coding, normalized evidence links, actor and technology provenance (including explicit unknowns or non-deployment context), a case–election relation, typed observations for decision-relevant comparable quantities, an agency-transfer pathway, and an explicit statement of the maximum conclusion warranted by the evidence. Boundary and null cases may remain in `candidates` or enter a labelled comparison layer when their main value is comparison rather than demonstrated harm.
-## Local validation
 ```bash
-node scripts/build-data.mjs
-node scripts/audit-v03.mjs > data/audit-v03.json
-cp data/audit-v03.json data/audit.json
-node scripts/validate-data.mjs
 ```
-A release must finish with zero blocking errors. Warnings must remain visible and receive a written disposition; missing negative-search coverage is not waived merely to produce a clean dashboard.

+# Contributing corrections and evidence
+Contributions are welcome when they make the evidence easier to inspect or a conclusion more cautious. Identify the stable record ID, propose one atomic change, provide the strongest available source and state exactly which field or inference it changes.
+Do not edit generated files in `data/`, `parquet/` or `release/` by hand. Change the declared source/seed layer, rebuild, and review the diff.
+## Minimum submission
 - Prefer an authoritative primary record plus independent corroboration.
+- Add a stable locator—page, paragraph, timestamp, table or cell—when the source permits it. An unarchived URL alone is page-level provenance.
+- Separate an underlying fact from an allegation, institutional assertion, interested-party measurement or researcher inference.
+- Separate creator, sponsor, distributor, platform, beneficiary, attributed state actor and responding institution.
+- Record denominator, unit, time window, uniqueness and method for every comparable quantity. Accounts, calls, completed calls, views, impressions, recipients and voters are not interchangeable.
+- Do not infer belief, behaviour, turnout, votes, seats or winners from content volume, reach, institutional alarm or official results.
+- Include relevant denials, counterevidence, acquittals, reversals and unresolved attribution.
+- State the maximum warranted conclusion and what evidence would weaken or falsify the proposed agency-transfer interpretation.
+Do not submit deceptive media, voter lists, targeting data, personal data, evasion guidance or operational playbooks.
 ## Promotion from `candidates`
+A lead enters the claim-coded corpus only after it has a stable `case_id`, an explicit `record_type`, atomic claims, normalized evidence links, actor and technology provenance, a case–election relation, typed observations where warranted, an explicit pathway assessment, and a bounded conclusion. Preparedness, transparent use, null and boundary cases must remain visibly labelled; inclusion does not itself establish manipulation.
+## Local checks
 ```bash
+npm ci
+npm run build:data
+npm run build:audit
+npm run build:parquet
+npm run validate
+npm run test:hardening
 ```
+These commands perform internal structural and semantic QA. Passing them does not verify source truth, settle an allegation, establish representativeness or validate a causal claim. A release must have zero blocking errors; every remaining warning requires a written disposition.
+## Release discipline
+Before publishing a numbered release:
+1. reconcile generated tables, Parquet, documentation, citation metadata and packaged artifacts;
+2. record hashes for the files included in the release manifest;
+3. create an immutable version tag and record its exact commit;
+4. test public Viewer and download links anonymously;
+5. update `CHANGELOG.md` without silently rewriting earlier versions.
+The workbook requires the external `@oai/artifact-tool` runtime and is not part of the default public npm environment. Treat CSV and Parquet as the canonical analytic artifacts. A DOI may be added only for a frozen, reconciled deposit; the project currently has none.

README.md CHANGED Viewed

@@ -2,7 +2,6 @@
 license: cc-by-4.0
 language:
   - en
-  - es
 pretty_name: AI, Elections and Agency Transfer Evidence Index
 size_categories:
   - 1K<n<10K
@@ -71,166 +70,174 @@ configs:
 # AI, Elections and Agency Transfer Evidence Index
-Version `0.3.1` is an auditable research corpus about AI-enabled election influence, political synthetic media, algorithmic distribution, coordinated networks, and the transfer of informational or representational control. The research cutoff is **12 August 2026**.
-The package is internally complete for the records and tables it declares. It is **not** a comprehensive catalogue of elections, a global prevalence sample, or a causal estimate of votes changed.
-## Browse and download
-- [Research view](https://huggingface.co/datasets/apol/ai-election-manipulation-cases/viewer/research_view/data) — the shortest usable entry point.
-- [Complete v0.3.1 package](https://huggingface.co/datasets/apol/ai-election-manipulation-cases/resolve/main/release/agency-transfer-election-cases-v0.3.1.zip?download=true) — CSV, Parquet, documentation, workbook, research ledgers, and validation scripts.
-- [Excel research workbook](https://huggingface.co/datasets/apol/ai-election-manipulation-cases/resolve/main/release/agency-transfer-election-evidence-index-v0.3.1.xlsx?download=true) — 24 research and QA sheets.
-- [Deterministic audit](https://huggingface.co/datasets/apol/ai-election-manipulation-cases/blob/main/data/audit-v03.json) — coverage, integrity, errors, and warnings.
-- [Citation metadata](https://huggingface.co/datasets/apol/ai-election-manipulation-cases/blob/main/CITATION.cff) — cite this release and the original evidence sources used for substantive claims.
 ## Start here
-1. `research_view.csv` — the shortest usable table: one row per agency-transfer pathway.
-2. `case_catalog.csv` — separates claim-coded records, screening leads, and empirical model studies.
-3. `pathways.csv` and `observations.csv` — structured mechanisms and typed quantities.
-4. `claims.csv` and `claim_evidence.csv` — the evidentiary basis for factual use.
-5. `official_election_metrics.csv` and `official_results.csv` — electoral context only.
-On Hugging Face, Data Studio reads the validated native files in `parquet/`; source CSV tables remain in `data/`. The Excel workbook and ZIP are in `release/`. Inside the ZIP, the workbook is also available under `workbook/`.
-## Release at a glance
-| Layer | Rows | Correct interpretation |
-|---|---:|---|
-| Claim-coded records | 10 | 2 election-wide cases, 2 observed networks, 3 bounded incidents, 1 campaign set, and 2 preparedness files |
-| Incident-count eligible | 8 | Excludes both preparedness files; still requires filtering by `record_type` |
-| Atomic claims | 96 | Occurrence, attribution, mechanism, reach, effects, response, counterevidence, and agency transfer are separate |
-| Claim-evidence sources | 70 | Primary records, official decisions, peer-reviewed research, technical reports, and independent reporting |
-| Official data sources | 31 | Electoral files and legal/administrative records kept in a separate manifest |
-| Screening leads | 47 | Comparative research queue; not included incidents until claim-coded |
-| Empirical model studies | 7 | Audits and experiments; never incident counts |
-| Agency-transfer pathways | 10 | One structured assessment per claim-coded record |
-| Typed observations | 24 | Selected decision-relevant metrics retain unit, denominator, time window, uniqueness, method, and causal status |
-| Official contests/bundles | 11 | Completed, scheduled, and conditional election contexts |
-| Official long-form metrics | 58 | Includes Moldovan protocol fields and legal numerator definitions |
-| Official result rows | 68 | Full national fields for Romania and Moldova; New Hampshire remains explicitly partial |
-Automated QA reports **0 blocking errors** and **6 explicit warnings**. Every warning concerns a regional frame without a systematic negative/null search; the warnings flag that prevalence inference is unsupported.
-## Why v0.3.x is methodologically different
-### Comparable units
-`record_type` prevents an election-wide institutional file, a distribution network, a single robocall, a campaign ad set, and preparedness monitoring from being treated as the same unit. `incident_count_eligible=false` excludes preparedness and empirical studies from incident counts.
-### Structured agency transfer
-The central analytic sequence is:
-`capability → controller → influence vector → target → affected agency → control shift → power recipient → harm pathway → observed outcome`
-`pathways.csv` records each link and its epistemic status. It does not create an arbitrary agency-transfer score.
-### Typed quantitative evidence
-`observations.csv` does not collapse accounts, posts, views, followers, calls, questions, users, voters, or votes. It normalizes selected decision-relevant quantities with denominator, window, uniqueness, method, source, and causal status. Numeric text can still appear only in claims when it is not suitable for comparison; absence from `observations` is not evidence that no number exists elsewhere.
-### Capability evidence is separate from deployment
-`model_evaluations.csv` includes point-in-time election-information audits, multilingual comparisons, voting-advice tests, open-weight Mistral/Mixtral evaluation, and randomized conversational-persuasion experiments. Experimental effects are causal only within the recorded design and outcome; they are not evidence of campaign deployment, actual ballots, or an election result.
-### A declared screening frame
-`sampling_frame.csv` records regional search status, languages, channels, counts, negative-search status, and known gaps. The candidate queue improves diversity and discovery, but does not turn purposive collection into representative sampling.
-## Important official-data corrections
-- **Moldova 2024:** the election protocol distinguishes voters receiving ballots (field `C`) from voters participating (field `D`). The Constitutional Court used `C/(A+B)` for the legal validity threshold in the presidential rounds and referendum. Both values are preserved. The 2025 parliamentary authority-reported rate instead uses `D/(A+B)`.
-- **Moldova 2025 parliament:** all 23 registered competitors are represented; 22 have valid votes, Moldova Mare is explicitly coded as excluded/ballots invalidated, and official seats are included.
-- **Romania:** the 2024 first-round tally was first confirmed and validated, then the entire process was voided. Tally status and legal-process chronology remain separate. The 2024 and 2025 first-round national candidate fields are complete.
-- **New Hampshire 2024:** 125,811 Democratic ballots and 123,996 presidential-contest votes are separate metrics. The 1,815 difference is not relabelled as invalid or undervote without an official classification. The all-party checklist is not used as a Democratic-primary turnout denominator.
-- **Brazil 2026:** the first date is an election bundle; a possible second round applies only to president and governors. Candidate data before the filing deadline are incomplete snapshots, not ballot qualification.
-- **United States 2026:** the national date bundles separate House, Senate, delegate, and special contests administered and certified by states. FEC finance/calendar files are not a national result or ballot-access authority.
-Official outcomes are descriptive context. They do not identify exposure, persuasion, or votes caused by manipulation.
-## Quick start
-```python
-from datasets import load_dataset
-view = load_dataset(
-    "apol/ai-election-manipulation-cases",
-    "research_view",
-    split="data",
-)
-pathways = load_dataset(
-    "apol/ai-election-manipulation-cases",
-    "pathways",
-    split="data",
-)
-observations = load_dataset(
-    "apol/ai-election-manipulation-cases",
-    "observations",
-    split="data",
-)
-```
-For local relational use, `data/datapackage.json` supplies field types, primary keys, and foreign keys. CSV and Parquet versions are available directly on the Hub and in the packaged release.
-## Join map
-- `claims.case_id → cases.case_id`
-- `claim_evidence.claim_id → claims.claim_id`
-- `claim_evidence.source_id → sources.source_id`
-- `pathways.case_id → cases.case_id`
-- `observations.pathway_id → pathways.pathway_id`
-- `observations.claim_id → claims.claim_id`
-- `analytic_record_claims` normalizes claims used by actors, technologies, content items, pathways, and events
-- `case_elections` normalizes case-to-election relations
-- `election_sources` normalizes election-to-official-source relations
-- `official_election_metrics.election_id → official_elections.election_id`
-Pipe-delimited fields remain only for backward compatibility or compact display. Prefer the normalized join tables for analysis.
-## Evidence semantics
-- `established_evidence`: the claim as worded is supported by a primary record or strong corroboration.
-- `established_as_campaign_report`: the dataset establishes that the campaign reported the measurement; it does not independently validate the underlying population quantity.
-- `strong_inference`: evidence supports the inference, but the link is not directly observed or finally adjudicated.
-- `open_question`: uncertainty is recorded rather than filled with an assumption.
-`coder_confidence_in_label` is confidence in the coding decision, not a probability that the claim is true. Use `claim_status`, `evidence_strength`, counterevidence, and the underlying source relationship.
-This release has one primary coder and no reported intercoder-reliability estimate. See `docs/methodology.md`; the numeric confidence field is legacy self-assessment metadata, not a calibrated scientific score.
-## Source preservation and locators
-The release retains public source URLs but does not redistribute source snapshots. Consequently, `archive_url`, `content_hash`, official-source `content_sha256`, `content_bytes`, and `source_schema_version` remain blank unless the exact referenced object was captured and verified. A blank is an explicit preservation gap, not proof that the public record is unavailable.
-Likewise, a blank `claim_evidence.locator` means the relationship is documented at page or document level; it must not be read as paragraph-level pinpoint verification. Researchers needing quote-level reproducibility should capture a lawful snapshot, record its hash, and add a stable page, paragraph, timestamp, table, or cell locator in a future release.
-## Suitable and unsuitable uses
-Suitable uses include comparative case research, evidence tracing, institutional-response analysis, model-capability comparison, official-context reconciliation, and hypothesis generation.
-Do not use this release as:
-- a global or regional prevalence estimate;
-- a causal estimate of votes, turnout, seats, or winners changed;
-- a single list of proven malicious “AI incidents”;
-- evidence that transparent political AI is necessarily manipulative;
-- training material for voter targeting, impersonation, deceptive content, or safeguard evasion;
-- a final account of elections or proceedings still ongoing at the cutoff.
-## Reproducibility
 ```bash
-node scripts/build-data.mjs
-node scripts/audit-v03.mjs > data/audit-v03.json
-cp data/audit-v03.json data/audit.json
-node scripts/validate-data.mjs
-npm ci
-node scripts/build-parquet.mjs
 ```
-`data/stats.json` records input and output SHA-256 hashes. `data/audit-v03.json` is deterministic, covers all 26 tables, and exits non-zero only for blocking errors. `data/schema.json` preserves column order and `data/datapackage.json` adds machine-readable types and relations. `package-lock.json` pins Parquet generation; `requirements-validation.txt` pins optional independent Frictionless validation.
-The dataset license covers the original coding and synthesis. Source materials remain subject to the terms of their publishers and issuing institutions. Cite both this dataset and the original sources for substantive claims.

 license: cc-by-4.0
 language:
   - en
 pretty_name: AI, Elections and Agency Transfer Evidence Index
 size_categories:
   - 1K<n<10K
 # AI, Elections and Agency Transfer Evidence Index
+**Version 0.4.0 · released 15 August 2026 · research cutoff 12 August 2026**
+This is a **purposive, exploratory evidence index**. It makes claims, sources, counterevidence, uncertainty, mechanisms, and selected measurements easier to inspect. It does **not** empirically prove the agency-transfer hypothesis, estimate how common election manipulation is, or identify votes or election outcomes caused by AI.
+The repository slug is inherited. The canonical title is the one above: the index also contains preparedness files, boundary comparisons, screening leads, official electoral context, and empirical model studies. Inclusion is not a finding that every record is manipulative or harmful.
 ## Start here
+- [Open the 10-row research view](https://huggingface.co/datasets/apol/ai-election-manipulation-cases/viewer/research_view/data) — fastest orientation.
+- [Read the Spanish quick guide](https://huggingface.co/datasets/apol/ai-election-manipulation-cases/blob/main/docs/guia-rapida-es.md) — concise interpretation guide.
+- [Read the methodology](https://huggingface.co/datasets/apol/ai-election-manipulation-cases/blob/main/docs/methodology.md) — units, coding and limitations.
+- [Read the v0.4.0 red-team report](https://huggingface.co/datasets/apol/ai-election-manipulation-cases/blob/main/docs/red-team-v0.4.0.md) — findings, fixes and residual limits.
+- [Inspect the claim–evidence table](https://huggingface.co/datasets/apol/ai-election-manipulation-cases/viewer/claim_evidence/data) — trace claims to source records.
+- [Download the complete v0.4.0 package](https://huggingface.co/datasets/apol/ai-election-manipulation-cases/resolve/v0.4.0/release/agency-transfer-election-cases-v0.4.0.zip?download=true) — CSV, Parquet, documentation, ledgers, scripts and workbook.
+- [Download the v0.4.0 Excel workbook](https://huggingface.co/datasets/apol/ai-election-manipulation-cases/resolve/v0.4.0/release/agency-transfer-election-evidence-index-v0.4.0.xlsx?download=true) — navigation and QA convenience artifact.
+- [Verify release checksums](https://huggingface.co/datasets/apol/ai-election-manipulation-cases/blob/v0.4.0/release/SHA256SUMS) — hashes for the frozen ZIP and workbook.
+- [Browse the Agency Transfer collection](https://huggingface.co/collections/apol/agency-transfer) — related research datasets.
+- [See citation metadata](https://huggingface.co/datasets/apol/ai-election-manipulation-cases/blob/main/CITATION.cff) — cite this index and the original sources.
+The coded tables and field values are in English. Spanish is limited to the quick guide and selected descriptive metadata; this is not a bilingual corpus.
+## What is in v0.4.0
+These counts describe the generated v0.4.0 build and must be refreshed if that build changes.
+| Layer | Rows | What the count means |
+|---|---:|---|
+| Claim-coded records | 10 | Mixed units: 2 election-wide files, 2 networks, 3 incidents, 1 campaign set and 2 preparedness files |
+| Incident-count eligible | 8 | Excludes preparedness; it is not a prevalence denominator |
+| Atomic claims | 100 | Occurrence, attribution, mechanism, reach, effects, response and inference remain separate |
+| Claim–evidence relations | 124 | `supports`, `premise` or `counterevidence` links, not 124 independent findings |
+| Sources linked to claims | 56 | Distinct source records used by `claim_evidence` |
+| Source records overall | 70 | Also includes records used elsewhere in the corpus |
+| Screening leads | 47 | Research queue; not included incidents |
+| Empirical model studies | 7 | Laboratory/audit evidence; not real-world deployments |
+| Pathway assessments | 10 | Researcher-coded analytic syntheses, not measured outcomes |
+| Typed observations | 22 | Selected quantities with unit, denominator, window, method and causal status |
+| Official data sources | 31 | Separate electoral and administrative provenance layer |
+The strongest pathway conclusion is deliberately limited: 4 records show an observed control mechanism, 2 show an observed attempted transfer, 2 remain insufficient, and 2 are preparedness records rather than incidents. **No record codes observed agency change or observed agency preservation/extension.** These categories describe evidence ceilings, not severity scores.
+## The 26 configurations
+| Use | Configurations |
+|---|---|
+| Orient and scope | `research_view`, `case_catalog`, `coverage_summary`, `sampling_frame` |
+| Claims and evidence | `cases`, `claims`, `sources`, `claim_evidence`, `analytic_record_claims` |
+| Mechanisms and provenance | `pathways`, `observations`, `events`, `case_sources`, `case_actors`, `technology_uses`, `content_items` |
+| Election relations and context | `case_elections`, `election_sources`, `official_elections`, `official_election_metrics`, `official_turnout`, `official_results`, `official_data_sources` |
+| Discovery | `candidates`, `watchlist` |
+| Empirical capability evidence | `model_evaluations` |
+Hugging Face Data Studio reads the native Parquet files in `parquet/`. Equivalent source tables are in `data/`; relational types, keys and foreign keys are declared in [`data/datapackage.json`](https://huggingface.co/datasets/apol/ai-election-manipulation-cases/blob/main/data/datapackage.json).
+Files under `research/` are transparent build inputs and historical working ledgers, not canonical v0.4.0 tables. Some retain superseded draft annotations; see [`research/README.md`](https://huggingface.co/datasets/apol/ai-election-manipulation-cases/blob/main/research/README.md) before using them.
+## Load the data
+### Hugging Face Datasets
+```python
+from datasets import load_dataset
+view = load_dataset(
+    "apol/ai-election-manipulation-cases",
+    "research_view",
+    revision="v0.4.0",
+    split="data",
+)
+```
+### pandas
+```python
+import pandas as pd
+url = "https://huggingface.co/datasets/apol/ai-election-manipulation-cases/resolve/v0.4.0/parquet/research_view.parquet"
+view = pd.read_parquet(url)
+```
+### Polars
+```python
+import polars as pl
+url = "https://huggingface.co/datasets/apol/ai-election-manipulation-cases/resolve/v0.4.0/parquet/research_view.parquet"
+view = pl.read_parquet(url)
+```
+Remove the pinned revision only if you deliberately want the mutable latest version.
+## Read the evidence correctly
+Each pathway keeps these fields separate: capability, controller, influence vector, target, affected agency, **hypothesized** control shift, possible power recipient, potential harm, and observed outcome. Their presence in one row does not imply that one caused the next.
+`pathways` makes that reasoning inspectable. It does not establish that every link occurred, and it does not produce a validated agency-transfer score. In particular:
+- `strong_inference` is a researcher inference, not a directly observed effect;
+- platform reach is not belief, behaviour, turnout or vote choice;
+- an institutional response is not proof of behavioural or electoral harm;
+- official election results are context, not causal evidence;
+- transparent, authorised, defensive and preparedness uses are not automatically manipulation.
+Legacy numeric confidence fields were removed because they were neither calibrated probabilities nor intercoder-reliability estimates. This release has one primary coder and reports no intercoder-reliability statistic.
+## Evidence provenance
+Use normalized joins for factual work:
+```text
+claims.case_id                 → cases.case_id
+claim_evidence.claim_id        → claims.claim_id
+claim_evidence.source_id       → sources.source_id
+observations.claim_id          → claims.claim_id
+pathways.case_id               → cases.case_id
+case_elections.election_id     → official_elections.election_id
+election_sources.official_source_id → official_data_sources.official_source_id
+```
+The index retains public URLs but does not redistribute source snapshots. Of 124 claim–source relations, 57 have a verified pinpoint covering the full claim, 47 have a verified pinpoint covering only a factual premise or part of the claim, and 20 remain `not_verified` because the cited object was inaccessible during the locator audit. `locator_coverage` preserves that distinction and `locator_note` explains every unresolved pinpoint; a nonblank pinpoint is not a frozen source object, and a partial locator must not be treated as support for the whole claim. All archive and content-hash fields remain unresolved preservation gaps. Cite and inspect the underlying source before relying on a substantive claim.
+## What the audit does—and does not do
+The automated audit is **internal structural and semantic QA**. It checks declared files, identifiers, relations, enums, date formats, selected consistency rules and recorded hashes within its stated scope. It does not establish that a source is still live, that a source supports every word of a claim, that an allegation is true, that the sample is representative, or that a causal interpretation is valid.
+The v0.4 auditor retains the historical filename `scripts/audit-v03.mjs` and compatibility output `data/audit-v03.json`; its internal auditor identifier is `audit-v04`, and `data/audit.json` is byte-identical. The old entry point `scripts/audit-data.mjs` is a read-only wrapper.
+Six visible warnings record regional frames without systematic negative/null searches. They are a reason not to infer prevalence, not evidence that no cases exist in those regions.
+## Appropriate use
+Use this index for evidence tracing, comparative case research, hypothesis generation, institutional-response analysis, source discovery, measurement reconciliation and bounded model-study comparison.
+Do not use it as:
+- proof of the agency-transfer thesis;
+- a global or regional incident count or prevalence estimate;
+- a causal estimate of votes, turnout, seats or winners changed;
+- a list of proven malicious AI incidents;
+- evidence that visible or authorised political AI is inherently manipulative;
+- operational material for voter targeting, impersonation, deception or safeguard evasion.
+## Builds, releases and citation
+The supported tabular pipeline is:
+```bash
+npm ci
+npm run build:data
+npm run build:audit
+npm run build:parquet
+npm run validate
+npm run test:hardening
+```
+After the externally generated workbook is frozen in `release/`, maintainers run:
 ```bash
+npm run build:package
+npm run verify:package
+npm run build:release-manifest
+npm run validate
 ```
+`npm run audit:links` is a separate, time-dependent network check; it is not evidence that a blocked or rate-limited page is absent. Within the declared runtime and inputs, the deterministic scripts rebuild and check the tabular and packaged artifacts they cover. That remains narrower than end-to-end source reproducibility: external pages are not frozen, 47 pinpoints cover only a factual premise or part of their claim, 20 relations have no verified pinpoint, and automated QA does not validate source truth.
+The Excel workbook is a convenience artifact, not the canonical analytic source. Its builder requires the `@oai/artifact-tool` runtime, which is not part of the public `package.json`; use CSV or Parquet unless that environment is available. The ZIP and workbook should be verified against the release manifest, not assumed equivalent merely because their filenames contain the same version.
+`main` is mutable. Use the `v0.4.0` revision for a fixed citation and record the exact commit hash used. No DOI has been assigned; none should be inferred from the repository URL. A DOI can be added only after a frozen revision is deposited and its identity is reconciled with the tag, citation metadata and release manifest.
+The CC BY 4.0 license covers the original coding and synthesis. Source materials remain under their publishers' or issuing institutions' terms. Cite both this index and the original evidence sources for substantive claims.

package-lock.json CHANGED Viewed

@@ -1,12 +1,12 @@
 {
   "name": "agency-transfer-election-cases",
-  "version": "0.3.1",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "agency-transfer-election-cases",
-      "version": "0.3.1",
       "dependencies": {
         "parquetjs-lite": "0.8.7"
       },

 {
   "name": "agency-transfer-election-cases",
+  "version": "0.4.0",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "agency-transfer-election-cases",
+      "version": "0.4.0",
       "dependencies": {
         "parquetjs-lite": "0.8.7"
       },

package.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "name": "agency-transfer-election-cases",
-  "version": "0.3.1",
   "private": true,
   "type": "module",
   "engines": {
@@ -9,8 +9,15 @@
   "scripts": {
     "build:data": "node scripts/build-data.mjs",
     "audit": "node scripts/audit-v03.mjs",
     "validate": "node scripts/validate-data.mjs",
-    "build:parquet": "node scripts/build-parquet.mjs"
   },
   "dependencies": {
     "parquetjs-lite": "0.8.7"

 {
   "name": "agency-transfer-election-cases",
+  "version": "0.4.0",
   "private": true,
   "type": "module",
   "engines": {
   "scripts": {
     "build:data": "node scripts/build-data.mjs",
     "audit": "node scripts/audit-v03.mjs",
+    "build:audit": "node scripts/build-audit.mjs",
     "validate": "node scripts/validate-data.mjs",
+    "validate:source": "node scripts/validate-data.mjs --source-only",
+    "build:parquet": "node scripts/build-parquet.mjs",
+    "build:package": "node scripts/build-package.mjs",
+    "verify:package": "node scripts/verify-package.mjs",
+    "build:release-manifest": "node scripts/build-release-manifest.mjs",
+    "audit:links": "node scripts/audit-links.mjs",
+    "test:hardening": "node scripts/test-hardening.mjs"
   },
   "dependencies": {
     "parquetjs-lite": "0.8.7"