AFOS-Analytics1 commited on
Commit
336afc6
·
verified ·
1 Parent(s): 853117c

daily mirror — 04/08/2026

Browse files
Files changed (3) hide show
  1. CHANGELOG.md +11 -0
  2. DATA_DICTIONARY.md +4 -2
  3. ERRATA.md +0 -69
CHANGELOG.md CHANGED
@@ -2,6 +2,17 @@
2
 
3
  All notable changes to this dataset. The data itself is dated and append-only; this log records **structural** changes (new files, schema, coverage).
4
 
 
 
 
 
 
 
 
 
 
 
 
5
  ## 2026-07-30, errata mechanism
6
 
7
  ### Added
 
2
 
3
  All notable changes to this dataset. The data itself is dated and append-only; this log records **structural** changes (new files, schema, coverage).
4
 
5
+ ## 2026-08-04, data-minimisation rule for the TSE registry
6
+
7
+ ### Changed
8
+ - **`polls/tse-registry.{csv,json}`: individual taxpayer numbers (CPF) are no longer carried.** Where the TSE Open Data source contains a CPF inside the free-text `methodology` or `sampling_plan`, the number is replaced by the literal marker `[CPF removido]`. This applies to the current file and to every file published from this date onward.
9
+
10
+ **Why.** The registry is published openly by the TSE under Lei 9.504/1997 art. 33; collecting it is correct and unchanged. A CPF, however, is not a field the TSE publishes: `statistician` is its own named column, and the number only ever reached the file because a pollster typed it into prose describing the sampling design. Nothing in this dataset is joined on CPF, so carrying it serves no analytical purpose. This is data minimisation, not a correction of a defect.
11
+
12
+ **What is unaffected.** The `cnpj` column and every company registration number are unchanged: a CNPJ identifies a legal entity, is published deliberately by the TSE, is filled in every row, and is the only reliable join key (legal names differ from the trade names institutes are known by). The `statistician` column is unchanged: it is professional activity and it is what lets a reader audit who signed off on a poll. Redaction is decided by **digit count**, never by label, because the TSE labels part of the free text `CNPJ/CPF` and fills it with a CNPJ.
13
+
14
+ **Scope.** Only the two long free-text columns can contain the marker. No numeric column, no date, no identifier and no row count changed.
15
+
16
  ## 2026-07-30, errata mechanism
17
 
18
  ### Added
DATA_DICTIONARY.md CHANGED
@@ -23,8 +23,8 @@ Official **TSE poll-registration registry** for the 2026 presidential cycle, bui
23
  | `conre` | string | — | Regional Statistics Council registration of the responsible statistician (`CD_CONRE`). |
24
  | `statistician` | string | — | Responsible statistician, named (`NM_ESTATISTICO_RESP`). |
25
  | `cost_brl` | number | BRL | Declared cost of the poll (`VR_PESQUISA`). |
26
- | `methodology` | string (long) | free text | **Full** methodology description (`DS_METODOLOGIA_PESQUISA`) — survey type, mode (in-person/phone), instrument. Up to ~3.8k chars; **not truncated**. |
27
- | `sampling_plan` | string (long) | free text | **Full** sampling/weighting design (`DS_PLANO_AMOSTRAL`) — universe, multi-stage cluster design, **and the demographic/geographic quota design (sex, age, education, income, region) with the exact quota percentages** when declared. Up to ~4k chars; **not truncated**. This is the registration-level weighting *design*. |
28
  | `control_system` | string (long) | free text | Internal field-control/quality-control system (`DS_SISTEMA_CONTROLE`). |
29
  | `municipality_data` | string | free text | Municipality-level breakdown declaration (`DS_DADO_MUNICIPIO`), when present. |
30
  | `uf` | string | 2-letter UF or `BR` | Federative unit (`SG_UF`). **Always `BR`** for presidential-office registrations — the TSE files all presidential polls under the national jurisdiction regardless of where the sample was drawn (see `scope`). |
@@ -32,6 +32,8 @@ Official **TSE poll-registration registry** for the 2026 presidential cycle, bui
32
  | `scope` | string | `national` \| `state` \| `unknown` | **AFOS-derived** sample-coverage label (not a native TSE field). `national` = declared universe spans more than one UF / the country; `state` = restricted to a single UF (a single municipality is within one UF, so municipal polls are `state`); `unknown` = the registration text does not declare a universe (left honest, never guessed). See provenance note below. |
33
  | `scope_source` | string | `methodology` \| `sampling_plan` \| `dado_municipio` \| `none` | Which registration field the `scope` was inferred from — for auditability/reproducibility. `none` ⇔ `scope = unknown`. |
34
 
 
 
35
  > ⚠️ **"Registered" ≠ "published".** A poll in the registry has been filed with the TSE; it may be delayed or never released. Confirm actual release against a primary source before citing numbers.
36
  >
37
  > 🔎 **What the TSE publishes vs not.** Public (in this file): methodology, sampling/weighting **design**, cost, contracting party, named statistician, fieldwork dates, sample size. **Not** in the open-data file: the per-candidate **results** and their demographic crosstabs (published by the institute, not the TSE), and the **complete questionnaire** (art. 33 VI) — that is an attachment in the TSE *PesqEle* system, not in the open-data CSV.
 
23
  | `conre` | string | — | Regional Statistics Council registration of the responsible statistician (`CD_CONRE`). |
24
  | `statistician` | string | — | Responsible statistician, named (`NM_ESTATISTICO_RESP`). |
25
  | `cost_brl` | number | BRL | Declared cost of the poll (`VR_PESQUISA`). |
26
+ | `methodology` | string (long) | free text | **Full** methodology description (`DS_METODOLOGIA_PESQUISA`) — survey type, mode (in-person/phone), instrument. Up to ~3.8k chars; **never truncated**. Sole edit: an individual taxpayer number (CPF) typed into the prose is replaced by `[CPF removido]` (see note below). |
27
+ | `sampling_plan` | string (long) | free text | **Full** sampling/weighting design (`DS_PLANO_AMOSTRAL`) — universe, multi-stage cluster design, **and the demographic/geographic quota design (sex, age, education, income, region) with the exact quota percentages** when declared. Up to ~4k chars; **never truncated**. This is the registration-level weighting *design*. Sole edit: an individual taxpayer number (CPF) typed into the prose is replaced by `[CPF removido]` (see note below). |
28
  | `control_system` | string (long) | free text | Internal field-control/quality-control system (`DS_SISTEMA_CONTROLE`). |
29
  | `municipality_data` | string | free text | Municipality-level breakdown declaration (`DS_DADO_MUNICIPIO`), when present. |
30
  | `uf` | string | 2-letter UF or `BR` | Federative unit (`SG_UF`). **Always `BR`** for presidential-office registrations — the TSE files all presidential polls under the national jurisdiction regardless of where the sample was drawn (see `scope`). |
 
32
  | `scope` | string | `national` \| `state` \| `unknown` | **AFOS-derived** sample-coverage label (not a native TSE field). `national` = declared universe spans more than one UF / the country; `state` = restricted to a single UF (a single municipality is within one UF, so municipal polls are `state`); `unknown` = the registration text does not declare a universe (left honest, never guessed). See provenance note below. |
33
  | `scope_source` | string | `methodology` \| `sampling_plan` \| `dado_municipio` \| `none` | Which registration field the `scope` was inferred from — for auditability/reproducibility. `none` ⇔ `scope = unknown`. |
34
 
35
+ > 🔒 **`[CPF removido]`, the only edited content in this file.** Where a pollster typed an individual taxpayer number (CPF) into the free-text `methodology` or `sampling_plan`, the number is replaced by that literal marker. Data minimisation: nothing here is joined on CPF, and it is not a field the TSE publishes (`statistician` is its own named column). Company registration numbers (**CNPJ**) are untouched, including inside the same free text, because a CNPJ identifies a legal entity and is the join key that makes a filing auditable. Redaction is decided by digit count, never by label. No numeric column, date, identifier or row count is affected. See CHANGELOG, 2026-08-04.
36
+ >
37
  > ⚠️ **"Registered" ≠ "published".** A poll in the registry has been filed with the TSE; it may be delayed or never released. Confirm actual release against a primary source before citing numbers.
38
  >
39
  > 🔎 **What the TSE publishes vs not.** Public (in this file): methodology, sampling/weighting **design**, cost, contracting party, named statistician, fieldwork dates, sample size. **Not** in the open-data file: the per-candidate **results** and their demographic crosstabs (published by the institute, not the TSE), and the **complete questionnaire** (art. 33 VI) — that is an attachment in the TSE *PesqEle* system, not in the open-data CSV.
ERRATA.md CHANGED
@@ -8,29 +8,6 @@ Scope note on "append-only": the file for the **current** date may be regenerate
8
 
9
  ## EN
10
 
11
- ### ERR-2026-002 · `polls/tse-registry.csv` + `.json` · personal data
12
-
13
- | | Value |
14
- |---|---|
15
- | **Affected field** | free text in `methodology` and `sampling_plan` |
16
- | **Published** | the CPF (Brazilian individual taxpayer number) of the responsible statistician, written into the prose of the filing |
17
- | **Correct** | `[CPF removido]` |
18
- | **Rows affected** | 7 occurrences, 4 distinct individuals, out of 533 filings |
19
- | **Detected** | 2026-08-04 |
20
- | **Status** | **Removed from the current file and from every file published after 2026-08-04. Revisions published before that date are preserved unchanged, deliberately** — see below. |
21
-
22
- **This is a minimisation decision, not a leak.** The registry comes from TSE Open Data (`pesquisa_eleitoral_2026_BRASIL.csv`), published openly under **Lei 9.504/1997 art. 33** and Resolução TSE 23.600/2019. Collecting it was correct and remains correct. The CPF is removed because it is **not necessary for this dataset's purpose**: nothing here is joined on CPF, and the LGPD requires processing only what the purpose needs.
23
-
24
- **Why the CPF specifically.** The TSE does not publish CPF as a field. It publishes `statistician` as its own column, by name. The CPF reached the file only because the pollster typed the number into the free-text prose of `methodology` / `sampling_plan`, a field meant to describe sampling design. It is incidental content, not deliberate publication.
25
-
26
- **What is kept, and why.** The institute **CNPJ** is kept. It is a company registration, public by law, and a column the TSE deliberately publishes and fills in 533 of 533 rows. It is also the only reliable join key: legal names differ from the names institutes are known by (`REAL TIME MIDIA LTDA` trades as *Real Time Big Data*). Removing it would destroy the auditability this dataset exists to provide. Careful reading required: the TSE labels part of the free text `CNPJ/CPF` and fills it with a CNPJ, so redacting by **label** would delete the institute identifier. The rule is **digit count**: 11 = CPF, removed; 14 = CNPJ, kept. The statistician's **name** is kept: it is professional activity, already its own column, and it is what lets a reader audit who signed off on a poll.
27
-
28
- **Why past revisions are not rewritten.** This dataset's whole claim is that the record stays exactly as distributed. Purging history would require collapsing all revisions into one, destroying the public trail of every daily mirror to remove a number the TSE itself still serves openly. The defect is declared here instead, which is what an auditable dataset does with a past defect.
29
-
30
- **Prevention.** `scripts/redigir-cpf-tse-registry.mjs` runs in the mirror workflow between the registry rebuild and the export, and a second `--check` pass fails the build if any 11-digit CPF survives.
31
-
32
- ---
33
-
34
  ### ERR-2026-001 · `data/divergence-2026-07-29.csv` · Renan Santos
35
 
36
  | | Value |
@@ -53,29 +30,6 @@ Scope note on "append-only": the file for the **current** date may be regenerate
53
 
54
  ## PT-BR
55
 
56
- ### ERR-2026-002 · `polls/tse-registry.csv` + `.json` · dado pessoal
57
-
58
- | | Valor |
59
- |---|---|
60
- | **Campo afetado** | texto livre de `methodology` e `sampling_plan` |
61
- | **Publicado** | o CPF do estatístico responsável, escrito dentro da prosa do registro |
62
- | **Correto** | `[CPF removido]` |
63
- | **Linhas afetadas** | 7 ocorrências, 4 pessoas distintas, em 533 registros |
64
- | **Detectado** | 04/08/2026 |
65
- | **Situação** | **Removido do arquivo atual e de todo arquivo publicado depois de 04/08/2026. As revisões publicadas antes dessa data ficam preservadas sem alteração, por decisão** — ver abaixo. |
66
-
67
- **Isto é minimização, não vazamento.** O registro vem do TSE Dados Abertos (`pesquisa_eleitoral_2026_BRASIL.csv`), publicado abertamente sob a **Lei 9.504/1997 art. 33** e a Resolução TSE 23.600/2019. Coletar foi correto e segue correto. O CPF sai porque **não é necessário à finalidade deste conjunto de dados**: nada aqui é cruzado por CPF, e a LGPD manda tratar só o que a finalidade exige.
68
-
69
- **Por que o CPF, especificamente.** O TSE não publica CPF como campo. Ele publica `statistician` como coluna própria, com o nome. O CPF chegou ao arquivo porque o pesquisador digitou o número dentro do texto livre de `methodology` / `sampling_plan`, campo destinado a descrever o desenho amostral. É conteúdo colateral, não publicação deliberada.
70
-
71
- **O que fica, e por quê.** O **CNPJ** do instituto fica. É registro de empresa, público por lei, e coluna que o TSE publica de propósito e preenche em 533 de 533 linhas. É também a única chave confiável de junção: o nome jurídico difere do nome pelo qual o instituto é conhecido (`REAL TIME MIDIA LTDA` atua como *Real Time Big Data*). Removê-lo destruiria a auditabilidade que justifica este conjunto de dados existir. Exige leitura cuidadosa: o TSE rotula parte do texto livre como `CNPJ/CPF` e a preenche com CNPJ, então redigir pelo **rótulo** apagaria o identificador do instituto. A régua é a **contagem de dígitos**: 11 é CPF e sai, 14 é CNPJ e fica. O **nome** do estatístico também fica: é atuação profissional, já é coluna própria, e é o que permite auditar quem assinou a pesquisa.
72
-
73
- **Por que as revisões passadas não são reescritas.** A afirmação central deste conjunto de dados é que o registro permanece exatamente como foi distribuído. Purgar o histórico exigiria colapsar todas as revisões numa só, destruindo o rastro público de cada espelhamento diário para remover um número que o próprio TSE continua servindo aberto. O defeito é declarado aqui, que é o que um conjunto de dados auditável faz com defeito passado.
74
-
75
- **Prevenção.** O `scripts/redigir-cpf-tse-registry.mjs` roda no espelhamento entre a reconstrução do registro e o export, e uma segunda passada com `--check` reprova a publicação se sobrar qualquer CPF de 11 dígitos.
76
-
77
- ---
78
-
79
  ### ERR-2026-001 · `data/divergence-2026-07-29.csv` · Renan Santos
80
 
81
  Este conjunto de dados é **datado e aditivo**. Arquivos de datas já encerradas **nunca são reescritos**, mesmo quando contêm erro conhecido. A correção é publicada aqui, para que o registro permaneça exatamente como foi distribuído e o defeito continue descobrível.
@@ -100,29 +54,6 @@ Este conjunto de dados é **datado e aditivo**. Arquivos de datas já encerradas
100
 
101
  ## ES
102
 
103
- ### ERR-2026-002 · `polls/tse-registry.csv` + `.json` · dato personal
104
-
105
- | | Valor |
106
- |---|---|
107
- | **Campo afectado** | texto libre de `methodology` y `sampling_plan` |
108
- | **Publicado** | el CPF (número de contribuyente individual brasileño) del estadístico responsable, escrito dentro de la prosa del registro |
109
- | **Correcto** | `[CPF removido]` |
110
- | **Filas afectadas** | 7 apariciones, 4 personas distintas, de 533 registros |
111
- | **Detectado** | 04/08/2026 |
112
- | **Situación** | **Eliminado del archivo actual y de todo archivo publicado después del 04/08/2026. Las revisiones publicadas antes de esa fecha quedan preservadas sin alteración, por decisión** — ver abajo. |
113
-
114
- **Esto es minimización, no una filtración.** El registro viene del TSE Datos Abiertos (`pesquisa_eleitoral_2026_BRASIL.csv`), publicado abiertamente bajo la **Lei 9.504/1997 art. 33** y la Resolução TSE 23.600/2019. Recolectarlo fue correcto y sigue siéndolo. El CPF se elimina porque **no es necesario para la finalidad de este conjunto de datos**: nada aquí se cruza por CPF, y la LGPD exige tratar solo lo que la finalidad requiere.
115
-
116
- **Por qué el CPF, específicamente.** El TSE no publica el CPF como campo. Publica `statistician` como columna propia, con el nombre. El CPF llegó al archivo solo porque el encuestador escribió el número dentro del texto libre de `methodology` / `sampling_plan`, un campo destinado a describir el diseño muestral. Es contenido colateral, no publicación deliberada.
117
-
118
- **Qué se conserva, y por qué.** El **CNPJ** del instituto se conserva. Es registro de empresa, público por ley, y una columna que el TSE publica a propósito y llena en 533 de 533 filas. Es además la única clave fiable de unión: el nombre legal difiere del nombre por el que se conoce al instituto (`REAL TIME MIDIA LTDA` opera como *Real Time Big Data*). Eliminarlo destruiría la auditabilidad que justifica la existencia de este conjunto de datos. Exige lectura cuidadosa: el TSE rotula parte del texto libre como `CNPJ/CPF` y la llena con un CNPJ, así que redactar por **rótulo** borraría el identificador del instituto. La regla es el **conteo de dígitos**: 11 es CPF y sale, 14 es CNPJ y se queda. El **nombre** del estadístico también se conserva: es actividad profesional, ya es columna propia, y es lo que permite auditar quién firmó una encuesta.
119
-
120
- **Por qué no se reescriben las revisiones pasadas.** La afirmación central de este conjunto de datos es que el registro permanece exactamente como fue distribuido. Purgar el historial exigiría colapsar todas las revisiones en una sola, destruyendo el rastro público de cada espejado diario para eliminar un número que el propio TSE sigue sirviendo abiertamente. El defecto se declara aquí, que es lo que un conjunto de datos auditable hace con un defecto pasado.
121
-
122
- **Prevención.** `scripts/redigir-cpf-tse-registry.mjs` corre en el espejado entre la reconstrucción del registro y el export, y una segunda pasada con `--check` reprueba la publicación si sobrevive algún CPF de 11 dígitos.
123
-
124
- ---
125
-
126
  ### ERR-2026-001 · `data/divergence-2026-07-29.csv` · Renan Santos
127
 
128
  Este conjunto de datos es **fechado y aditivo**. Los archivos de fechas ya cerradas **nunca se reescriben**, ni siquiera cuando contienen un error conocido. La corrección se publica aquí, para que el registro permanezca exactamente como fue distribuido y el defecto siga siendo detectable.
 
8
 
9
  ## EN
10
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
11
  ### ERR-2026-001 · `data/divergence-2026-07-29.csv` · Renan Santos
12
 
13
  | | Value |
 
30
 
31
  ## PT-BR
32
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
33
  ### ERR-2026-001 · `data/divergence-2026-07-29.csv` · Renan Santos
34
 
35
  Este conjunto de dados é **datado e aditivo**. Arquivos de datas já encerradas **nunca são reescritos**, mesmo quando contêm erro conhecido. A correção é publicada aqui, para que o registro permaneça exatamente como foi distribuído e o defeito continue descobrível.
 
54
 
55
  ## ES
56
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
57
  ### ERR-2026-001 · `data/divergence-2026-07-29.csv` · Renan Santos
58
 
59
  Este conjunto de datos es **fechado y aditivo**. Los archivos de fechas ya cerradas **nunca se reescriben**, ni siquiera cuando contienen un error conocido. La corrección se publica aquí, para que el registro permanezca exactamente como fue distribuido y el defecto siga siendo detectable.