Add files using upload-large-folder tool

README.md

@@ -28,16 +28,17 @@ security and security-relevant venues, with 13,987 abstracts and a BibTeX entry
 every record.
 
 - **Code / tool:** <https://github.com/sidneibarbieri/topvenues-tool>
-- **Current tool release:** [`v1.4.0`](https://github.com/sidneibarbieri/topvenues-tool/releases/tag/v1.4.0)
+- **Seven-minute demonstration:** [MP4](assets/demo/topvenues-sf-demo-v1.5.0.mp4)
+  with [Brazilian Portuguese](assets/demo/topvenues-sf-demo-v1.5.0.pt-BR.vtt)
+  and [English](assets/demo/topvenues-sf-demo-v1.5.0.en.vtt) captions.
 - **Pinned source of truth:** the gzipped SQLite snapshot shipped with the
   tool; this dataset is a faithful Parquet export of the same named snapshot.
 
 ## Release identity
 
-- **Profile:** `security-20-v3`
-- **Snapshot SHA-256 (gzip):** `3fd34ac5cb67e528ad9789f52e1a1d2302c51f5e37044bc7d688af976ac36ff5`
-- **Snapshot source release tag:** `v1.3.0` (the corpus bytes are unchanged in
-  tool release `v1.4.0`)
+- **Profile:** `security-20-v4`
+- **Snapshot SHA-256 (gzip):** `bcb762c1c9b1f8ce6f075a8c1a23d68310caec853b0cc8ce3f42931e43c370c5`
+- **Source release tag:** `v1.5.0`
 - **Snapshot record count:** 14,859
 - **Identity rule:** exact canonical DOI or stable landing page, plus Crossref-confirmed DOI aliases
 - **Same-metadata groups retained:** 4
@@ -59,7 +60,7 @@ security = corpus.filter(lambda paper: paper["area"] == "security")
 
 This archived capture demonstrates the **Abstract contains** workflow on the
 accepted paper's frozen `security-20` profile; it is interface evidence, not a
-count claim for `security-20-v3`. Open the
+count claim for `security-20-v4`. Open the
 [high-resolution PDF](assets/topvenues-abstract-search.pdf) for close inspection.
 
 ## Scope and curation
@@ -135,11 +136,7 @@ dataset card does not grant rights in third-party abstract text.
   author = {Barbieri, Sidnei and Ferraz, Agney Lopes Roth and
             Pereira J{\'u}nior, Louren{\c{c}}o Alves},
   year   = {2026},
-  version = {v1.4.0},
-  url    = {https://github.com/sidneibarbieri/topvenues-tool/releases/tag/v1.4.0}
+  version = {v1.5.0},
+  url    = {https://github.com/sidneibarbieri/topvenues-tool/releases/tag/v1.5.0}
 }
 ```
-
-The dataset revision remains `security-20-v3`; `v1.4.0` advances the research
-interface, watchlist and preprint-monitoring workflows, manual-audit tooling,
-and reviewer packaging without rewriting the frozen corpus.

assets/demo/topvenues-sf-demo-v1.5.0.en.vtt

@@ -0,0 +1,181 @@
+WEBVTT
+
+00:00:00.000 --> 00:00:03.324
+A literature review can fail before screening begins.
+
+00:00:03.324 --> 00:00:12.466
+If the population of eligible papers comes from changing portals, undocumented searches, or private exports, another researcher cannot reconstruct the same denominator.
+
+00:00:12.466 --> 00:00:24.100
+TopVenues turns that first step into an executable artifact: a declared cybersecurity corpus, a fixed local snapshot, and tools for inspecting every aggregate back down to individual records.
+
+00:00:30.000 --> 00:00:38.804
+This August 2026 release contains fourteen thousand eight hundred fifty-nine records from twenty declared venues, covering 2019 through 2026.
+
+00:00:38.804 --> 00:00:44.365
+Ninety-four point one percent have searchable abstracts, and every record has BibTeX.
+
+00:00:44.365 --> 00:00:46.219
+The snapshot is immutable.
+
+00:00:46.219 --> 00:00:55.950
+After installation, its manifest, database, tests, search, export, and web health check can all be verified offline with one reproduction command.
+
+00:01:00.000 --> 00:01:04.547
+Start in Search with a research question, not with a ranking.
+
+00:01:04.547 --> 00:01:09.507
+Here I search for L L M security using full-text BM25 ranking.
+
+00:01:09.507 --> 00:01:12.400
+The result is not a recommendation list.
+
+00:01:12.400 --> 00:01:20.667
+It is an inspectable candidate set: title, authors, venue, year, class, abstract preview, source link, and citation record remain visible.
+
+00:01:20.667 --> 00:01:24.800
+Substring filters are also available when exact textual mention matters.
+
+00:01:30.000 --> 00:01:40.620
+For reference mapping, I can restrict the same question to the security top four: ACM CCS, IEEE Security and Privacy, USENIX Security, and NDSS.
+
+00:01:40.620 --> 00:01:43.718
+That returns seventy-eight records in this snapshot.
+
+00:01:43.718 --> 00:01:50.355
+A systematic review should normally begin with all venues unless its protocol justifies a restriction.
+
+00:01:50.355 --> 00:01:56.550
+Either way, the exact subset can be exported as CSV, JSON, or ready-to-cite BibTeX.
+
+00:02:00.000 --> 00:02:03.957
+Insights provides distributions, but the charts are not decorative.
+
+00:02:03.957 --> 00:02:08.793
+Every venue, year, class, and coverage bar is a navigation control.
+
+00:02:08.793 --> 00:02:14.509
+Clicking a bar transfers its value into Search and opens the supporting records.
+
+00:02:14.509 --> 00:02:25.500
+That drill-down is important: a researcher can challenge an aggregate immediately, inspect outliers, and export the evidence instead of trusting a dashboard number without provenance.
+
+00:02:30.000 --> 00:02:34.496
+The topic view separates raw volume from share of each year's corpus.
+
+00:02:34.496 --> 00:02:43.112
+For L L M in the top four, it finds two hundred eighty-two papers, led by CCS, USENIX, NDSS, and Security and Privacy.
+
+00:02:43.112 --> 00:02:47.233
+The share curve prevents corpus growth from masquerading as topic growth.
+
+00:02:47.233 --> 00:02:55.850
+And 2026 is visibly marked as partial, so this chart supports monitoring without pretending an incomplete year is comparable to a completed one.
+
+00:03:00.000 --> 00:03:02.439
+Researcher Radar answers three different questions.
+
+00:03:02.439 --> 00:03:05.285
+Any-author frequency shows recurring names across collaborations.
+
+00:03:05.285 --> 00:03:09.351
+First-author frequency emphasizes who most often leads the author list.
+
+00:03:09.351 --> 00:03:18.702
+Last-author frequency can be useful in fields where that position often reflects group leadership, but the tool never assumes that convention is universal.
+
+00:03:18.702 --> 00:03:24.800
+Paper count is the transparent default; optional tier weighting is clearly labeled as a heuristic.
+
+00:03:30.000 --> 00:03:39.684
+Selecting a name opens the evidence behind the row: annual publication trajectory, first- and last-author positions, direct coauthors, active years, and main venues.
+
+00:03:39.684 --> 00:03:43.895
+DBLP identity suffixes are preserved to reduce accidental homonym merges.
+
+00:03:43.895 --> 00:03:49.789
+These views reveal whom to read and monitor within a declared topic and scope.
+
+00:03:49.789 --> 00:03:54.000
+They do not measure citations, seniority, authority, or scientific quality.
+
+00:04:00.000 --> 00:04:08.071
+For ongoing surveillance, download a portable watchlist containing the topic, authors, tier scope, and paper IDs already observed.
+
+00:04:08.071 --> 00:04:12.107
+A later snapshot can compute deterministic additions and removals.
+
+00:04:12.107 --> 00:04:19.282
+The emerging-activity table compares recent and earlier publication rates as a descriptive signal, not a forecast.
+
+00:04:19.282 --> 00:04:28.250
+The arXiv handoff searches an author's name for possible preprints; every match remains an unverified candidate until identity is checked.
+
+00:04:30.000 --> 00:04:32.622
+Evidence makes the claim boundary explicit.
+
+00:04:32.622 --> 00:04:37.867
+Sidnei Barbieri manually reviewed a deterministic, venue-stratified sample of two hundred records.
+
+00:04:37.867 --> 00:04:44.860
+One hundred sixty-nine abstracts satisfied all three criteria: complete, uncontaminated, and matched to the sampled paper.
+
+00:04:44.860 --> 00:04:55.350
+That is an eighty-four point five percent usable rate, with a ninety-five percent Wilson interval from seventy-eight point eight to eighty-eight point nine percent.
+
+00:05:00.000 --> 00:05:12.466
+The audit was performed on the previous snapshot and transferred only after an exact check showed the same fourteen thousand eight hundred fifty-nine paper IDs and identical abstract text.
+
+00:05:12.466 --> 00:05:16.334
+Version four changes ten titles, none in the sample.
+
+00:05:16.334 --> 00:05:21.922
+Coverage is still uneven: for example, ESORICS and WOOT have substantial missing abstracts.
+
+00:05:21.922 --> 00:05:28.800
+Missing text is disclosed and excluded from abstract-dependent conclusions; it is never treated as negative evidence.
+
+00:05:30.000 --> 00:05:32.645
+Dataset lifecycle separates exploration from maintenance.
+
+00:05:32.645 --> 00:05:36.171
+A released profile cannot be refreshed in place.
+
+00:05:36.171 --> 00:05:47.631
+New collection from live APIs must create a successor profile, record provenance, compare added and removed IDs, rerun quality gates, and receive a new semantic release.
+
+00:05:47.631 --> 00:05:52.479
+On Linux and macOS, reproduce dot sh exercises the reviewer path.
+
+00:05:52.479 --> 00:05:58.650
+Native Windows uses reproduce dot P S one and closes every SQLite handle deterministically.
+
+00:06:00.000 --> 00:06:02.483
+A productive workflow is therefore simple.
+
+00:06:02.483 --> 00:06:05.380
+Begin broad when constructing a systematic-review denominator.
+
+00:06:05.380 --> 00:06:09.518
+Narrow to the top four when mapping canonical security conversations.
+
+00:06:09.518 --> 00:06:13.243
+Inspect the records behind every trend or author signal.
+
+00:06:13.243 --> 00:06:16.554
+Export the candidate set with its snapshot identity.
+
+00:06:16.554 --> 00:06:19.451
+Save a watchlist for the next release.
+
+00:06:19.451 --> 00:06:26.900
+And document each restriction as a protocol choice, because TopVenues supports scientific judgment; it does not replace it.
+
+00:06:30.000 --> 00:06:41.346
+TopVenues gives researchers three things that ordinary search portals do not provide together: a reproducible denominator, an inspectable research workbench, and a forward-looking monitoring handoff.
+
+00:06:41.346 --> 00:06:52.238
+The code, immutable corpus, audit evidence, documentation, and seven-minute demonstration are published with release one point five point zero on GitHub and Hugging Face.
+
+00:06:52.238 --> 00:06:59.500
+Reproduce the snapshot, challenge the claims, and build the review from evidence you can carry forward.

assets/demo/topvenues-sf-demo-v1.5.0.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7eab33e28e6d411f55fa250ddc8d3afcff3086c6392ba403cdee4335d8deb3d
+size 16689981

assets/demo/topvenues-sf-demo-v1.5.0.pt-BR.vtt

@@ -0,0 +1,181 @@
+WEBVTT
+
+00:00:00.000 --> 00:00:03.825
+Uma revisão de literatura pode falhar antes mesmo da triagem.
+
+00:00:03.825 --> 00:00:13.006
+Se a população de artigos elegíveis vier de portais mutáveis, buscas não documentadas ou exportações privadas, outro pesquisador não consegue reconstruir o mesmo denominador.
+
+00:00:13.006 --> 00:00:24.100
+O TopVenues transforma essa primeira etapa em um artefato executável: um corpus declarado de cibersegurança, um snapshot local fixo e ferramentas para rastrear cada agregado até os registros individuais.
+
+00:00:30.000 --> 00:00:37.477
+Esta versão de agosto de 2026 contém 14.859 registros de vinte venues declaradas, cobrindo 2019 a 2026.
+
+00:00:37.477 --> 00:00:44.514
+Noventa e quatro vírgula um por cento possuem abstracts pesquisáveis, e todos os registros possuem BibTeX.
+
+00:00:44.514 --> 00:00:46.274
+O snapshot é imutável.
+
+00:00:46.274 --> 00:00:55.950
+Após a instalação, manifesto, banco, testes, busca, exportação e integridade da interface podem ser verificados offline com um único comando de reprodução.
+
+00:01:00.000 --> 00:01:04.650
+Comece em Search com uma pergunta de pesquisa, não com um ranking.
+
+00:01:04.650 --> 00:01:08.913
+Aqui eu busco por segurança de LLM usando ranqueamento textual BM25.
+
+00:01:08.913 --> 00:01:12.012
+O resultado não é uma lista de recomendações.
+
+00:01:12.012 --> 00:01:20.150
+É um conjunto de candidatos inspecionável: título, autores, venue, ano, classe, trecho do abstract, link da fonte e citação permanecem visíveis.
+
+00:01:20.150 --> 00:01:24.800
+Filtros por substring também estão disponíveis quando a menção textual exata importa.
+
+00:01:30.000 --> 00:01:39.900
+Para mapear referências, posso restringir a mesma pergunta ao top-4 de segurança: ACM CCS, IEEE Security and Privacy, USENIX Security e NDSS.
+
+00:01:39.900 --> 00:01:42.600
+Isso retorna 78 registros neste snapshot.
+
+00:01:42.600 --> 00:01:48.900
+Uma revisão sistemática normalmente deve começar com todas as venues, salvo justificativa do protocolo.
+
+00:01:48.900 --> 00:01:56.550
+Em qualquer caso, o subconjunto exato pode ser exportado como CSV, JSON ou BibTeX pronto para citação.
+
+00:02:00.000 --> 00:02:03.762
+Insights apresenta distribuições, mas os gráficos não são decorativos.
+
+00:02:03.762 --> 00:02:09.197
+Cada barra de venue, ano, classe e cobertura funciona como controle de navegação.
+
+00:02:09.197 --> 00:02:15.049
+Ao clicar, o valor é transferido para Search e os registros correspondentes são abertos.
+
+00:02:15.049 --> 00:02:25.500
+Esse detalhamento é essencial: o pesquisador pode questionar um agregado, inspecionar outliers e exportar a evidência, em vez de confiar em um número sem proveniência.
+
+00:02:30.000 --> 00:02:35.744
+A visão de tópicos separa volume bruto da participação no corpus de cada ano.
+
+00:02:35.744 --> 00:02:42.310
+Para LLM no top-4, encontra 282 artigos, liderados por CCS, USENIX, NDSS e Security and Privacy.
+
+00:02:42.310 --> 00:02:47.644
+A curva de participação evita confundir crescimento do corpus com crescimento do tópico.
+
+00:02:47.644 --> 00:02:55.850
+E 2026 é marcado como parcial, permitindo monitoramento sem fingir que um ano incompleto é comparável a um ano fechado.
+
+00:03:00.000 --> 00:03:02.480
+Researcher Radar responde a três perguntas diferentes.
+
+00:03:02.480 --> 00:03:06.023
+A frequência em qualquer posição mostra nomes recorrentes nas colaborações.
+
+00:03:06.023 --> 00:03:10.274
+A frequência como primeiro autor destaca quem mais aparece liderando a lista.
+
+00:03:10.274 --> 00:03:20.194
+A frequência como último autor pode ser útil em áreas onde essa posição costuma refletir liderança de grupo, mas a ferramenta não assume que essa convenção seja universal.
+
+00:03:20.194 --> 00:03:24.800
+Contagem de artigos é o padrão; ponderação por tier é uma heurística explícita.
+
+00:03:30.000 --> 00:03:39.836
+Selecionar um nome abre a evidência por trás da linha: trajetória anual, posições de primeiro e último autor, coautores diretos, anos ativos e principais venues.
+
+00:03:39.836 --> 00:03:44.951
+Os sufixos de identidade do DBLP são preservados para reduzir fusões de homônimos.
+
+00:03:44.951 --> 00:03:50.459
+Essas visões revelam quem ler e acompanhar dentro de um tópico e escopo declarados.
+
+00:03:50.459 --> 00:03:54.000
+Elas não medem citações, senioridade, autoridade ou qualidade científica.
+
+00:04:00.000 --> 00:04:07.504
+Para vigilância contínua, baixe uma watchlist portátil com tópico, autores, escopo de tier e IDs já observados.
+
+00:04:07.504 --> 00:04:11.477
+Um snapshot futuro pode calcular adições e remoções deterministicamente.
+
+00:04:11.477 --> 00:04:18.539
+A tabela de atividade emergente compara taxas recentes e anteriores como sinal descritivo, não como previsão.
+
+00:04:18.539 --> 00:04:28.250
+O encaminhamento ao arXiv busca possíveis preprints pelo nome do autor; cada resultado permanece candidato não verificado até a identidade ser conferida.
+
+00:04:30.000 --> 00:04:33.226
+Evidence torna explícito o limite das alegações.
+
+00:04:33.226 --> 00:04:39.679
+Sidnei Barbieri revisou manualmente uma amostra determinística e estratificada por venue de duzentos registros.
+
+00:04:39.679 --> 00:04:47.975
+Cento e sessenta e nove abstracts atenderam aos três critérios: completos, sem contaminação e correspondentes ao artigo amostrado.
+
+00:04:47.975 --> 00:04:55.350
+Isso representa taxa utilizável de 84,5%, com intervalo de Wilson de 95% entre 78,8% e 88,9%.
+
+00:05:00.000 --> 00:05:10.634
+A auditoria foi feita no snapshot anterior e transferida somente após uma verificação exata confirmar os mesmos 14.859 IDs e textos de abstract idênticos.
+
+00:05:10.634 --> 00:05:15.065
+A versão quatro altera dez títulos, nenhum deles na amostra.
+
+00:05:15.065 --> 00:05:21.268
+A cobertura ainda é desigual: por exemplo, ESORICS e WOOT têm muitos abstracts ausentes.
+
+00:05:21.268 --> 00:05:28.800
+Texto ausente é declarado e excluído de conclusões dependentes de abstract; nunca é tratado como evidência negativa.
+
+00:05:30.000 --> 00:05:32.914
+Dataset lifecycle separa exploração de manutenção.
+
+00:05:32.914 --> 00:05:37.769
+Um perfil publicado não pode ser atualizado no mesmo lugar.
+
+00:05:37.769 --> 00:05:50.881
+Nova coleta em APIs deve criar um perfil sucessor, registrar proveniência, comparar IDs adicionados e removidos, repetir os controles de qualidade e receber uma nova versão semântica.
+
+00:05:50.881 --> 00:05:53.308
+Linux e macOS usam reproduce.sh.
+
+00:05:53.308 --> 00:05:58.650
+O Windows nativo usa reproduce.ps1 e fecha deterministicamente cada conexão SQLite.
+
+00:06:00.000 --> 00:06:02.306
+Um fluxo produtivo é, portanto, simples.
+
+00:06:02.306 --> 00:06:06.149
+Comece amplo ao construir o denominador de uma revisão sistemática.
+
+00:06:06.149 --> 00:06:09.607
+Restrinja ao top-4 ao mapear conversas canônicas de segurança.
+
+00:06:09.607 --> 00:06:14.219
+Inspecione os registros por trás de cada trend ou sinal de autoria.
+
+00:06:14.219 --> 00:06:17.293
+Exporte os candidatos com a identidade do snapshot.
+
+00:06:17.293 --> 00:06:19.983
+Salve uma watchlist para a próxima versão.
+
+00:06:19.983 --> 00:06:26.900
+E documente cada restrição como escolha de protocolo, pois o TopVenues apoia o julgamento científico; não o substitui.
+
+00:06:30.000 --> 00:06:42.328
+O TopVenues oferece três coisas que portais de busca comuns não fornecem em conjunto: um denominador reproduzível, uma bancada de pesquisa inspecionável e um encaminhamento para monitoramento futuro.
+
+00:06:42.328 --> 00:06:52.015
+Código, corpus imutável, evidência de auditoria, documentação e demonstração de sete minutos são publicados na versão 1.5.0 no GitHub e Hugging Face.
+
+00:06:52.015 --> 00:06:59.500
+Reproduza o snapshot, questione as alegações e construa a revisão sobre evidências que possam ser levadas adiante.

assets/topvenues-abstract-search.pdf

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:67b65e27df72e3586ea9fcc93087fb5a499073b8d2901df0269d101ed6e41c14
-size 215476
+oid sha256:91b2be82ea168383cfefe8734c48e391e311af7f9b00d6d6b96104c5b756fccc
+size 137001

train-00000-of-00002.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:21d8bae64d10472c9311846e90fee86ca256bead6915572b25d5445e29258b14
-size 12338128
+oid sha256:70e04058e179ca8643d7e7a026948b8b74a5d00cd5c27f39a84ab752d0c7a7cc
+size 12337968

train-00001-of-00002.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2b070a95a9de9e1a33a2efcc4b767633bba61db4f0922b43cc4c117c6eb6fc66
-size 2636482
+oid sha256:5df704e616dfcaa43b262a0bcf6b86b4356afeca7c1d8ec28bb807835d470f14
+size 2636788