https://www.record.pt/fora-de-campo/detalhe/adepto-do-sporting-volta-a-tentar-matar-em-emboscada-com-fogo-apos-sair-da-cadeia

>Jovem de 27 anos esteve no ataque a carros de rivais do FC Porto após jogo de hóquei em patins em Alvalade
Tem 27 anos, nasceu no seio de uma família de classe alta com casa na avenida de Roma. Mas já tem uma condenação por tentativa de homicídio associada à violência de claques, no caso do Sporting, e está novamente a ser julgado pelo mesmo crime, juntamente com outros seis jovens. Chama-se Pedro Fazenda e é o principal arguido da emboscada com fogo posto que deixou cinco adeptos do FC Porto com queimaduras graves. Para o Ministério Público é o espelho de "uma atitude que revela uma personalidade extremamente imatura e violenta, movida por ódios fúteis, avidez em maltratar, mesmo que matar, quem se configura como inimigo".

Como devem os clubes lidar com este tipo de adeptos?

Porque não os expulsam pura e simplesmente?

Que tipo peculiar de loucura os atinge e como pode ser evitada?

Fico perplexo quando vejo este tipo de notícias e ainda mais quando vejo a permissividade com que os clubes lidam com esta subcultura que vive nas claques de apoio dos clubes.

https://pt.euronews.com/my-europe/2026/06/12/europa-abre-quatro-inqueritos-aos-snipers-de-fim-de-semana-em-sarajevo

Num livro-investigação, Ezio Gavazzeni reconstrói um caso dos anos 90 nunca plenamente apurado. "Os snipers de fim de semana" expõe a rede que levava europeus a disparar sobre civis em Sarajevo. O autor explica à Euronews em que ponto está a busca dos responsáveis.

Quatro ministérios públicos europeus investigam uma organização que, durante a guerra na antiga Jugoslávia, levava cidadãos abastados a Sarajevo, para os transformar em atiradores furtivos, num macabro passeio turístico reservado sobretudo a fanáticos por armas, caça e, em alguns casos, extrema-direita.

Num livro-investigação, Ezio Gavazzeni reconstrói um caso dos anos 90 nunca plenamente apurado. "Os snipers de fim de semana" expõe a rede que levava europeus a disparar sobre civis em Sarajevo. O autor explica à Euronews em que ponto está a busca dos responsáveis.

Quatro ministérios públicos europeus investigam uma organização que, durante a guerra na antiga Jugoslávia, levava cidadãos abastados a Sarajevo, para os transformar em atiradores furtivos, num macabro passeio turístico reservado sobretudo a fanáticos por armas, caça e, em alguns casos, extrema-direita.

Pelo menos três clientes da rede são investigados em Milão, outros dois na Áustria, e há inquéritos abertos também na Bélgica e na Bósnia e Herzegovina; um processo poderá ser formalizado em breve na Suíça, assim como uma interpelação no Parlamento Europeu.

Quem o revela é Ezio Gavazzeni, autor do livro_I cecchini del weekend. L’inchiesta sui safari umani a Sarajevo_ (editado pela PaperFIRST), atualmente em digressão por Itália para o apresentar.

Muitos “partiam à sexta-feira à tarde de via Mecenate” em Milão, de autocarro para o aeroporto de Trieste, conta à Euronews. Alguns usavam o de Parma e outros seguiam por mar a partir de Ancona, muitas vezes com primeira paragem em Belgrado ou em países vizinhos, para depois continuarem por terra até à frente da guerra civil.

...

The world’s 100 largest asset managers crossed $100 trillion in assets under management for the first time.

This chart shows the top 50, which together hold about $93 trillion in AuM.

A few things that stood out:

BlackRock and Vanguard alone manage about $26 trillion.

BlackRock added about $2.5 trillion in AuM from 2024 to 2025.

Vanguard added about $1.9 trillion.

Together, the two largest managers accounted for nearly one-third of the total AuM growth across the top 50.

BNP Paribas Asset Management was another major mover, with AuM nearly tripling after the AXA Investment Managers acquisition.

The chart groups firms by ownership type: independent, bank-owned, and insurance-linked asset managers.

https://www.cmjornal.pt/economia/detalhe/estrangeiros-que-recebem-apoio-do-rsi-sao-menos-de-5

>O valor máximo deste apoio de combate à pobreza pago a um casal com um filho é de 544, 63 euros.
O número médio mensal de beneficiários do Rendimento Social de Inserção (RSI) estrangeiros foi, no primeiro trimestre, de 7685 indivíduos, segundo dados da Segurança Social que, desde o final de 2025, passou a desagregar informações relativas a estes residentes. Face ao número total de beneficiários, os estrangeiros representam menos de 5% do total de pessoas que recebe este apoio destinado a combater a pobreza extrema.
A informação estatística da Segurança Social relativa aos três primeiros meses deste ano dá conta de um universo total de beneficiários do RSI de 161 171, cerca de 11 500 menos do que no mesmo período de 2025. Deste universo, 7685 são beneficiários estrangeiros. Este apoio, que deverá integrar a nova Prestação Social Única (PSU), tem um valor máximo para um titular de 247,56 euros, 173,29 euros pelos adultos seguintes e de 123,78 por menor, ou seja, um casal com um filho receberá, no máximo, 544, 63 euros. Na prática, a média por indivíduo ronda 160 euros. 
Os dados divulgados permitem concluir que os cidadãos da Ucrânia constituíam, no primeiro trimestre deste ano, o grupo estrangeiro mais numeroso no RSI, com uma média mensal de 2516 beneficiários, seguido do Brasil, com 958, e de Angola com 774 beneficiários. No polo extremo da tabela, os grupos menos numerosos, encontram-se os cidadãos do Bangladesh (93 beneficiários de RSI), Índia (75) e Nepal (47).
Ainda segundo dados oficiais, o Brasil é o país com o maior número de residentes em Portugal, com cerca de 485 mil cidadãos, cinco vezes mais do que o número de residentes da Índia, que é já o segundo país mais representado.

https://github.com/IoakeimKyrgiafinis/WorldCup2026ML-MonteCarloSimulationPrediction

There is a lot of discussion about how the High temperature in USA venues will affect the 2026 World Cup tournament.

This is a humble attempt in estimating how heat might affect outcomes, deriving heat effect from studies in the way shown in the pictures, trying to account for both heat at each specific venue and every national team's assumed acclimatization to heat, while also penalizing high pressure playstyle teams due to the assumed effect heat has on high intensity performance, favoring the more tactical-style teams.

Backtesting is done in 2022 World Cup, model fails to predict Argentina as a winner because of its favoring of high-valued Squads, but it captures the teams that were the strong Bookmaker Odds' favorites at the given time.

Any and all feedback for fixes of possible mistakes and further development are more than welcome!

Full methodology below same as seen in github repo.

This project combines squad market valuations, historical match results, age factors, bookmaker odds, and environmental factors (heat stress, pressing intensity) to simulate the 2026 FIFA World Cup 10,000 times and estimate each team's probability of winning the tournament. The approach is inspired by Groll et al. (2019) and Zeileis et al. (2026).

In short, a Random Forest Classifier is trained on an 80/20 train/test split. Train dataset features are extracted from publicly available Kaggle Datasets (Jürisoo, 2023), (davidcariboo, Transfermarkt. (2024)). The features extracted are the following:

Feature	What It Measures
total_value_diff	Squad market value gap (€)
avg_value_diff	Average player quality gap
gini_diff	Value distribution inequality gap
form_diff	Recent win rate gap (last 5 games)
prime_diff	Age-prime score gap

*We compute the prime age score of each player in the following way using Gaussian Peak curve:

prime score(age) = e^(-(age - 25.5)^2 / (2 * 3.5^2))

where 25.5 is the peak age and 3.5 is the sigma (spread). (Branquinho et al., 2025)

The Random Forest Classifier model is trained on these features on club matches from 2005 onwards, not international matches. This is done because there are far more club matches available than international matches, which happen infrequently. The assumption is that football outcomes are affected in the same way by the features for both clubs and national teams.

The target variable (result) takes three values: 1 (home win), 0 (draw), and -1 (away win). Once trained, the model outputs three class probabilities for each result: P(home win), P(draw), and P(away win).

Match Probability Engine

For the 2026 World Cup, the raw model output is not used directly. It goes through three sequential adjustments before becoming the final match probability.

1. Heat Stress Adjustment (α=0.002)

There is a lot of discussion about how the high temperatures present mainly in USA venues in June-July are going to affect player performance in the tournament. We try to account for it by first implementing a heat stress adjustment.

• Each team has a baseline temperature (team_baseline_temp) representing their typical training climate.
• Each venue has an expected match-day temperature (venue_heat).

Heat stress is the gap between the venue temperature and the team baseline—a Senegalese team playing in Houston in July experiences very little additional stress compared to a Norwegian team. The heat difference between the two teams is computed and applied as a small adjustment to win probability (α=0.002 per degree Celsius of differential). The adjustment is clamped between 0.01 and 0.99 so probabilities never reach zero or absolute certainty.

Extraction of α=0.002:

Mohr et al. (2012) observed that a massive temperature swing from a temperate baseline (∼21°C) to extreme tournament heat (∼43°C) caused a 7% total performance drop for unacclimatized players. The delta between those two test environments was exactly 22°C (43−21=22). Taking that 7% total drop (0.07) and dividing it across that temperature gap yields:

0.07 total performance deficit / 22°C temperature delta ≈ 0.0031

Accounting for modern acclimatization techniques, this scaling factor is smoothed down to a baseline of 0.002 per degree Celsius.

2. Tactical Pressing Penalty (α=0.003)

Next, we account for the fact that the participating teams have different playstyles. We extract each team's pressing intensity from analyst reports and penalize the high-pressure teams (Tor-Kristian Karlsen (2026)), based on the argument that heat is going to affect them at a higher rate. In extreme heat, a high-pressing team faces a double penalty: their tactical style becomes harder to maintain. The pressing adjustment models this interaction.

Each team has a pressing_intensity score (0 to 1). The adjustment scales the pressing differential by heat severity (venue temperature divided by 40, normalized).

Tactical Nudge = 0.003 × ΔPressing Intensity × Heat Severity

Example: A high-pressing team (Austria, 0.95) playing a low-pressing team (Qatar, 0.10) in Dallas (38.5°C) would have their win probability nudged down by approximately 0.003 × 0.85 × 0.96 = 0.0024—small but meaningful across the tournament bracket.

Extraction of α=0.003:

A high-pressing system relies entirely on sustaining continuous high-intensity running to choke the opponent's space. Conversely, a low-pressing system saves physical energy by sitting in a passive shape and focusing on possession mechanics. The paper proves that extreme heat creates a severe tactical disadvantage for high-intensity movement while rewarding a slower, cleaner passing style.

To extract the mathematical "exchange rate" of this tactical trade-off, we evaluate the friction between physical decay and technical gains recorded by Mohr et al. (2012), dividing the passing efficiency gain (+8%) by the high-intensity running loss (-26%):

Tactical Exchange Rate = Passing Success Gain / High-Intensity Running Loss

Tactical Exchange Rate = 0.08 / 0.26 ≈ 0.3076

Shifting the decimal two places to the left to scale it down safely from a raw physical efficiency metric into a percentage-point modifier for a probability outcome loop (0.3076 × 0.01) yields exactly 0.003 when rounded.

3. Bookmaker Odds Blending

After heat and pressing adjustments, the model probabilities are blended with bookmaker odds. This is done based on the argument that bookmaker odds encapsulate an enormous amount of information that a Machine Learning model trained purely on historical data cannot capture (squad news, injury reports, tactical adjustments, etc.).

American odds are converted to implied probabilities using the standard formula, then normalized to sum to 1 across all teams. For each match, the relative winner odds of the two teams determine the odds-implied head-to-head win probability.

The final blended probability uses odds_weight = 0.6:

• 60% weight to the bookmaker-implied probability
• 40% weight to the model probability (after heat and pressing adjustments)

The draw probability uses the model's draw estimate as its odds anchor (since outright tournament winner odds don't price individual match draws), then blends with the same 40/60 split. All three probabilities are renormalized to sum to 1 after blending.

Expected Goals (λ)

For every match, expected goals (λ) are computed for each team from their squad market value differential:

• λa = max!(0.5, 1.5 + value diff / 10^9)
• λb = max!(0.5, 1.5 − value diff / 10^9)

The baseline of 1.5 represents an average international match goal rate. The value differential shifts this—a €500M squad advantage adds 0.5 expected goals. The floor of 0.5 ensures no team's expected goals collapse to an unrealistic level.

Goals are sampled from a Poisson distribution—the standard model for discrete count data like football scores. Crucially, rejection sampling is used rather than clamping.

The naive approach (sampling goals, then forcing the winner to have more by subtracting 1 from the loser) distorts the distribution, creating an artificial pile-up at scorelines like 1-0, 2-1, 3-2. Rejection sampling instead draws two independent Poisson samples and accepts them only if they are consistent with the simulated match outcome. With realistic lambdas, this converges in very few tries. If the sampler fails to converge within 500 attempts (extremely rare), a minimal fallback score is used (1-0, 0-1, or 0-0 for the respective outcome).

Tournament Simulation

Group Stage

Each of the 12 groups plays a full round-robin: every team faces every other team once (6 matches per group). For each match, the outcome (home win / draw / away win) is drawn from the cached probabilities, and a Poisson score is generated. Points (3/1/0), goal difference, and goals for are all accumulated.

Final group standings are sorted by points, goal difference, and goals for—exactly the FIFA tiebreaker order. The top two teams advance as group winner and runner-up. The third-place team's record is saved for the best-third-place ranking.

Best Third-Place Teams

In a 48-team World Cup with 12 groups of 4, 8 third-place teams also advance to the Round of 32. The 12 third-place finishers are ranked by the same criteria (points, goal difference, goals for) and the top 8 advance. These are stored as best8 and slotted into the bracket in the official FIFA-specified positions.

Knockout Rounds

From the Round of 32 onwards, all matches are single-elimination. The bracket is hard-coded to match the official FIFA 2026 World Cup bracket structure, with each match numbered 73-104 and assigned to its official venue. For knockout matches, a draw in 90 minutes leads to a 50/50 penalty shootout coin flip. This is a simplification—in reality, the stronger team has a slight penalty advantage—but it is a reasonable approximation since penalty shootouts are largely unpredictable.

Monte Carlo Engine Execution

The full tournament simulation is run 10,000 times. Each run is independent—group draws, scores, and knockout results are all re-sampled from scratch. The only shared state is the matchup_cache (pre-computed probabilities), which is deterministic and identical across all runs.

After 10,000 simulations, each team's win count is divided by 10,000 to produce a win probability percentage. The results are sorted from highest to lowest probability. 10,000 iterations is sufficient for stable probability estimates at the top of the table (±0.5 for teams with 10 win probability). For very low-probability teams (below 1%), more simulations would reduce noise further, but the absolute differences at that level are not practically meaningful.

Limitations

• Training Data: Training on club data to predict international matches means the feature space is shared, but the context differs (squad size, player familiarity, tactical system cohesion).
• Static Squad States: No live injury or suspension modeling is integrated; a key player being suspended or injured for a knockout stage match cannot be captured.
• Deterministic Shootouts: Penalty shootouts are simulated as a static 50/50 coin flip, which ignores proven team and goalkeeper performance metrics during spot-kicks.
• Simplified Seeding Rules: The model places the best 8 third-place teams into bracket slots strictly by their ranking order, whereas FIFA's official seeding matrix uses more complex, group-dependent path-blocking constraints.
• Outright to Match Probability Conversions: Converting tournament outright winner odds to localized head-to-head match probabilities assumes that relative outright odds accurately approximate isolated match-level win distributions.

Tool: Python (CatBoost match model + Monte Carlo, pandas, matplotlib).

Source: my own match-prediction model trained on historical results, 10,000 full-tournament simulations. I publish the daily snapshot as timestamped, signed CSVs here: https://github.com/uanalyse/world-cup-2026-predictions

Interactive version with the full bracket and per-team chances: https://uanalyse.co.uk/world-cup-2026

Reading the chart: France top at 12.0%, Spain and Argentina tied at 9.8%, then a tight pack down to Brazil at 5.6%. The top two only combine for about 22%, and the expanded 48-team format is what spreads it this wide. Happy to answer anything on the method.

Faço voluntariado a apoiar famílias pobres.

Muitas delas não conhecem alguns apoios que podiam usufruir e estou a pensar fazer um livrinho com todos os apoios que elas podem usufruir.

Vou escrever aqui os que me conheço e me lembro agora. Peço que me ajudem a ajudar as famílias pobres, indicando os que me possam estar a esquecer.

1. RSI - Rendimento Social de Inserção
2. Cartão social eletrónico para a compra de alimentos (em fase de implementação)
3. Apoio à renda - Porta 65 Jovem e Porta 65+. Algumas câmaras municipais também têm programas para apoio à renda
4. Botija de Gás Solidária
5. Algumas Câmaras têm apoios para a compra de medicamentos
6. Tarifas sociais da água (depende da câmara municipal) e da luz
7. SASE para o material escolar dos alunos carenciados (apenas 16€)
8. Subsídio por Assistência de Terceira Pessoa - É um subsídio destinado a famílias que recebam o Abono de Família com bonificação por deficiência e que precisam de acompanhamento permanente.
9. Atestado multiuso de incapacidade
10. Fundo de garantia em caso de incumprimento das responsabilidades parentais
11. Apoio jurídico, nomeação de um advogado, em caso de necessidade
12. Há câmaras municipais que têm apoio para a recuperação de casas degradadas (normalmente pedido pelo inquilino).
13. Complemento Solidário para Idosos (com o agradecimento ao GentlyGliding pela lembrança)
14. Cheque-dentista
15. Complemento por dependência, em situações de pessoas dependentes de 3os ou acamadas (lembrança de morphedrine)
16. nalguns locais ser sócio dos bombeiros dá direito a descontos na farmácia

Muitas famílias pobres desconhecem algumas destas coisas. Principalmente as imigrantes.

E fico com a ideia de que ainda me faltam algumas coisas.

Ajudem-me a ajudar.

https://portaldaqueixa.com/brands/pingo-doce/complaints/pingo-doce-chocado-com-o-desperdicio-de-comida-que-vai-para-o-lixo-sem-sequer-deixar-os-funcionarios-ou-associacoes-levar-127637925