The Method

How AFOS Analytics Works

A didactic guide to navigating the platform

Introduction, Why AFOS exists

Every day you open a newspaper and read "Poll X says candidate Y has 37%". In another, "32%". Which one to believe?

The problem: electoral polls measure declared intent, what a person says they will do. But intent changes, polls have bias, and in Brazil this has gone wrong several times (2018 and 2022 had major surprises).

AFOS's solution: instead of trusting ONE source, the platform cross-references three independent sources in real time:

When these three sources agree, the forecast is robust. When they diverge, it's a sign something is moving, and that's extremely valuable information.

How data is cross-referenced (the method)

AFOS doesn't do formal statistics (regression, Bayesian models). It does something different and more useful day-to-day: a structured narrative cross-reference with explicit rules.

Golden rule: convergence vs divergence

For each important question (e.g., "who wins the first round?"), the platform compares the values from the 3 sources:

The gold is in divergence

When Polymarket and polls diverge, the reason is investigated:

• Poll above, market below → either the poll is outdated/biased, or the market knows something (imminent operation, leaking scandal)
• Market above, poll below → either the market anticipates a turnaround, or it's speculation with low volume

Real example, the Candidate D surge and retreat

Those who read AFOS daily arrive at conclusions earlier, because they see the movement while it's happening, not after.

Understanding ↑↓pp variations

"pp" = percentage point. It's the difference between two percentages. Different from "percent".

Example: Candidate A had 40%, today has 42%. They went up 2 percentage points (2pp). In relative terms, that's a 5% growth (2 on top of 40).

Why small variations matter

1. Liquidity: the presidential Polymarket has US$ 54 million at stake. A 0.8pp change means roughly US$ 432 thousand was repriced. It's not opinion, it's real financial commitment.

2. Speed: 0.8pp in 1 day seems small. If the pace holds: 5.6pp per week; 24pp per month; complete reversal in 5 months. Small and persistent movement beats large and isolated movement.

3. Anticipation: when the market moves, it moves before the newspaper consensus. 48 hours later, you'll read analysts saying what the market already said.

Interpretation table

Mental rule in one sentence: "A 1pp move is a tweet. 3pp is an interview. 5pp+ is a done deal."

The home page (landing)

When you open afos-analytics.com, you land on the home page, the public entry point. It gathers, in one place, everything that orients a visitor before stepping into the dashboard:

• Top bar (header strip), the AFOS Analytics logo + direct shortcuts to Daily · Tradeoff · Global, a language switcher (PT/EN/ES), a theme toggle (light/dark), and the Dashboard button to enter the app.
• Top badge, just below the header strip, a clickable Harvard Dataverse · DOI pill that links to the permanent, citable academic deposit of the Brazil 2026 dataset (DOI 10.7910/DVN/2D0UK7).
• What AFOS does, plus live stats (monitored countries, polling institutes, market refresh frequency).
• "Our analyses", a grid with the four main shortcuts: The Method (this guide), AFOS Daily, AFOS Tradeoff, and AFOS Global.
• Global coverage, a strip with the monitored elections (Brazil, France, Germany, United Kingdom, Canada, and more).
• Email signup, the "get weekly analyses in your inbox" block: an email field + opt-in consent (you tick the box) to receive alerts, summaries, and updates. No spam, unsubscribe anytime. Right after confirming, a welcome screen lets you choose the language you want to receive (Português, English, or Español), and that preference sets the language of your Daily and Tradeoff emails.
• Lean footer, the platform name, a one-line description, the disclaimer (no Polymarket affiliation), and a row of icons: Dashboard + GitHub, LinkedIn, X, Bluesky, Product Hunt, 🤗 Hugging Face, and 🏛️ Harvard Dataverse. (The full four-block footer, with detailed lists, is the one on the dashboard and other pages, described in section 14.)

1. Header (top of the page)

At the top of the screen you see the AFOS Analytics logo and three navigation buttons:

The header is present on all pages. It's the navigation anchor.

2. AFOS Daily, Daily Synthesis

Right below the header you see the AFOS Daily card: a light-blue block with today's date, an excerpt of the day's narrative synthesis (up to 2 lines) and a "Read synthesis →" button.

It's the only piece of the dashboard that is not raw data, it's a curated narrative (~700 words, 4-minute read) cross-referencing the three signals (market, poll, news) with an inline link for every claim. Published once a day, at the end of the day.

When it's worth reading: when you want to understand why the dashboard numbers moved, not just how much. The dashboard shows the day's snapshot; AFOS Daily tells the story, with auditable sources for you to verify each step.

AFOS Tradeoff, Weekly Technical Brief

If the AFOS Daily is the everyday read, the AFOS Tradeoff is the weekly deep dive: a technical brief published every Monday, covering the week just closed. It's built for the professional market reader (research, institutional desk, treasury, buy-side), focused on pricing and divergence rather than narrative.

While the Daily tells the day's story, the Tradeoff measures the week's move (weekly Δ) from persisted snapshots and lays everything out in a fixed, fast-scan structure for readers who need an actionable signal.

When to read: when you want the week's read rather than the day's snapshot, where the market leads or lags the polls, what actually changed, and what is noise. The Daily is the daily pulse; the Tradeoff is the weekly X-ray.

AFOS Global, the Method Worldwide

If the Daily and the Tradeoff read Brazil today, AFOS Global shows that the same reading holds in any election. The Global page has two layers. At the top, the Validated cases: elections already held, where we cross the prediction market with the polls and check the result against what actually happened at the ballot box. Below, the Live layer: the odds map of ongoing elections, with market pricing only, without the polls and divergence layer.

Why validated cases matter: they are the proof of the method. In an election that has already been counted, you can see, in hindsight, when the market saw what the poll did not, and when the market got it wrong. There are six today: Brazil, Peru, Colombia, Chile, Germany and Canada, from the Americas to Europe.

Each validated case opens in two complementary screens:

• The country page carries the Divergence analysis: a candidate-by-candidate table with three columns, Poll (vote share), Market (probability of winning) and Divergence (the difference in percentage points). When a number deserves a caveat, the table annotates it rather than hiding it. In Peru, for example, Carlos Álvarez's 31.6% was a one-day spike in a thin market, and is flagged as such.
• The election page shows the "Who wins?" card: the snapshot of the market on election day (pre-result, with the market already closed) and the accumulated volume in dollars, the same "real money" reading as the Brazilian dashboard.

In practice, Chile's divergence table looks like this:

The winner is precisely the name the market most overpriced relative to the poll. That is the pattern AFOS pursues: divergence as signal, not as noise.

Open data: each validated case has a public dataset on Hugging Face, under CC BY 4.0 license, with the polls, the odds and the calculated divergence. Anyone can audit or reproduce the analysis.

Why cover the world: wherever there is an election, there is a signal. The same method applied from the Americas to Europe shows the reading is not a Brazilian fluke, it is structural.

Structural context, the country's frame

Below the electoral signals, every validated case and the Brazil dashboard carry a Structural context block: official, open World Bank indicators that frame the country alongside the market signal, not as a predictor of it. Three groups: Governance (six Worldwide Governance Indicators, 0 to 100 scale: political stability, voice and accountability, rule of law, government effectiveness, regulatory quality, control of corruption), Economy (population, GDP, GDP per capita, inflation) and Education (public education spending and expected years of schooling). They are annual, structural indicators, they change slowly and do not predict the electoral outcome, they are background for the reading. Source: World Bank (WGI via Data360 and WDI v2), open and citable.

The cross-reference graph, the method became a navigable map

On every validated case and, above all, on the Brazil dashboard, the cross-reference now has an Obsidian-style force-directed graph. The election sits at the center, surrounded by the four pillars (markets, polls, press and structural context). The divergence is the star: each candidate is linked by a thin edge colored by magnitude (red = high, amber = medium, green = convergence), with the Δpp written on the edge itself. On validated cases a "real result" node points in green to whoever actually won; candidate nodes use party colors.

On the dashboard the graph is the platform's navigable brain. Hover a node and it and its connections light up blue; click and each node takes you somewhere: data nodes open the matching folder in the open Hugging Face dataset, context nodes scroll to the on-page context card, and the shortcuts lead to the products (Daily, Tradeoff, Global, White Paper, Governance), the dashboard sections and Harvard Dataverse (collection + DOI). It is the whole method, open data, live election and platform, in a single map, in three languages.

3. The 6 Polymarket Cards, Instant dashboard

Right after the header, six cards appear side by side summarizing Polymarket's most important markets at the moment. Each card shows a percentage (probability priced by the market) with the variation relative to the previous day (↑↓pp).

4. Electoral Polls

Below the Polymarket cards, you find the electoral polls section.

How polls arrive here

All polls registered with Brazil's TSE (Superior Electoral Court) are downloaded automatically every day. The database holds more than 150 indexed polls and grows by about 2 to 4 new polls registered per week, a pace that accelerates as the electoral cycle advances.

Monitored institutes

The platform tracks 17+ Brazilian institutes. The most frequent in the last month:

Knowing not only what polls say, but what they're about to say

Under Brazilian legislation, every institute is required to register each poll with the TSE before publishing it, with a unique protocol, field dates (when it's being administered), expected publication date, sample size, and cost. This registration is public and becomes available in the TSE's official database the moment the institute submits it.

This is where AFOS intelligence kicks in: the platform runs automatic ingestion cycles throughout the day querying the TSE directly. When a new poll is registered in the official database, within hours it's already processed, cross-referenced with Polymarket, and displayed on your screen, without depending on a journalist covering it or an official note from the institute.

What you see for each poll

• Institute (e.g., Paraná Pesquisas, Datafolha, Quaest)
• TSE protocol (unique, auditable identifier)
• Field: dates when interviewers are collecting responses
• Expected publication: when the institute will release results
• Sample: number of respondents
• Status: "published" (already out) or "active field" (still being administered)

Real example today (Apr 19)

Paraná Pesquisas, national, field Apr 21-23 (in progress), expected publication Apr 24, sample 1,680

You know 5 days in advance that on Thursday there will be a national Paraná Pesquisas poll with nearly 1,700 respondents.

Institute evaluation criteria

Beyond listing polls, AFOS displays a dashboard card called "Monitored Institutes, Reliability", where each institute receives a 1-to-5-star rating. This rating serves as an editorial weighting ruler to help readers decide how much to trust each source when polls diverge.

Nature of the rating: it is a qualitative editorial evaluation, not an automatically calculated score. It reflects public consensus in the Brazilian electoral market (analysts, specialized journalists, methodological literature). It works as an honest first approximation; evolution toward a quantitative score is on the AFOS roadmap post-election cycle.

The 5 criteria considered

Star scale and interpretation

Sources consulted for the classification

• Official TSE electoral results, public comparison between predictions and ballot box results in previous cycles (2018, 2022)
• Brazilian methodological literature, ABEP (Brazilian Association of Research Companies), academic articles on electoral poll accuracy, methodological analyses at FGV and Poder360
• Specialized journalistic consensus, reference political analysts (Folha, Estadão, O Globo, Poder360) who historically weight institutes in similar fashion
• Public TSE history, registration protocols, sample size, declared cost, publication frequency (all public and auditable)
• The institutes' own websites, declared methodology, disclosure of questionnaires, transparency about weighting and stratification

5. In-Depth Analysis (of the top 4 candidates)

This is the richest section and the one that requires slower reading.

The analysis is divided into 4 sections: Candidate A, Candidate B, Candidate C, and a section grouping candidates D, E, and F. Each section has three blocks: STRENGTHS, WEAKNESSES, and ANALYSIS.

🟢 "STRENGTHS" block

Everything that is working in the candidate's favor that day, with source, date, and outlet cited. It's not opinion, it's auditable data.

Today's Candidate A example: "POLYMARKET (Apr 19): 39.5% (stable) while Candidate B deflates. 2nd-place Poly Candidate A RISES 17% (↑1pp). Folha: 'Candidate A intensifies agenda aimed at women'. Poder360: 'Women become central focus'."

🔴 "WEAKNESSES" block

Everything working against the candidate, with the same depth as strengths blocks. AFOS is symmetric.

Today's Candidate A example: "Candidate B keeps minimal lead. Paraná Pesquisas São Paulo: runoff Candidate B 48.1% × Candidate A 40.3%. BNews: Candidate A drop / Candidate B growth in the Northeast. Gazeta do Povo: 'How Candidate A melted their 2022 advantage'."

🔵 "ANALYSIS" block

The stitch. How strengths and weaknesses connect, and what it means strategically at that moment.

6. Comparative Table

A single table summarizing candidate by candidate:

7. Candidate Profiles

This section presents each candidate in individual cards with five fields:

• Name, party, age, current role
• Polymarket: current percentage in the presidential market
• Poll: most recent voting intention number
• Position: ideology summarized (center-left, liberal right, etc.)
• ⚠️ Risk: the day's summary, what might change, what's under pressure, what favors

8. Countries (dashboard buttons)

The dashboard has a row of country buttons, direct shortcuts to the profile of each monitored election (15 countries: Brazil, USA, France, Germany, UK, Canada, Colombia, Chile, Peru, among others). It's the quick way to leave Brazil and open another country without going back to the home page.

The full international layer, the validated cases, the country and election pages, and the open datasets, is explained in the AFOS Global block (right after Daily and Tradeoff). Here on the dashboard, the buttons are just the navigation shortcut.

9. Live Elections News 120'

A live feed showing news published in the last 120 minutes related to monitored elections. Sources include Google News, major Brazilian portals, and international agencies.

How it works: every 30 minutes, a bot fetches news in 6 different categories (presidential election, specific candidates, scandals, polls, government approval, state races) and in the platform's 3 languages (PT-BR, EN, ES). The feed displays the most relevant in chronological order.

10. Political Climate

A dedicated panel showing the general climate of the race through four simultaneous lenses:

• Right: what's working for/against right-wing candidates
• Left: same for left-wing candidates
• Third way: how candidates outside the Candidate A / Candidate B axis are moving
• Consolidated Polymarket: the single number of the day summarizing the scenario

11. INSS Scandal and the Incumbent's Family Case

Specific card about the biggest economic scandal of 2026, the fraud of undue INSS (Brazilian pension) deductions, and the ramifications involving a family member of the incumbent.

What it shows: text structured in 4 blocks:

• Current context of the case (day's updates)
• Institutional dynamics (Congress, PF, PGR, Supreme Court)
• Supreme Court impact (impeachment probability on Polymarket)
• Political field (how the scandal affects Candidate A vs Candidate B)

12. Banco Master Scandal Impact

Card focused on the Banco Master case and the plea bargain of the executive involved, another economic scandal unfolding in chapters.

What it shows:

• Latest developments (e.g., Central Bank approved the executive's control after initially rejecting)
• Institutional tensions (Federal Police × Attorney General's Office, Congressional Inquiry × Supreme Court)
• Cross-reference with Polymarket (does the market believe in a justice impeachment?)
• Electoral consequences

13. Supreme Court Credibility, Electoral Impact

Card dedicated to reading the Supreme Court as an electoral actor, because the Court, though it doesn't vote, decisively influences elections.

What it shows:

• Justice by justice (Justice 1, Justice 2, Justice 3, Justice 4): what each is doing
• Nexus: how individual actions connect into institutional strategy
• Analysis: interprets Polymarket, does the market expect rupture (impeachment) or accommodation?

Today's example: "Supreme Court impeach falls 11.5% (↓1.5pp, -4.5pp in 2 days). Court wants to tighten inquiry rules. Justice 1 may recuse themselves in the BRB case."

Translation: the Court is protecting itself. The market senses there will be no impeachment, which reduces risk for candidates who would bet on institutional rupture.

14. Footer (page bottom)

This is the footer on the dashboard and other pages (the home page has a leaner version, noted in the landing section). It is organized into four lean blocks, each with a clear purpose. No link in the footer points to an empty page, each one delivers something specific.

Block 1, Navigation

Shortcuts to the main platform areas:

• Dashboard, main application with the 6 Polymarket cards, analyses, and thematic cards
• Global Map, interactive D3.js visualization of the 14+ monitored countries
• Latin America, regional hub with Brazil, Colombia, Chile, and Mexico
• Europe, regional hub with France, Germany, and the United Kingdom

Block 2, Open Source

Full transparency about the project, following the reference standard in open-source software:

• Apache 2.0 License, use, modification, and redistribution permitted with attribution
• ⭐ GitHub, public repository with auditable source code
• Security, responsible disclosure policy for vulnerabilities
• Contributing, guide for external developers to submit improvements
• Code of Conduct, community coexistence rules (Contributor Covenant)
• Governance, how editorial integrity is enforced through code (automated validators + versioned rules) rather than case-by-case human review
• Trademark, usage policy for the "AFOS Analytics" name and mark

Block 3, Get in Touch

Four email channels segmented by purpose:

• 📧 Contact, press, partnerships, and general matters
• 💬 Support, help using the platform
• 🔒 Security, confidential vulnerability reports
• 👤 Founder, direct contact with the founder

Block 4, Social networks and open data

A row of icons connects to AFOS's official presences, each a real and up-to-date channel:

• ⭐ GitHub, auditable source code and commit history
• LinkedIn, the founder's profile (André Felipe) with institutional updates
• X (@AFOS_Analytics), real-time distribution of signals and market moves
• Bluesky, the same coverage on the open ecosystem
• Product Hunt, the product's page in the tech community
• 🤗 Hugging Face, the open dataset of divergence (markets × polls × press), updated daily under a CC BY 4.0 license, the footer's data link goes straight to it
• 🏛️ Harvard Dataverse, the AFOS Analytics collection of curated, citable academic datasets (Brazil 2026 and USA 2024), each a permanent, versioned snapshot with its own DOI

Bottom row

Bottom line with platform identification, data sources with real frequencies ("Polymarket 5min, 17+ TSE Institutes, Google News 30min"), Polymarket non-affiliation disclaimer, and "back to top" button.

Behind the platform

Data arrives on its own

Everything you read comes from automated pipelines that run 24 hours a day:

• Every 30 minutes: Polymarket is queried and percentages updated
• Every 30 minutes: news is collected in 6 thematic categories and 3 languages
• Daily: new polls registered with the TSE are downloaded and indexed
• Twice daily (12pm and 6pm BRT): complete cross-referencing of the 3 sources is executed, comparing current state with the previous day (↑↓pp variations), and persisted in the database to form an auditable history

Analyses generated by AI from public data

The in-depth analyses (the Strengths, Weaknesses, Analysis blocks, the Cross-reference, and the four thematic cards) are generated by artificial intelligence that:

1. Reads the current values from the 3 sources
2. Compares with values from the previous day
3. Consults the most relevant news from the last 24 hours
4. Applies the convergence/divergence rules described earlier
5. Writes the resulting narrative, citing sources, dates, and outlets

All data used is public and auditable, anyone can verify Polymarket, TSE polls, or cited news.

Permanence and citability: beyond the daily, auditable mirror on Hugging Face, the Brazil 2026 dataset has a curated, citable academic version on the Harvard Dataverse (DOI 10.7910/DVN/2D0UK7), a permanent, versioned snapshot with a fixed reference for anyone who wants to cite AFOS in research.

User profiles

When AFOS isn't useful (honest limitations)

No platform is useful for every question. Being honest about what AFOS doesn't deliver is what separates a serious tool from a vague promise.

AFOS doesn't replace formal statistical research

If you need margin of error, confidence interval, or controlled scientific sampling (in plain language: numbers with certified mathematical precision and auditable sampling methodology), the source is the polling institute (Datafolha, Quaest, IBGE, etc.). AFOS consolidates and cross-references this data, but doesn't produce new polls.

AFOS doesn't predict results with quantitative precision

The cross-reference is structured narrative, not statistical model. The platform doesn't deliver predictions with calculated mathematical precision. It delivers direction, pace, and convergence, qualitative readings useful for supporting decisions, but that don't replace the formal mathematical modeling that academics and quantitative funds use.

AFOS depends on the quality of prediction markets

In countries where Polymarket doesn't have active markets or has markets with very low liquidity (below US$ 100 thousand in volume), the market signal becomes noisy. AFOS flags these cases, but data confidence drops proportionally.

AFOS is not investment or voting recommendation

It's structured information to support decisions. Decisions about portfolio, bets, or votes are the sole responsibility of the user. The platform doesn't work with clients, doesn't receive commission, and has no declared conflict of interest, precisely so it doesn't have to recommend anything.

Current coverage is restricted to 14+ countries

Countries outside this list don't have a specific collection pipeline. The global map shows aggregates, but the depth of analysis (polls × Polymarket × news cross-reference) only exists where the infrastructure is ready. Expansion is continuous, but not universal.

What makes AFOS different from Google News or a newspaper

Trilingual glossary of political terms

Brazilian politics has terms with no direct translation (TSE, STF, first round, technical tie, BolsoMaster, tarifaço) that can stall a reader from outside the country. AFOS keeps a dedicated glossary that explains each one in 3 languages (PT-BR · EN · ES).

Start here

If this is your first visit, this is the fastest way to extract value in 5 minutes:

Updated April 2026