To compare with the polls, we use the Ranking dos Políticos polling aggregator, which averages the main Brazilian pollsters in valid votes. On election eve it had Lula at 45.3%, Flávio Bolsonaro at 43.2% and the gap between them at 2.1 points. Across the final three weeks of the campaign, our forecasts were closer to the result than that average on all three forecasts.

Every Mantic forecast linked below is a snapshot frozen before the vote.

Lula: 45.12% forecast, 45.16% result

Our final forecast put Lula at 45.12%, with an 80% range of 42.7% to 47.6%, and a median of 45.17%. He won 45.16%: a miss of 0.04 points on our most likely value and 0.01 points on the median. The polling average missed by 0.14 points.

Flávio: 44.69% forecast, 47.03% result

We forecast 44.69% for Flávio, with an 80% range of 41.9% to 47.9%. He won 47.03%. The polling average had him at 43.2%: it missed by 3.8 points, we missed by 2.3.

Our forecast started from the polling average and added two adjustments: late consolidation of minor candidates’ voters, and a partial correction for the polls’ underestimate of the right in 2018 and 2022.

Subscribe now

The gap: Lula +1.05 forecast, Flávio +1.87 result

With other candidates in the race, it was worth independently forecasting the gap between the two front-runners. We forecast Lula ahead by 1.05 points, with an 80% range from Flávio +3.5 to Lula +5.2. Flávio won by 1.87 points. The polling average had Lula ahead by 2.1: it missed the gap by 4.0 points, we missed by 2.9.

Over the last two weeks of the campaign, our forecast of Lula’s lead stayed between 1.4 and 2.4 points, while the polling average ranged from 2.5 to 3.9.

Not just on election morning

We compared each of our updates from 14 September (16 September for the gap) to 2 October with the polling average on the day before. Using our medians, our average miss was 1.0 points on Lula against 1.6 for the polls, 5.2 against 6.5 on Flávio, and 4.0 against 5.0 on the gap.

On Flávio and on the gap, we were closer than the polling average at every update. On Lula, at six of eight. Our Flávio forecast rose from 38.7% on 14 September to 42.7% on 2 October, above the polling average throughout.

What’s next: the 25 October runoff

We now give Flávio an 80.8% chance of winning the presidency, against 18.9% for Lula. On 2 October, before the first round, we had him at 57.8%.

We featured Lula’s first-round vote share before the vote in our latest edition of Friday Forecast, the Election Bonanza, with forecasts on the US midterms, Brazil, France, Israel and Nigeria. For how our forecasts compare with prediction markets over a longer period, see our backtest against Polymarket.

1. Lula’s first-round vote share and his chances of reelection ↗

Lula polls at 43–46% in the first round, below an outright majority. Even if he finishes first, he may struggle to find the extra votes needed to win, as smaller candidates’ voters are expected to favor Flávio in the runoff. Explore on Mantic ↗

• 42%. A substantial underperformance against polling suggests weaker underlying support or adverse polling error carrying into the runoff. Right-leaning minor-candidate voters make recovery harder, although campaign changes leave a comeback possible.

• 43%. This remains below the polling-based baseline. The models carry part of that disappointment into the runoff and allow Flávio to consolidate eliminated candidates’ voters. Polling uncertainty and subsequent campaigning preserve a recovery path.

• 44%. A smaller polling disappointment leaves Lula competitive but disadvantaged. Models link the result to weaker runoff support and estimate that most minor-candidate transfers favor Flávio.

• 45%. Roughly matching polling provides little evidence of a collapse or surge. With an outright Lula victory ruled out, the decisive factors are Flávio’s support, voter transfers, turnout and the runoff campaign.

• 46%. Matching or slightly exceeding expectations improves inferred electoral strength. That makes the runoff approximately even, but the rightward tilt of minor-candidate transfers prevents a strong Lula advantage.

• 47%. Stronger-than-baseline performance positively updates Lula’s underlying support. Some strength may reflect consolidation of both leading candidates rather than gains against Flávio, so transfers and campaigning remain consequential.

• 48%. A clear polling outperformance supports a persistent strength or turnout advantage. The models also invoke the 2022 precedent. Close runoff polling, uncertain transfers and incomplete persistence keep defeat plausible.

2. Democratic Senate wins in states Trump carried by at least 10 points ↗

Democrats are competitive in several states Trump won by at least 10 points in 2024. These races can move together, and a favorable national swing could deliver several major Senate wins to make the Congress fully Democrat. Explore on Mantic ↗

• Favorable September polling across five key battlegrounds (Alaska, Ohio, Texas, Iowa, and Kansas) was weighed against prediction market odds that favored Republicans in Iowa and Kansas, anchoring my expectations around three wins.

• Potential independent victories in Nebraska and Montana contribute little to the count, as both candidates explicitly rejected party affiliation and are unlikely to fulfill the requirement to caucus with Senate Democrats.

• Accounting for shared national polling error rather than treating races as independent widens the outcome spread, accommodating both multi-seat Democratic waves and near-total Republican sweeps.

• My forecast is right-skewed, with an 80% range of 1 to 5 wins and a 90th percentile of 6 wins, as long-shot upsets in deep-red states like Florida and South Carolina stretch the upper tail.

• The main remaining uncertainty is whether late Republican consolidation and systematic polling errors across red states will emerge over the closing weeks to erase narrow Democratic leads.

Subscribe now

3. The far-right’s combined first-round vote share in France ↗

The far-right bloc polls near 40%, but early polls overstated it by 2–5 points in past races. What will be their actual share on election day? Explore on Mantic ↗

• September 2026 surveys across seven polling organizations put the combined far-right bloc near 40%, led by Marine Le Pen after a July 2026 court ruling restored her eligibility.

• I reduced this baseline by nearly two points rather than projecting polls directly, as early surveys overestimated the far-right bloc by two to five points in 2007, 2017, and 2022.

• While minor candidates face signature hurdles to reach the ballot, I deducted only a fraction of a point for ballot attrition because most of their supporters transfer to other qualifying candidates.

• My distribution is roughly symmetric around 38.8%, with an 80% range from 33.5% to 44.0%, allowing for tail risks from a major campaign setback to a late surge.

• The primary remaining uncertainty is how candidate momentum and polling error evolve over seven months of campaigning, alongside Le Pen’s pending appeal before the Court of Cassation and potential replacement by Bardella.

4. Who Israel’s president first tasks with forming a government ↗

In 2021 Netanyahu received the first mandate with 52 recommendations, yet Bennett became prime minister. The opposition’s pact may unify its votes, but winning the first invitation still differs from winning power. Explore on Mantic ↗

• Mainstream polling puts the Zionist opposition slightly ahead with Yashar leading Likud at 22 to 23 seats, though right-leaning polls show the governing coalition reaching 60 to 63 seats.

• Opposition leaders signed a cooperation pact making a unified recommendation likely in roughly 79% to 82% of scenarios, but disagreements between Eisenkot, Bennett and Lieberman over candidate selection keep fragmentation possible.

• Because the first mandate depends on initial recommendations rather than the ability to pass a confidence vote, Netanyahu retains a consequential 40% probability while Bennett is largely relegated to later coalition talks.

• The forecast is heavily concentrated on Eisenkot (54%) and Netanyahu (40%), with Bennett holding 5% and Lapid and other contenders assigned small residual chances.

• Remaining uncertainty turns on whether Arab party disqualifications are overturned on appeal, whether Arab factions endorse Eisenkot, and whether small swing lists cross the electoral threshold.

5. A runoff in Nigeria’s 2027 presidential election ↗

Nigeria has held seven presidential elections since 1999 without a runoff. Will this be the election to break the streak? Explore on Mantic ↗

• With the ruling party controlling 31 of 36 governorships and Bola Tinubu having reached 25% across roughly 30 states in 2023, an incumbent plurality almost certainly meets the constitutional spread rule, weighing heavily against a runoff.

• A fractured opposition makes a runoff possible, but that outcome hinges almost entirely on Peter Obi winning a national plurality without sufficient geographic spread, after clearing the 25% threshold in only 16 to 19 jurisdictions in 2023.

• The models diverged on opposition strength: one predicted a 15% runoff chance by treating Tinubu as an 80% plurality favorite, whereas the other saw a 23% chance by modeling a tighter race where Obi captures nearly 33% of the first-ballot vote.

• My forecast leans decisively toward a first-round victory, but preserves a 19% chance of a runoff to capture the scenario where an opposition plurality lacks sufficient geographic spread.

• Key uncertainties center on whether Rabiu Kwankwaso’s northern base transfers cleanly to Obi, how effectively APC governors deliver their states, and whether the electoral commission requires 24 or 25 qualifying jurisdictions.

The majority of trading volume on Polymarket is concentrated in sports and crypto, but there are also markets on more sober themes such as geopolitics, national and regional elections, central bank decisions, technological developments, and armed conflict. One example is when traffic through the Strait of Hormuz returns to normal.

We’re often asked how Mantic’s accuracy compares to Polymarket. To investigate this, we don’t need Mantic to make trades — instead, we can treat Polymarket’s midpoint price (i.e. what you see on their website) as the Polymarket prediction and compare against Mantic on standard forecasting metrics like Brier score.

The experiment

We have built our prediction engine to be backtestable, meaning it can be run today as though it were some date in the past. For example, we can run our system as though it were April and ask ‘When will Keir Starmer stop being Prime Minister?’

We take Polymarket prices at 10,073 observations from 4,139 markets over 817 resolved events between February and August 2026, sampling at volume snapshots set at open, $1k, $10k, $100k, $1M, and $10M volume, and ran the Mantic system to compare forecasts.2

This post includes five summary findings. Reach out for the full report.

Mantic is significantly more accurate than Polymarket up to $100k of traded volume

We are significantly more accurate3 than Polymarket until a question reaches $100k of traded volume across its markets. At open, we’re closer to the outcome on 84% of markets; our win rate is still 72% at $1k and 62% at $10k. There is not a significant difference in accuracy from $100k but our win rate stays above 50%.

Mantic is better calibrated

Calibration measures how well a forecaster’s odds actually match up with how often things happen.4 We’re significantly better calibrated than Polymarket over questions with at least $1k traded across their markets, with an expected calibration error5 (ECE) of 0.022 against Polymarket’s 0.058, where lower is better.

Subscribe now

Mantic’s edge is largest where price discovery is limited, and grows when it disagrees with the market

Our advantage is strongest for markets with lower volume (top left), lower liquidity (bottom left), and higher bid-ask spread (top right), i.e. where price discovery is limited. We also find the more we disagree, the bigger our advantage (bottom right).

Taking positions in the direction of Mantic’s disagreement yields positive payoffs

Taking a one-share position in the direction of our disagreement and holding until resolution yields +15 percentage points (pp) on average at market prices, or +3.8 pp when limited only to disagreements outside the spread and entering at the bid or ask. Ordered by market volume (left), the cumulative payoff rises at every volume level. By snapshot (right), the payoff is significantly positive through the $100k snapshot.6

Mantic’s forecasts are diversifying to market prices even when its accuracy advantage is no longer significant

What happens if we combine the two? We can mix our forecasts in an ensemble that weights the Mantic prediction with the Polymarket price, choosing the weight that gives the best accuracy.7 At open and $1k, the best ensemble ignores the Polymarket price entirely, and at $10k it puts 96% of the weight on Mantic. At $100k, where the two are equally accurate on their own, the weight splits roughly evenly and the ensemble beats both on held-out events.

The scale of AI forecasting

It’s very hard to durably out-perform highly liquid markets over the long-term (but we’ll try!). Either way, what’s most exciting is the scale of AI forecasting. We’re extending PM-level accuracy to all important issues, free from constraints. Come join the mission!

1. When an autonomous AI system directly causes a human death ↗

Documented destructions of live production systems show agents can act destructively. The next threshold is a death directly caused by an action no human specifically approved. Explore on Mantic ↗

• Existing AI-linked fatalities, including military drone strikes, chatbot-advised suicides, and hospital triage delays, are excluded because the criteria demand an independently executed action directly causing death.

• Evidence of autonomous capabilities, including the Jadepuffer ransomware attack, rogue agent swarms, and nine documented destructions of live production systems, demonstrates a rising operational hazard.

• Weighed against that, physical-access constraints and ongoing physician oversight in clinical pilots, such as Utah’s prescription-renewal program, limit immediate fatal pathways.

• My 80% range spans from 21st April 2027 to 18th July 2036, leaning toward later years with a heavy tail driven by compounding hazard growth.

• The key remaining uncertainty is whether strict direct causation can be proven in complex real-world incidents, leaving open the possibility that a qualifying event never occurs.

2. Congress passes an AI-development ban or pause, conditional on AI-caused death before October 2026 ↗

Sanders and Casar’s new AI-pause bill lacks bipartisan support but a future proposal may be more consensual. A fatal incident could force Congress to act, although fears of conceding the AI race to China would remain. Explore on Mantic ↗

If an AI causes its first human death before 1 October 2026 (“If no”):

• A fatal incident would sharply increase political pressure for action, but major institutional hurdles remain.

• Congressional scheduling limits immediate action, with the House adjourned through the midterms and major AI policy potentially slipping into the next Congress.

• The question requires an actual ban or pause on model development. Deployment restrictions, safety audits and export controls would not qualify.

• The only matching legislative proposal lacks broad bipartisan support or committee progress.

• A post-election lame-duck session or early 2027 emergency rider offers a possible route. Congress’s preference for deployment safeguards over research bans nevertheless limits the prospects.

If the first AI-caused death is on or after 1 October 2026 (“If yes”):

• No qualifying development-ban bill has passed either chamber. The Sanders–Casar proposal lacks committee traction or floor scheduling, while other initiatives focus on deployment, auditing or licensing.

• Congressional recess and the short lame-duck session constrain near-term passage.

• An outright development pause faces industry opposition, concerns about international technological competition and congressional gridlock.

• A severe crisis or national-security incident after 2026 is the forecast’s most plausible trigger for passage.

• International agreements or repeated capability scares could gradually increase pressure in later congressional cycles.

Subscribe now

3. A decommissioned US nuclear plant receives final federal approval to restart ↗

A $1.9bn loan backs the Duane Arnold nuclear plant, but its restart target is 2029. For approval this year, the race rests on Crane’s licensing review, not the availability of funding. Explore on Mantic ↗

• I largely set aside Palisades because the NRC restored its operational status in August 2025, before the November 1, 2025 eligibility cutoff, making Crane Clean Energy Center the leading eligible candidate.

• Crane received permission to possess new fuel in July 2026 and requested safety evaluation sign-offs by November 30, 2026, but the precedent of multi-month regulatory reviews places expected operating reauthorization in early to mid-2027.

• NextEra’s Duane Arnold serves as a funded backstop with a 1.9 billion dollar loan, but its target restart in early 2029 points to an NRC decision around 2028 rather than 2026.

• My forecast is right-skewed with only an 18% chance of approval before the late-2026 deadline, as the distribution centers on spring 2027 and rises to over 80% by the end of 2027.

• The key remaining uncertainties are whether forthcoming NRC actions on Palisades could qualify under an alternative interpretation, or whether licensing reviews and legal contentions delay Crane further.

4. The outcome of Google’s search-monopoly appeal ↗

The government says Google’s default deals foreclosed roughly half the search market. Google argues they are not exclusive, putting the legal treatment of defaults at the heart of its appeal. Explore on Mantic ↗

• The D.C. Circuit’s Microsoft precedent strongly supports upholding the core monopoly-maintenance finding, and historical appeals in the circuit consistently left district court liability judgments intact.

• Weighing against affirmance, Google raises legal challenges arguing that default agreements are not exclusive dealing, echoing decisions like Qualcomm where appellate courts reversed government antitrust victories.

• Aggressive enforcement by the Department of Justice and state co-plaintiffs makes a pre-ruling settlement unlikely, restricting the chance of an alternative resolution to 4%.

• My distribution concentrates 72% of its weight on affirmance, with the remaining possibility leaning toward outright reversal or vacatur at 17% rather than a liability remand at 8%.

• The principal remaining uncertainty is the randomly assigned three-judge panel, whose legal interpretation of causation and default contracts will determine the outcome when a ruling issues around mid-2027.

5. The next country to land humans on the Moon ↗

NASA has two contracted lander providers; China plans a simpler two-launch mission. Tested transport gives the US a head start, but orbital refuelling could erode it. Explore on Mantic ↗

• The American lead rests on flight-tested Orion transport and two contracted lander providers, though Starship orbital refuelling demonstrations, Blue Origin pad repairs, and Axiom suit tests present significant delay risks.

• China offers a credible rival timeline targeting 2029 to 2030 with a simpler refuelling-free architecture, but its Long March 10 rocket, Mengzhou capsule, and Lanyue lander have yet to fly.

• All other national operators collectively hold under 1% because foreign participation on NASA flights does not confer operator credit and independent programs like India’s target much later dates.

• My forecast reflects heavily overlapping landing timelines rather than a decisive lead, with median landing dates separating the two front-runners by only a few months between late 2029 and mid-2031.

• The primary uncertainty is whether complex American orbital cryogenic transfers and spacesuit development slip past China’s simpler two-launch mission before its new flight hardware encounters unexpected setbacks.

AI is the unexpected guest. With models ever more powerful, the public is now concerned by the risks the technology is posing. This gives reason for cooperation as China is often invoked as the main reason why it is impossible for the labs to slow down. A credible agreement would make that argument weaker.

Our five forecasts examine what the summit is likely to deliver on trade, AI and Taiwan.

Will the export-control truce be extended, and for how long?

Why it matters

The duration of the truce affects access to critical minerals and advanced technology, which has consequences for manufacturing costs, supply chains and business investment. A longer extension would give businesses more planning certainty while a short one would make them exposed to renewed confrontation.

Mantic rationale

• Preparatory talks remained split three days before the summit, with Washington publicly offering three to six months while Beijing sought over two years, and the November expiration allows both sides to continue negotiating.

• Weighed against the impasse, pre-summit analyst expectations and the precedent of the 2025 Busan agreement supported an extension of exactly 12 months as a natural compromise.

• One model favoured an extension of less than 12 months based on Washington’s live offer, whereas the other placed the most weight on no qualifying announcement within the September window.

• The main remaining uncertainty is whether the leaders bridge the duration dispute during the summit or issue a general political statement that leaves legal terms for both suspensions unspecified.

Will both countries cut tariffs?

Why it matters

Since Liberation Day, tariffs have had a negative impact on both countries, albeit for different reasons. Farmers, manufacturers but also consumers in general have been especially hit in the US. For Trump, a positive announcement could also help sway impacted voters in the coming midterms.

Mantic rationale

• Negotiations over a symmetric 30 billion dollar reciprocal tariff package covering US energy and agriculture alongside Chinese manufacturing inputs, potentially applying most-favored-nation rates, provide concrete momentum for a bilateral deal.

• Against this, past summits often produced only vague framework language, and historical research shows zero of 15 comparable high-level US-China trade negotiations met such strict specificity criteria.

• The models diverged on the September 20 preparatory talks: one estimated a 65% chance of agreement based on technical readiness, while the other leaned toward No at 45% because negotiators still had not finalized a product list.

• The principal remaining uncertainty is whether any summit announcement publishes explicit bilateral rate cuts below current levels by September 30, rather than defaulting to purchase pledges, truce extensions, or deferred frameworks.

Will AI talks get a concrete agenda?

Why it matters

AI risks have recently become mainstream with many labs asking for help from the government to allow them to pace the speed of progress. An agreed agenda would give the two governments a concrete process for addressing these risks and the labs an horizon to look towards to pace the advances of artificial intelligence.

Mantic rationale

• Preparatory talks between US Treasury Secretary Scott Bessent and Vice Premier He Lifeng on September 20 already identified the bilateral leads and agreed on a dialogue with a Shenzhen follow-up.

• Current public statements propose an AI incident-notification mechanism without explicitly naming an enumerated hazard like cyberattacks, though an uncontradicted US statement during the summit window would suffice.

• Weighing against an explicit agreement, both leaderships have political reasons to avoid threat framing, and only two of five summit-linked channel announcements since 2016 specified both counterparts and an agenda.

• The main remaining uncertainty is whether official summit releases will explicitly enumerate qualifying risks like cyberattacks or military escalation, rather than relying on generic national-security incident phrasing.

Will the summit produce a specific AI safeguard?

Why it matters

A concrete joint safeguard could reduce a shared danger, such as AI-enabled cyber misuse or military escalation. There is now a concrete proposal: on September 20, Bessent proposed notifications for AI incidents affecting national security but China’s acceptance remains unclear. This would show whether cooperation can move beyond statements and meetings while the two countries continue to compete in AI.

Mantic rationale

• Preparatory negotiations placed a concrete AI incident-notification mechanism on the table, which directly matches qualifying examples in the criteria and could be satisfied even by an uncontradicted unilateral assertion.

• However, Chinese readouts acknowledged only general AI dialogue without endorsing the proposal, and merely continuing or operationalizing the dialogue agreed upon in May 2026 is explicitly excluded from qualifying.

• Historical precedent also tempers expectations, as only one of five comparable security meetings produced concrete risk-reduction deliverables, while both leaders’ public postures emphasize technological dominance over safety constraints.

• The primary remaining uncertainty is whether China will privately accept the notification mechanism at the summit, and whether the final joint wording establishes a specific operational commitment rather than vague intentions.

Will the US commit to restricting arms sales to Taiwan?

Why it matters

Taiwan could be more lightly affected, e.g. with delays, but an explicit restriction would mark a stark change in US policy, affecting Taiwan’s access to defense capabilities and how it plans for its security. That, in turn, would send a worrying signal to China and to Asia-Pacific as a whole on the reliability of the US in the region.

Mantic rationale

• Reports indicate the administration is weighing a quiet postponement of pending Taiwan arms sales past the summit, while official testimony denies linkage to China talks and previews center on trade and AI.

• While ten bilateral leader summits since 2013 yielded no qualifying public arms restrictions, Trump’s May 2026 remarks describing a $14 billion package as a negotiating chip held in abeyance provide a concrete precedent.

• The primary remaining uncertainty is whether Trump makes an unscripted statement that establishes a genuinely new commitment to delay or restrict sales rather than merely reiterating existing policy.

2026 opens the era of superhuman forecasting.

This summer, for the first time ever, Mantic beat all 676 human participants, including professional forecasters, in the Metaculus Cup. Only last year, experts believed this achievement was still a decade away.

The Metaculus Cup is the preeminent forecasting tournament: 500-1000 people compete over who can make the most accurate predictions. Participants submit probabilistic estimates on around 60 different world events, and score higher for assigning greater weight to the eventual outcome.

For example, for the Colombian presidential election, Mantic identified Abelardo de la Espriella as the favourite a week before the human community did. Mantic outperformed the crowd in predicting Spider-Man box office revenues, and numerous other questions across geopolitics and technology.

In the same competition, Mantic beat all the other AI startups, and beat all frontier models running on a standard harness provided by the competition organisers. Models like GPT-6 and Kimi K3 provide a strong foundation, but need the right scaffolding and fine-tuning to maximise their forecasting capabilities.

Supercharging the world’s decision-making

Companies, investors, and governments are facing a world of increasing volatility and uncertainty. Against this backdrop, decision-makers need the best possible technology for understanding their environment and acting with confidence.

Mantic exists to serve these people. Our product helps users create and track personalised forecasts across the issues that affect their work, and includes MCP and API access.

Some of the world’s leading hedge funds are already using Mantic to predict market-moving events across the domains of geopolitics, macroeconomics, and business. In particular, Mantic has excelled at predicting the timing of events. For example, Mantic was more pessimistic than oil markets about a reopening of the Strait of Hormuz.

Fortune 500 customers are adopting Mantic to inform their strategic decision-making around M&A and the release of new products. We receive inbound interest across an abundant set of applications, from supply chain management and demand forecasting to executive-level planning.

We’ve raised $25mn to scale

Today, we’re announcing a $25mn fundraise, led by Radical Ventures, with participation from Balderton, Thinking Machines, DRW, FT Ventures, M12 (Microsoft’s venture capital fund), Episode 1, Charlie Songhurst, and Thomas Wolf.

With this capital, we are significantly scaling up the team and the compute and data powering Mantic’s forecasts; and we have growing capacity to work with ambitious partners.

Join the mission

We’re actively hiring across AI research, engineering, product development, sales, and operations.

We’re also onboarding new customers, especially traders in financial markets.

To get involved, please reach out at contact@mantic.com.

1. The number of House seats Republicans win in the 2026 midterms ↗

Democrats lead the generic ballot by about 7 points with Trump’s approval near 38%, yet redistricting across states means they still need a 3 to 4 point margin to take the chamber. It raised the floor without changing the ending. Explore on Mantic ↗

• Republicans entered the cycle defending a narrow majority of 220 seats won in November 2024, meaning a net loss of just three seats hands control to Democrats.

• National generic ballot polling shows Democrats leading by margins between 5.7 percentage points and 8.1 percentage points, creating an adverse national wave environment for the incumbent party.

• President Donald Trump faces deeply underwater approval ratings of 35% to 39%, driven by public dissatisfaction with inflation and the ongoing conflict with Iran that began on February 28, 2026.

• Mid-decade redistricting across ten states created a net increase of 12 Republican-leaning seats and shifted the median district to Cook PVI R+3, providing a structural cushion that prevents a collapse on the scale of 2010 or 2018.

• The official resolution excludes independents who caucus with Republicans, which deducts Kevin Kiley of California from the official Republican tally if he wins as an independent and sets my expectation slightly below raw market indices.

2. Republican House seats conditional on the timing of the Fed’s next hike ↗

August inflation at 3.4%, payrolls up 162,000 and three July dissents put futures near 90% for a September hike. How much does a pre-election hike punish the incumbent party? Our forecaster believes this would cost the Republicans 5 additional seats. Explore on Mantic ↗

• If the Fed hikes before 1 November, the critical pivot in my thinking is conditioning on an interest rate hike occurring shortly before the election. Against a backdrop of depressed consumer sentiment and persistent inflation concerns, a late monetary tightening acts as an overt signal of economic distress.

• I determined this would sharpen voter dissatisfaction with the incumbent party, creating an additional electoral drag, so I shifted the distribution toward lower Republican totals and treated a sustained Republican majority as an improbable upper-tail scenario.

• If the hike instead lands on or after 1 November, taking the absence of a pre-election interest rate hike into account merely confirms my baseline economic backdrop without providing a strong catalyst either way, leaving a balance between a sizable wave loss and map-insulated retention.

• In both branches the baseline assessment rests on generic congressional ballot polling, current presidential approval ratings, prediction markets and nonpartisan race raters. Midterms historically penalize the party holding the White House, especially when executive approval is low, and national polling aggregates point toward strong Democratic momentum.

• I tempered expectations of a complete historical wipeout by accounting for structural cushions: recent mid-decade redistricting has created a more favorable baseline for Republicans, shifting the median congressional district to the right and providing a defensive floor in several tossup races.

• Professional handicappers and race-by-race models reflect this dynamic by projecting a more resilient defense than national polls alone would indicate, which is why an unfavorable national voting environment translates into fewer lost seats than historical wave elections might suggest.

Subscribe now

3. Starship Flight 14 achieves orbit ↗

Thirteen flights, none to orbit on purpose, and Flight 13 already relit a Raptor in space in July. The largest risk on 22 September is not an explosion but SpaceX’s own rule: controllers skip the insertion burn unless the ship can prove it could deorbit. Explore on Mantic ↗

• SpaceX has scheduled the launch for September 22, 2026, during a 75-minute window opening at 12:15 UTC from Starbase, Texas, marking the program’s first intentional orbital insertion attempt.

• The flight profile plans an initial ascent followed by a 19-second orbital insertion burn around 25 minutes after liftoff, inserting the upper stage and 26 Starlink V3 satellites into an altitude of approximately 275 km.

• The vehicle upper stage demonstrated the core propulsion requirements on Flight 13 on July 24, 2026, including nominal ascent cutoff, payload bay deployment, and a successful 14-second in-space Raptor relight.

• The remaining 34% probability of failure reflects strict mission commit criteria where flight controllers will only execute the insertion burn if they verify sufficient redundancy for a subsequent deorbit burn, as well as the historical failure risks inherent in maiden orbital attempts for new rocket architectures.

4. When the Paramount–Warner Bros. Discovery merger closes ↗

Every regulator has cleared this, the DOJ included, so the $111bn deal now turns entirely on twelve state attorneys general in one Oakland courtroom. Trial starts 2 March 2027, and Paramount pays $7m a day from 1 October just to wait. Explore on Mantic ↗

• Consummation is currently blocked by a binding joint stipulation filed on July 24, 2026 in the U.S. District Court for the Northern District of California, which establishes that the transaction will not close until the earlier of five days after the merits determination or June 1, 2027.

• U.S. District Judge Araceli Martinez-Olguin has scheduled a 12-day bench trial running from March 2, 2027, through March 19, 2027, following a final pretrial conference on February 24, 2027, with post-trial briefing extending into April 2027.

• A merits decision issued in mid-to-late May 2027 represents the most likely procedural pathway to satisfy the five-day closing condition under the stipulation, fitting inside the merger agreement outside date which automatically extends to June 4, 2027.

• All government agency approvals have been secured worldwide, including clearance from the U.S. Department of Justice on June 12, 2026, the European Commission on July 22, 2026, the United Kingdom Competition and Markets Authority on August 6, 2026, and Mexican regulators on August 14, 2026, leaving only state and guild litigation.

• Current prediction market pricing aligns with this calendar, assigning only a 21% likelihood to a closing by December 31, 2026, while pricing a 72% probability that Paramount completes the acquisition by June 30, 2027.

5. The IEA forecast for global electricity demand growth in 2027 ↗

The agency’s mid-year update already had demand growth accelerating from 3.0% in 2025 to 3.6% this year, with data centres going from 485 to 950 TWh by 2030. This is where the AI build-out stops being a narrative and becomes a published number. Explore on Mantic ↗

• The baseline for this forecast is established by the agency’s Electricity Mid-Year Update 2026, published on 23 July 2026, which projected worldwide electricity demand to grow by 3.8% in 2027, accelerating from 3.6% in 2026 and 3.0% in 2025.

• Structural tailwinds support robust growth into 2027, with total consumption projected to rise from 28,600 TWh in 2025 to 30,700 TWh by 2027, driven by electric vehicles, heat pumps, air conditioning, and data centers expanding from 485 Terawatt-hours (TWh) in 2025 to 950 TWh by 2030.

• Macroeconomic conditions provide an upward demand impulse, with global gross domestic product growth projected by the International Monetary Fund to strengthen from 3.0% in 2026 to 3.4% in 2027.

• Counteracting upside factors, geopolitical conflict in the Middle East caused the loss of nearly 20% of global LNG supply through the Strait of Hormuz, with damage to Qatari infrastructure eliminating 17% of its LNG export capacity and triggering power curtailments in Pakistan and Bangladesh.

1. The winner of the 2026 Brazilian presidential election ↗

Jair Bolsonaro is barred from running after his conviction, so his son Flávio carries the name — into a runoff the polls now call a dead heat. Does the family’s legal trouble cost votes or buy sympathy? Brazil votes 4 October. Explore on Mantic ↗

• I predict Luiz Inácio Lula da Silva will win the 2026 Brazilian presidential election with 53% probability.

• First-round polling gives Lula a consistent advantage, such as 38% against Flávio Bolsonaro’s 33% in Datafolha, supported by incumbency and procedural milestones like approved candidate registrations ahead of the October 4, 2026 vote.

• Head-to-head second-round polling for the anticipated October 25, 2026 runoff indicates a statistical tie, demonstrated by 41% to 41% in Quaest and 46% for Flávio versus 45% for Lula in BTG/Nexus.

• Runoff upside for both frontrunners remains constrained by rejection rates exceeding 45%, while Polymarket mirrors this razor-thin edge at 53% for Lula as voter-transfer polling indicates right-leaning third-party support consolidates behind Flávio.

2. Brazilian Amazon deforestation in 2029 ↗

Amazon clearing surged under one Bolsonaro and has fallen three years running under Lula, to a series-low 2,874 km² of alerts. Is the forest’s fate just a policy setting the next president can switch back? Explore on Mantic ↗

If Lula wins:

• Recent high-frequency monitoring data indicates a substantial downward trajectory, with near-real-time alerts falling to multi-year lows. This strong momentum, paired with federal policy continuity—such as active environmental enforcement, targeted credit restrictions, and international conservation funding—largely rules out a return to peak historical deforestation levels.

• However, I recognized significant structural barriers that prevent deforestation from dropping to zero. Brazil’s Forest Code legally permits private landowners to clear a set percentage of native vegetation, creating a statutory floor beneath which rates cannot easily fall. Additionally, emerging headwinds—notably the collapse of private grain agreements, rising agricultural expansion in key states, and ongoing staffing vacancies in enforcement agencies—threaten to slow or partially reverse recent progress.

• Ultimately, my thinking balanced the strong momentum of federal enforcement against statutory clearing baselines and growing agricultural pressures, anticipating that annual deforestation will level off near record lows rather than continuing a steep downward path.

If Flávio Bolsonaro wins:

• I anchored my starting point on recent official INPE data and early alert trends, which show clear-cutting reaching near-record lows ahead of the political transition. However, I expect this downward trajectory to reverse sharply under a new administration. Historically, the prior Bolsonaro presidency saw environmental enforcement rolled back, budgets restricted, and annual forest loss surge rapidly within three years. I expect similar dynamics to unfold by 2029, compounded by the recent legal collapse of the private Soy Moratorium, which is projected to unleash substantial additional clearing for agriculture.

• At the same time, I accounted for crucial dampening factors that should prevent clearing from returning to the extreme peaks seen in the early 2000s. These include strict international supply-chain regulations, foreign trade tariffs linked to environmental standards, corporate anti-deforestation commitments, and newly designated protected areas. Balancing rapid policy rollback against external market constraints, my forecast reflects a substantial rebound in deforestation from recent lows while recognizing realistic commercial ceilings.

Subscribe now

3. Brazil’s primary fiscal balance as a percentage of GDP in 2029 ↗

Brazil is running a 1.19%-of-GDP primary deficit where the law demands a 1.25% surplus by 2029, and interest alone eats 8.8% of GDP. Both camps promise discipline — but 90% of the budget is untouchable. Who blinks? Explore on Mantic ↗

If Lula wins:

• My core reasoning centered on reconciling official targets with practical implementation realities under a continued Lula administration. The IMF and Brazilian Treasury establish a baseline of modest primary surpluses, while official government targets point toward a more aggressive fiscal adjustment. However, private economists remain noticeably more skeptical, pointing to rigid mandatory spending obligations—such as social programs, pension commitments, and minimum-wage indexation—that make sharp expenditure cuts difficult to deliver.

• I concluded that policy continuity under the current fiscal framework provides guardrails against severe fiscal deterioration, but ongoing political incentives toward social spending will likely cause the primary balance to lag official targets. As a result, my thinking centered on a gradual, revenue-driven consolidation path that hovers near overall fiscal balance, while incorporating downside risks from potential macro shocks or structural revenue shortfalls, and upside potential from successful tax collection reforms or commodity-driven growth.

If Flávio Bolsonaro wins:

• My analysis was explicitly conditioned on a scenario where Flávio Bolsonaro wins the 2026 presidential election. I anticipate that a right-of-center administration would prioritize fiscal credibility, spending restraint, and market-friendly reforms, pointing toward an improved primary balance compared to current baseline projections. This expectation is supported by the historical precedent of the previous Bolsonaro administration, which achieved primary surpluses outside of the pandemic shock.

• However, I tempered this optimism due to significant structural constraints. Over 90% of Brazil’s federal spending is legally mandatory, and constitutional budget earmarks leave little room for discretionary cuts without substantial congressional cooperation. Furthermore, right-wing administrations often face political incentives for tax relief or populist spending near elections, which could erode fiscal gains. While my central expectation tilts toward a modest surplus achieved through gradual fiscal consolidation, I maintained substantial downside risk to account for chronic underperformance on targets, congressional gridlock, and potential macroeconomic shocks or recessions.

4. The effective tariff rate on United States imports from Brazil in September 2027 ↗

The 25% tariffs on Brazilian goods were aimed at Lula, not at Brazil. So if the opposition wins in October, does Washington still have a reason to keep them? Explore on Mantic ↗

If Lula wins:

• My central scenario assumes the broad continuation of current punitive measures, particularly given the political backdrop of Luiz Inácio Lula da Silva remaining in power, which reduces the likelihood of a near-term bilateral breakthrough. The current policy structure combines Section 232 metal tariffs with newly enacted Section 301 duties. However, the realized effective rate is heavily mitigated by substantial product exemptions covering major export staples such as crude oil, coffee, and aircraft. Furthermore, I factored in dynamic trade diversion: as high duties take effect, U.S. importers progressively substitute away from heavily taxed Brazilian products, increasing the relative share of exempt or lower-tariff goods and suppressing the overall realized duty rate.

• I balanced this baseline against significant upside and downside risks. On the downside, ongoing litigation before the Court of International Trade, potential WTO challenges, and technical carve-outs could invalidate or narrow the newest duties, leaving primarily metal tariffs and baseline rates. On the upside, persistent political friction could prompt retaliatory escalation or the revocation of existing exemptions.

If Flávio Bolsonaro wins:

• Historically, effective tariff rates on Brazilian goods were minimal before rising sharply following recent punitive measures, specifically bilateral, politically charged Section 301 actions alongside broader sectoral duties like metals tariffs under Section 232.

• The central driver of my analysis is the anticipated political rapprochement following a Flávio Bolsonaro presidential victory. Given the close ideological and personal alignment between the Bolsonaro family and U.S. leadership, the primary political motivation behind the country-specific trade penalties would effectively dissolve. With roughly eight months in office before the target date, an aligned administration would have ample time to negotiate significant tariff relief.

• Consequently, I placed the greatest likelihood on scenarios involving substantial de-escalation, where country-specific punitive duties are dismantled. However, I factored in that global sectoral tariffs on metals and standard baseline duties would likely persist as a floor unless a comprehensive bilateral agreement includes explicit exemptions or quotas. I also retained modest probability on bureaucratic delays leaving current restrictions in place.

5. When the next US hyperscaler announces a $1bn+ data centre commitment in Brazil ↗

Microsoft and Amazon committed R$25bn between them in 2024, then stopped. AWS says there is no new money until the vote is settled. Does a Bolsonaro win open the tap, or is it the tax code holding it shut? Explore on Mantic ↗

If Lula wins:

• My core reasoning centers on the removal of political and regulatory overhangs. The Brazilian Senate recently passed REDATA, a major federal tax incentive package designed to attract data center investment. While Microsoft recently hinted at imminent announcements following these legislative advances, industry reporting indicates that US hyperscalers are largely postponing high-profile commitments until after the Brazilian presidential election and the US midterms.

• Under the stipulated condition of political continuity via a Lula election victory, I expect REDATA to receive swift presidential sanction, clearing the primary federal roadblock. Coupled with immense, record-level global AI and cloud infrastructure budgets across Amazon, Microsoft, and Alphabet, I believe conditions will become highly favorable shortly after the elections conclude.

• Consequently, I focused my timeline heavily on the late 2026 to first-half 2027 window. I also accounted for potential delays extending into late 2027 or beyond, primarily driven by lingering negotiations over state-level tax exemptions, power availability, or corporate decisions to lease third-party colocation facilities rather than making standalone, billion-dollar public commitments.

If Flávio Bolsonaro wins:

• A central factor in my thinking is the explicit signal from industry leadership that US tech giants are delaying high-profile capital deployment until after the political uncertainty of Brazil’s national elections and the US midterms clears. I weighed this against pending legislative milestones, notably the advancement of the REDATA federal tax framework and unresolved state-level tax reliefs, which have historically presented a major cost hurdle for imported equipment.

• Conditioned on a Flávio Bolsonaro victory, I expect an accelerated timeline driven by an incoming administration likely to offer a friendlier investment climate, regulatory clarity, and tighter diplomatic alignment with the United States. While massive global cloud budgets and hints of near-term announcements provide strong underlying pressure, near-term pre-election announcements remain unlikely. Instead, I see the highest likelihood of a major commitment clustering around the post-election transition and the first half of the new presidential term, when tax decrees can be finalized and ceremonial investment pledges made, tapering into a multi-year background rate as capacity expands.

1. How The New York Times v. OpenAI and Microsoft copyright case is resolved ↗

The Times has spent $28m so far, and a sanctions motion over allegedly destroyed OpenAI training logs is due 10 September. With Anthropic’s $1.5bn settlement setting the price of walking away, will this reach a verdict? Explore on Mantic ↗

• I predict a 54% probability that The New York Times v. OpenAI and Microsoft copyright case will reach a settlement before a trial verdict is issued.

• Intense procedural pressure from a July 2026 sanctions motion alleging that OpenAI destroyed training data logs makes a pretrial resolution highly attractive to avoid adverse evidentiary rulings.

• The New York Times has already spent $28 million on litigation, creating a strong financial incentive to convert these costs into a revenue-generating licensing deal similar to the $1.5 billion Anthropic benchmark.

• OpenAI’s existing licensing agreements with the Associated Press and News Corp establish a viable commercial path for a resolution, even as the Times continues to aggressively refine its claims in the August 21, 2026 amended complaint.

2. When the IAEA next inspects Natanz, Fordow or Isfahan ↗

Iran has kept inspectors out for over 14 months, leaving 440.9 kg of 60% enriched uranium unverified, and declared “wartime conditions” on 26 August. The IAEA board meets 21 September, so when does anyone get back in? Explore on Mantic ↗

• My 25th percentile prediction for an IAEA on-site inspection at Natanz, Fordow, or Isfahan is 31st October 2027.

• I expect the current diplomatic impasse to persist after the Islamabad Memorandum framework collapsed in July 2026, leaving no active roadmap for restoring oversight.

• I believe that “wartime conditions” declared by Iranian leadership on August 26, 2026, and the requirement for Islamic Consultative Assembly approval create significant political and security barriers to access.

• I anticipate technical delays because physical damage to Natanz necessitates a new inspection protocol to manage safety and verification at previously targeted facilities.

• I note that market signals show an 88% probability of no inspection by late 2026, confirming a consensus that the standoff will extend well into 2027.

  Subscribe now

3. When the IAEA next inspects Iran’s nuclear sites if the fighting stops ↗

The two charts below condition on when the Iran–United States war first goes 30 consecutive days without a direct kinetic attack, at a January 2027 threshold and a November 2026 one. Explore on Mantic ↗

If the fights stop by January 1, 2027:

If the fights stop by November 1, 2026:

• I analyzed current reports showing that IAEA inspections at Natanz, Fordow, and Isfahan have been halted for over a year following military strikes. Iran has explicitly stated that any return to these sites requires a new, specific security protocol and parliamentary approval, rather than a simple return to previous arrangements.

• I weighed the “outside view”—market prices and historical delays—against the “inside view” that a ceasefire would likely be bundled with or quickly followed by nuclear concessions to secure sanctions relief. While a ceasefire removes the immediate excuse of “wartime conditions,” I expect a lag of several months for the negotiation of new inspection protocols and the physical preparation of damaged sites.

• Consequently, my reasoning prioritizes the period shortly after the ceasefire takes effect, particularly aligned with IAEA Board of Governors meetings in 2027, which often serve as catalysts for diplomatic breakthroughs. We also accounted for a “long tail” scenario where, despite a ceasefire, political friction or site damage leads to a protracted stalemate lasting years.

4. When Japan next intervenes to buy yen ↗

Tokyo spent a record ¥15.4tn buying yen in July and August, its first joint operation with the US Treasury in three decades. With the dollar back at 160 and a September BoJ hike 92% priced, when does the MOF return? Explore on Mantic ↗

• My median prediction for the next confirmed Japanese yen-buying intervention is 26th November 2027.

• I expect a prolonged pause in direct market action as the Bank of Japan assumes responsibility for currency defense through monetary policy normalization, with markets pricing a 92% probability of a rate hike in September 2026.

• The record ¥15.4 trillion intervention in mid-2026 established a psychological floor near 160, but the depletion of foreign exchange reserves to $1.287 trillion necessitates a long observational period before another physical market entry.

• I anticipate that future action will depend on U.S. Treasury coordination, which currently faces a high threshold as U.S. officials view recent moves as “pretty well contained” following the first joint currency operation in nearly three decades.

5. The gold price at the end of October 2026 ↗

Gold settled at $4,464 on 31 August after central banks bought a record 289 tonnes in the second quarter. But Kevin Warsh has turned hawkish and a September hike is 60% priced, so where does October close? Explore on Mantic ↗

• My most likely prediction for the gold price at the end of October 2026 is 4,485.69 USD per troy ounce, with a 50% chance of falling between 4,200.00 and 4,720.00 USD.

• I expect sustained safe-haven demand to counter monetary headwinds as the 2026 Iran War maintains a high geopolitical risk premium, further supported by record central bank net demand which hit 289 tonnes in Q2 2026.

• Recent hawkish signals from Fed Chair Kevin Warsh have introduced significant downside risk, with a 60% probability of a September rate hike likely to cap near-term gains despite China’s 21-month buying streak.

• I have aligned my forecast with institutional targets such as Goldman Sachs’ $4,900 year-end projection, while accounting for the market’s current contango structure where December 2026 futures trade at $4,529.90.

1. When an EU state next reintroduces border controls aimed at another member state ↗

Italy’s NB 499 filing on 3 August was the first Schengen notification to name a fellow member state, and Spain retaliated five days later. Broad German and Austrian controls lapse on 15 September, so who names a neighbour next? Explore on Mantic ↗

• My 25th percentile prediction for when an EU/Schengen state will next notify the European Commission of a new reintroduction of internal border controls targeting another named member state is 17th November 2026.

• I believe the 3rd August 2026 notification (NB 499) by Italy against Spain has normalized the practice of explicitly naming member states in the NB register, significantly lowering the diplomatic cost for future targeted actions.

• I expect potential triggers for new notifications in late autumn 2026, coinciding with the 15th September expiration of broad controls for Germany and Austria, as states are increasingly pressured to provide more specific, rather than generic, threat justifications.

• I see a high risk of escalation from Northern European states like Denmark, which has already signaled readiness for total border closures following the unprecedented Ceuta crisis that saw 80,000 individuals enter the enclave.

2. The average Q4 2026 price of jet fuel if the Strait of Hormuz stays closed ↗

Gulf Coast jet fuel settled at $3.51 a gallon in early August while Hormuz transits ran at 6.07 ships a day, and distillate stocks sit 10% below their five-year average going into winter. If the strait is still shut, what does Q4 cost? Explore on Mantic ↗

• My most likely prediction for the average Q4 2026 price of jet fuel is 4.27 USD/gallon, with a 50% chance of falling between 3.72 and 4.52 USD/gallon.

• This forecast relies on a sustained blockade pushing Brent crude to 120 USD per barrel. I expect jet fuel crack spreads to exceed 50 USD per barrel because the loss of Middle Eastern refinery exports has severely constrained global supply.

• Distillate inventories are 10% below historical norms, providing no cushion for Q4 heating demand. I expect prices to revert to the 4.20 to 4.50 USD range once the market fully accounts for the indefinite nature of the July 12 closure.

• Daily transits through the Strait of Hormuz remain at just 6.07 vessels, far below the 20-ship threshold necessary for normalization. I believe the combination of this logistical blockade and reduced global refining capacity will maintain price pressure despite potential demand destruction.

Subscribe now

3. The average year-over-year change in the US CPI airline fares index in Q4 2026 conditional on jet fuel hitting $5 ↗

Jet fuel averaging $5 a gallon in Q4 is roughly a 10% scenario, and fares are lapping a soft late-2025 base with fuel near a fifth of airline costs. If that $5 print lands, how hot does the fares index run? Explore on Mantic ↗

• I analyzed the “denominator”—the airline fare index levels from late 2025. Because those values were relatively low, even if airfares stayed flat from their mid-2026 levels, the year-over-year change would already be positive.

• I examined historical data and recent SEC filings from major carriers like United, Delta, and American. These documents consistently showed a strong intent and ability to pass higher fuel costs onto consumers to maintain profit margins. Given that fuel typically accounts for about a fifth of operating expenses, a jump to $5 per gallon creates massive upward pressure on ticket prices.

• I accounted for industry-wide supply issues, such as aircraft delivery delays and carrier exits, which limit the number of available seats. This lack of supply makes it easier for airlines to maintain high fares even if some travelers are deterred by the cost.

• While I considered “demand destruction”—the idea that people might stop flying if prices get too high—the necessity of fuel cost recovery in a constrained-supply environment led us to expect substantial year-over-year increases.

4. The outcome of the Surface Transportation Board’s review of the Union Pacific–Norfolk Southern merger ↗

Union Pacific’s $85bn bid for Norfolk Southern sits in abeyance while the Board digests a 400-page supplemental filing from July, with seven state attorneys general and rival railroads against it. Does the Board clear it, and on what conditions? Explore on Mantic ↗

• I predict there is a 68% probability that the Surface Transportation Board (STB) will issue a final approval subject to substantive conditions for the Union Pacific and Norfolk Southern merger by December 31, 2027.

• Approval is likely tied to extensive voluntary commitments including gateway pricing, service oversight, and a binding MOU with Canadian National that addresses key competitive concerns.

• The current procedural abeyance allows for the review of massive supplemental data filed in July 2026, putting the board on a timeline similar to the recent 16.5-month CP/KCS precedent.

• Significant hurdles remain under strict 2001 merger rules, as seven state attorneys general and rival railroad coalitions challenge the deal, while the Environmental Impact Statement process poses a 16% risk of pushing the final decision into 2028.

• Strong capital positions and a $2.5 billion reverse termination fee make abandonment highly unlikely, signaling management resolve to finalize the transcontinental network.

5. When Meta and Anthropic sign a cloud computing deal ↗

Anthropic was reported in July to be in early talks to lease up to $10bn of Meta’s compute, and Meta has since built a compute division on $130–145bn of 2026 capex. It just signed a rival $10bn deal with Volta, so when does this one land? Explore on Mantic ↗

• My 25th percentile prediction for Meta and Anthropic to finalize a cloud computing infrastructure deal is 16th February 2027.

• I predict Meta’s shift toward commercial cloud services, signaled by the new Meta Compute division and former AWS leadership, drives this timeline. Meta’s projected 2026 capex of $145 billion creates the surplus capacity required for the rumored $10 billion lease.

• I expect Anthropic’s confidential IPO filing to accelerate deal finalization as the company must prove compute scalability to investors. Although talks were preliminary in mid-2026, securing Meta’s infrastructure is a strategic priority alongside their existing $10 billion Volta deal.

• I anticipate that regulatory monitoring by the CMA and internal competition for capacity will prevent an earlier agreement. Anthropic’s new partnerships with Macquarie and GIC provide them leverage in negotiations, likely delaying a final binding contract until the first quarter of 2027.

  Thanks for reading! Subscribe to keep up with Mantic.

Here’s a question that drove a lot of disagreement in the tournament: how many million net barrels of crude oil will be drawn from the U.S. Strategic Petroleum Reserve (SPR) during May 2026?

The forecasters disagreed about how to extrapolate the recent trend in drawdowns from the SPR: some saw the drop as the start of a ramp up, others thought it was a blip that would regress back to the mean. Differences in opinion of this kind, once we know who was right, are the fuel for improving AI forecasting.

Series 1 is wrapping up. Congrats to our top-scoring question askers:

1. Michael Vella (mrvella14): $1647

2. Marcos Ortega (MarcosO): $1606

3. Nicolò Bagarin (Nicolò_Bagarin_404_NOT_FOUND): $1461

4. Phillip Godzin (pgodzin): $1330

5. Capital (Capital): $1313

We’ve had lots of positive feedback from entrants.

The tournament was super (!) insightful into the frontier of forecasting…

Benjamin Yeoh, global equities investor.

We asked the highest-scoring question-writers what they found worked for generating disagreement and distilled these into the following four strategies.

Subscribe now

1. Test information asymmetries

The AI forecasters disagreed when the forecast relied on niche or difficult-to-retrieve data. This provides a strong incentive for the developers to expand their data access.

The bots … disagreed more when high-quality information was sparse, such as niche indices, data behind logins, or values published atypically (e.g., inside an image rather than a table). However, pushing this too far caused the bots to default to similarly conservative estimates, eliminating the disagreement.

Nicolò Bagarin, Superforecaster.

The forecasters were asked how many models on Hugging Face Hub will have more than 500 downloads in the 30 days ending June 30, 2026? And told the question would be resolved using the Hugging Face API.

The Preseen system was able to directly access the API and realised the resolution was overwhelmingly likely to resolve above the upper bound at 22,000. Two other systems were able to access the required data indirectly, whilst the other forecasters all failed to find useful information. The difference in data access split the field.

2. Identify a regime shift

The participants were incentivised to find fast-moving areas, where there’s a reason to believe there’ll be a break from the trend. These don’t have obvious answers, leading to higher disagreement.

For questions where you can get most of the way with only public, easily accessible information, the questions that worked best related to some qualitatively new change, for example some kind of regime change with no clear precedent. Models couldn’t fall back on a shared base rate, or disagreed about what the base rate should even be.

My CVE count question [about cyber vulnerabilities] got a lot of disagreement because models differed in how much the release of Claude Fable would change things, and some didn’t factor it in at all since it wasn’t mentioned in the question background and it’s not obvious to go search about new AI model releases when asked about the number of vulnerabilities some tracker will show in the next month.

Suhas Hariharan, Quant Developer.

The question Suhas is referencing resolved very quickly which caught some forecasters by surprise:

3. Require decomposition

Some participants probed the AI forecasters’ ability to break down a forecasting problem into sub-questions, finding variance in the approaches taken.

My highest scoring question asked on what date the model holding the highest mean reward on the Long-Horizon Terminal-Bench leaderboard (as of a fixed date) was publicly announced.

To answer the question a system first has to take a view on which model (or which lab’s upcoming release) tops the leaderboard, and the different candidates map to announcement dates far apart from each other. So the question is really asking systems to pick between a few disparate worlds, and when they commit to different worlds their distributions barely overlap.

Suhas Hariharan, Quant Developer.

4. Exploit bugs

Some participants were effective at sniffing out edge cases where the AI forecasters, due to technical problems, failed catastrophically.

The most effective strategy was to find and exploit small bugs in how bots translated their predicted probability distributions into probability density functions.

Nicolò Bagarin, Superforecaster.

Nicolò benefitted from this strategy in his question asking what seasonally adjusted labor-force participation rate will the BLS report for U.S. women ages 25–54 in its August 7th, 2026 release?

One forecaster had a bug in converting their belief about the outcome into a probability distribution over discrete buckets, resulting in outputting low probability mass in four bins. This registers as large disagreement with the rest of the field.

Epistemic disagreement?

Otherwise, some participants commented that most of the disagreement was not from differences in reasoning, and they’d like the tournament to surface these differences.

Much of the divergence resulted from various types of forecaster noise rather than genuine epistemic disagreement though. With the bots improving and the format being refined, I look forward to Series 2 and beyond to surface more of the disagreements the format is designed to find.

Michael Vella, AI engineer.

There were some minor things with the reward function being hackable in some ways, which led to a bit of a meta around some qualitatively less interesting questions (in my opinion). But it seems like this is being rapidly iterated on and will be worked out by the next tournament with new and better reward functions.

Suhas Hariharan, Quant Developer.

I was ultimately surprised by the bots' ability to navigate complex forecasting problems using reasonable assumptions, meaning they rarely showed significant disagreement on genuinely difficult questions.

Nicolò Bagarin, Superforecaster.

Overall

Overall feedback was positive though we still have work to do on best aligning incentives and we’re excited to continue innovating in this space. Sign up for Series 2 here!

I normally have to wait weeks or months for a forecast to resolve, but here I could see my disagreement score in a couple of hours. This increased my ability to iterate quickly.

Marcos Ortega

Overall thought this went very smoothly and I think has created a fascinating benchmark.

Suhas Hariharan, Quant Developer.

Tournaments like this help expose the boundaries of [LLMs’] forecasting capabilities and highlight clear paths for improvement.

Nicolò Bagarin, Superforecaster.

The immediate feedback of disagreement score without needing to wait for resolution is definitely unique and useful as a question writer.

Phillip Godzin

I personally disliked the disagreement score incentives this tournament, which incentivised repeating questions in which one or two bots get significantly wrong. Although I believe Mantic has addressed this with their new scoring system in Series 2.

Capital

Overall, I remain excited by Crucible’s potential. Having an adversarial environment with financial incentives for the actors is powerful, and a public forecasting setup like this is unique as far as I’m aware. Series 1 successfully exposed bot weaknesses and personally helped me patch issues with date forecasts and source retrieval.

Michael Vella

1. The combined capital expenditures of Alphabet, Amazon, Meta and Microsoft in calendar Q4 2027 ↗

Combined quarterly spending already hit $165bn in Q2 2026, and consensus for 2027 cloud capex has reached $1.2 trillion, most of it locked into long-term commitments. Power bottlenecks could still push the cash into 2028, so how big does a single quarter get? Explore on Mantic ↗

• My most likely prediction for the combined capital expenditures of Alphabet, Amazon, Meta, and Microsoft in Q4 2027 is 346.25 billion USD, with a 50% chance of falling between 300.00 and 382.00 billion USD.

• This forecast reflects an AI infrastructure supercycle where market consensus for 2027 cloud spending has reached $1.2 trillion. Combined quarterly spending already hit $165 billion as of Q2 2026, as demand for AI workloads consistently outstrips available compute capacity.

• The scale of these outlays is anchored by nearly $2.4 trillion in combined long-term purchase commitments and data center lease obligations, suggesting spending is increasingly locked into contractually-mandated execution phases.

• I accounted for accounting shifts like Microsoft reclassifying $15 billion to operating leases. While corporate profitability is robust, physical bottlenecks like power shortages could shift the timing of these cash outflows into 2028.

2. US real GDP growth over the 4 quarters to Q4 2027 if the AI data center capex is lower than anticipated ↗

If AI infrastructure spending is what is holding growth up, a capex slowdown should be visible in GDP. This is where that link gets priced instead of argued about. Explore on Mantic ↗

• My most likely prediction for US real GDP growth over the four quarters ending Q4 2027 is 2.24%, with a 50% chance of falling between 1.45% and 2.99%.

• I view massive capital expenditure on artificial intelligence infrastructure as a critical tailwind, with AI-related spending reaching nearly 2% of total U.S. GDP and fueling up to $5.5 trillion in total investment through 2030.

• I expect growth to be tempered by a significant fiscal cliff in 2027, as scheduled cuts to Medicaid estimated at $900 billion over the next decade begin to dampen consumer spending power and labor demand.

• I anticipate the Federal Reserve will pivot to a more accommodative stance, likely delivering 50 basis points in rate cuts as inflation normalizes, though I remain cautious of energy market volatility following the spike in oil prices caused by regional conflict.

Subscribe now

3. When China and the Philippines get into a military clash ↗

Second Thomas Shoal is the likeliest place a US treaty obligation gets tested this decade. The timing matters more than the odds, because it sets how long anyone has to prepare. Explore on Mantic ↗

• My median prediction for a military clash between China and the Philippines is 21st September 2027.

• The structural decay of the BRP Sierra Madre at Second Thomas Shoal is the primary trigger, as its impending collapse forces the Philippines into reinforcement missions that Beijing has pledged to block.

• Operational density reached record levels in 2026, with Chinese Coast Guard ship-days at Scarborough Shoal nearly doubling to 156 per month, significantly raising the likelihood of a violent collision.

• I believe the recognition of Chinese martyrs and the deployment of the Typhon missile system have shortened the escalation timeline by increasing domestic political pressure and strategic friction.

• Recent transitions to hand-to-hand combat and boarding actions show that both sides have abandoned non-violent standoffs for a high-risk gray zone of physical conflict.

4. Starship Flight 14 achieves orbit ↗

Every Starship flight so far has deliberately stopped short of orbit. Clearing that line is what turns a test programme into a working launch business, and Starlink V3 depends on it. Explore on Mantic ↗

• I predict Starship Flight 14 will achieve orbital insertion, defined as an osculating perigee of at least 80 km, with a 68% probability.

• This mission marks a critical shift from suborbital testing to operational deployment, as Flight 14 is intended to place Starlink V3 satellites into a stable orbit rather than the transatmospheric paths used in previous flights like Flight 13.

• Transitioning to the Block 3 architecture with Raptor 3 engines provides improved thrust-to-weight ratios and reduced weight, granting the performance margins necessary for the first successful circularization burn in the program’s history.

• While technical reliability is high, the final FAA launch license for the new orbital reentry path is still pending, and any engine relight failure during the upper-stage burn remains a primary risk factor for mission success.

5. When a major U.S. city first has 10% of its taxis be autonomous vehicles ↗

Robotaxi progress gets reported in ride counts and funding rounds. Share of a city’s actual taxi fleet is the number that would show drivers being displaced at scale. Explore on Mantic ↗

• My 25th percentile prediction for when a major U.S. city will first have 10% of its taxis be autonomous vehicles is 26th November 2027.

• I anticipate this milestone will be reached as Waymo aggressively scales its fleet following a significant $16.0 billion investment from Alphabet and the recent import of over 3,200 new vehicles to support densification in mature markets like San Francisco and Phoenix.

• While Waymo already captures 15% to 19% of rider spending in its primary hubs, I expect the transition to a 10% active vehicle share will be slower than revenue growth because robotaxis must directly substitute for a large pool of 5,000 to 10,000 human-driven TNC vehicles per city.

• I expect rapid deployment in Texas cities like Austin and Dallas due to House Bill 100, which limits local municipal restrictions, alongside the market entry of the Tesla Cybercab and Zoox’s expansion under federal exemptions.

1. When shipping through the Strait of Hormuz recovers for at least a month ↗

Hormuz traffic is still running under 5% of the daily transits the strait handled before the war, and mines and war-risk premiums stand in the way of recovery. When does a third of that volume hold for a full month? Explore on Mantic ↗

• My median prediction for Strait of Hormuz transits to recover to a 7-day moving average of at least 50 for 30 consecutive days is 24th November 2027.

• Current commercial traffic remains at just 4.43 transits daily, roughly 6% of pre-war volumes. A mandatory 30-day demining phase required for safe passage ensures that the recovery clock cannot begin for at least a month following any successful diplomatic accord.

• Prohibitive insurance premiums of 7.5% of hull value and Lloyd’s clauses that terminate cover for vessels paying Iranian fees create a structural deadlock. I anticipate normalization of these financial frameworks will take several months beyond a diplomatic settlement.

• Data reliability is hampered by 62% of tankers operating with disabled AIS transponders. I expect the required AIS-based recovery threshold to be met only after regional security stabilizes enough for carriers to resume standard reporting practices.

2. The number of Bundibugyo Ebola cases in 2026 ↗

With 4,470 confirmed cases, this is already the second-largest Ebola outbreak on record, and roughly 580 more are being added every week in an Ituri province where armed conflict keeps contact tracers out. How high does the 2026 count climb? Explore on Mantic ↗

• I predict the most likely number of confirmed 2026 Bundibugyo Ebola cases is 8,000, with a 50% chance of falling between 6,240 and 12,855.

• With 4,470 confirmed cases as of August 12, this is the second-largest Ebola epidemic in history, concentrated in Ituri Province where over 80% of infections occur.

• Transmission is largely driven by community spread, as 60-70% of new cases occur outside contact lists and M23 armed conflict prevents health workers from reaching high-risk zones.

• My forecast assumes the weekly growth of 580 cases will slow by late 2026 due to Phase 3 vaccine trials and increased international funding, though lack of a specific Bundibugyo vaccine remains a major risk.

3. A Bundibugyo Ebola case native to the US conditioned on the total number of cases ↗

Boarding bans and medevac rules divert the most likely routes for the virus to reach US soil, but a bigger epidemic means more chances to slip through. How much does the risk shift between a 15,000 and a 25,000-case outbreak? Explore on Mantic ↗

Number of 2026 Bundibugyo Ebola cases over 25,000

• I used the 2014 West Africa Ebola epidemic as a primary historical anchor. During that event, despite its massive scale, only four cases were laboratory-confirmed within the US. I adjusted this baseline by weighing the current strict travel controls, including boarding prohibitions for non-citizens from affected regions and mandatory health screenings for returning residents. I also considered the question only tracks cases first confirmed in the US, infected humanitarian workers medevaced to and diagnosed in third countries (like Germany) are excluded.

• The crux of my reasoning balances the high likelihood that these containment measures will keep the count at zero against the mathematical risk inherent in a 25,000-case outbreak. Such a large epidemic suggests the virus could spread to neighboring countries, potentially bypassing specific travel bans. While a traveler could unknowingly import the virus during its incubation period, I expect US hospital protocols to isolate any such case immediately, preventing the kind of secondary clusters seen in the past.

Number of 2026 Bundibugyo Ebola cases over 15,000

• My primary point of reference was the 2014 West Africa Ebola epidemic, which resulted in a handful of cases within the United States. However, I identified several critical differences between that event and the current scenario. Key to my thinking was the implementation of much stricter travel protocols, including 21-day flight bans from the primary outbreak regions and mandatory screening at designated U.S. airports.

• While these measures significantly reduce the likelihood of an undetected importation, I recognized that a 15,000-case scenario implies a significant failure in global containment. This would likely lead to spread in neighboring African nations where travel restrictions might be less robust or slower to implement. Consequently, while we lean toward zero domestic cases, we accounted for the possibility of a single importation or limited secondary transmission, while remaining skeptical of a large-scale domestic cluster.

Subscribe now

4. The Bank of Japan raises its policy interest rate above 1% in 2026 ↗

The Bank of Japan’s July minutes show members pushing for a faster pace, with one formal dissent calling for an immediate move to 1.25%, and spring wage talks settled at a 31-year high. Does the policy rate clear 1% before the year is out? Explore on Mantic ↗

• I predict the Bank of Japan will raise its policy interest rate strictly above 1% in 2026 with an 89% probability.

• The central bank’s board has shifted significantly hawkish, with the July Summary of Opinions revealing multiple members advocating for faster hikes and a formal dissent by Hajime Takata proposing an immediate move to 1.25%.

• Robust wage growth serves as the primary catalyst, as June 2026 nominal wages climbed 3.4% and spring negotiations secured a 31-year high increase of 5.37%, validating the virtuous cycle between wages and prices.

• External and inflationary pressures remain high, with 86% of economists forecasting a 1.25% rate and the July 2026 Outlook Report projecting CPI will accelerate clearly above 2%, necessitating further normalization despite potential political wariness regarding debt-servicing costs.

5. The directional guidance SK Hynix provides for its 2027 capital expenditure during its Q3 2026 earnings call ↗

SK Hynix shares suffered a downturn in July as investors started to doubt how long the AI spending boom can last. Will management still promise to spend more next year, or keep quiet until it has to? Explore on Mantic ↗

• I predict a 56% probability that SK Hynix will not provide explicit directional guidance for its 2027 capital expenditure during the Q3 2026 earnings call.

• Historical precedent supports this non-disclosure, as management has withheld directional capex comparisons in 8 of the last 10 Q3 calls, typically opting to focus on operational targets like HBM4E ramp-ups and sold out capacity status.

• Significant pressure from a 48% share price drawdown and investor demands for bigger payouts favors a cautious tone regarding future capital outlays over providing a specific higher spending floor.

• A 25% probability for higher guidance remains possible due to board-approved KRW 54.3 trillion fab investments and CEO warnings of a severe supply crunch by 2027.

1. The number of seats the Republican Party holds in the U.S. House of Representatives after the 2026 midterm elections ↗

The war with Iran and a slowing economy have dragged Trump’s approval to a record-low 34%, and Democrats now lead the generic ballot by six points. Redrawn maps give Republicans a cushion, but how deep do the losses run? Explore on Mantic ↗

• My most likely prediction for the number of seats won by the Republican Party in the 2026 midterm elections is 201.43, with a 50% chance of falling between 187.80 and 216.82.

• I anticipate severe political headwinds will drive significant seat losses, as the party currently faces a record-low presidential approval rating of 34% and a slowing economy where second-quarter GDP growth dipped to 1.5%.

• I expect the Republican Party to retain a structural floor despite these challenges, thanks to mid-decade redistricting that provided a net gain of 10 additional red-leaning seats in key states like Texas and Florida.

• I have incorporated market signals and handicapper data showing a likely Democratic takeover, with Polymarket data and the Crystal Ball ratings suggesting Republicans will fall approximately 17 seats short of their current 220 seat baseline.

2. The number of seats of the Republican Party in the House of Representatives after the midterms conditional on the price of the WTI crude ↗

Oil is the fastest route from a foreign war to a domestic ballot, and voters feel it at the pump long before they see it in an inflation print. How much does the House forecast move if WTI hits $85? Explore on Mantic ↗

WTI Crude price: above $85

• By conditioning my forecast on oil prices remaining elevated through late October, I assumed that voters would face persistent inflation and high costs at the pump leading directly into the election. This environment generally penalizes the incumbent party, as pocketbook issues are highly salient to undecided voters and can dampen turnout for the president’s base.

• I analyzed the gains Republicans achieved through mid-decade redistricting, which created more favorable district lines in several states. I also factored in the party’s significant fundraising advantage and recent legal changes allowing for more coordinated campaign spending.

WTI Crude price: below $85

• I reasoned that avoiding an energy-price shock shortly before the election would stabilize consumer sentiment and reduce the political damage caused by inflation. This economic stability acts as a brake on the opposition’s momentum, making a massive landslide against the incumbent party less probable. By combining these historical headwinds with the moderating effect of stable energy prices and redistricting advantages, I arrived at a forecast that balances the typical midterm disadvantage against these specific protective factors.

Subscribe now

3. The amount of global insured losses from natural catastrophes in 2026 ↗

First-half losses came in at just USD 44 billion, some USD 22 billion below the five-year average, and a strengthening El Niño should keep the hurricane season quiet. Does the second half stay that calm? Explore on Mantic ↗

• My most likely prediction for global insured natural catastrophe losses in 2026 is 100.64 USD billion, with a 50% chance of falling between 91.72 and 112.66 USD billion.

• Actual losses for the first half of 2026 reached only USD 44 billion, tracking USD 22 billion below the five-year average for H1 and providing a favorable starting position for the annual total.

• A strengthening El Niño is expected to suppress North Atlantic hurricane activity, with meteorological forecasts indicating a likely below-normal season that reduces the primary driver of catastrophic second-half payouts.

• I anticipate that structural inflation, including a 60% rise in reconstruction costs and higher asset values, will keep the loss floor elevated even if major event frequency remains low.

• I remain cautious of the “Super El Niño” peak in Q4, which creates significant risks in Asia-Pacific for typhoons and floods that could trigger the forecast upper-bound scenarios.

4. When Benjamin Netanyahu is no longer Prime Minister of Israel ↗

Israel votes on 27 October with Netanyahu’s bloc polling right on the edge of a majority. He could then become a caretaker PM or he may well hang on. When does he actually leave office? Explore on Mantic ↗

• I predict Benjamin Netanyahu will cease to be the Prime Minister of Israel by January 1, 2027, with a 36% probability.

• The primary driver for this forecast is the timing of the national election on October 27, 2026, which occurs just seven days before the resolution date. Israeli law dictates that a sitting prime minister remains in a caretaker capacity until a successor is sworn in, a process that has historically required a minimum of 20 days.

• Although recent polls show Netanyahu’s bloc falling short of a majority with only 48 to 53 seats, the fragmented nature of the opposition makes an unprecedented one-week transition of power highly improbable.

• The low probability of departure accounts for extreme outliers, such as a sudden medical emergency or an immediate resignation resulting from ongoing plea-deal mediation regarding his corruption trial.

5. When Safe Superintelligence (SSI) releases their first model ↗

An investor recently mentioned on a podcast that SSI would release their first model in August. CEO Ilya Sutskever had said in the past that they would not release any product before reaching a safe superintelligence. Explore on Mantic ↗

• My 25th percentile prediction for Safe Superintelligence (SSI) to release its first AI model is 13th October 2026.

• I based this timeline on investor reports from August 2026 regarding a planned summer or autumn release, but I have pushed the prediction to October to account for standard technical slippage and safety red-teaming common in frontier model development.

• I have integrated the 5 billion dollar investment from Nvidia and the Q3 2026 production timeline for the Vera Rubin hardware into my forecast, as these represent the primary logistical constraints for a public launch.

• I predict that Ilya Sutskever’s July 2026 confirmation of research worthy of scaling up suggests the firm has reached a maturity level that permits a public demonstration, even as they maintain a straight shot strategy that avoids typical commercial APIs.

1. Marine Le Pen’s vote share in the second round of the 2027 French presidential election ↗

After her judgement was overturned, Marine Le Pen said she would be the presidential candidate for the Rassemblement National. Explore on Mantic ↗

• I observe that current polling consistently places her in a winning position against centrist opponents at 51% to 54%, supported by expansion into historically resistant retiree and executive demographics.

• I incorporate a significant fade factor because the ‘republican front’ consolidation typically triggers a late decline for the far-right, aligning with Polymarket’s lower implied win probability.

• Structural dissatisfaction remains a primary driver, as 84% of voters distrust political parties and economic anxiety persists among middle-class voters.

• I highlight critical risks including a pending Cour de Cassation ruling on her eligibility and potential campaign financing failures.

2. Marine Le Pen’s vote share in the second round of the presidential election conditional on her opponent ↗

Marine Le Pen’s score shifts depending on her opponent in the second round of the presidential election. Below are Philippe, Mélenchon and Attal. Explore on Mantic ↗

Le Pen v Mélenchon: Le Pen 55%

• The most critical factor in my thinking was the expected breakdown of the traditional “republican front”—the historical tendency for mainstream voters to unite against the far-right. In a contest against the far-left, my research suggests this defensive alliance would likely collapse.

• I found that many centrist and right-wing voters, who previously voted to block Le Pen, now view her opponent with equal or greater skepticism. In this specific pairing, these voters are much more likely to abstain or switch their support to Le Pen rather than back a far-left candidate.

Le Pen v Philippe: Le Pen 48%

• I relied heavily on polling from mid-2026, which shows Le Pen in a significantly stronger position than in previous election cycles, often leading her opponent in hypothetical matchups. This reflects a “rising floor” of support fueled by widespread political distrust, economic dissatisfaction, and the normalization of her party.

• Because Philippe is viewed as a relatively popular, stable centrist, I anticipated he would be more effective at attracting tactical votes from the left and center-right than other potential candidates.

Le Pen v Attal: Le Pen 50%

• I considered the “normalization” of the National Rally and a clear upward trajectory in Le Pen’s previous second-round performances. Current head-to-head polling shows her in a significantly stronger position than in previous cycles, fueled by high levels of political distrust and economic dissatisfaction.

• I adjusted for the historical “polling fade” observed in past French elections. I accounted for the likely reactivation of the “Republican Front”—the traditional mobilization of voters from the left and center-right to block the far-right in the final round.

Subscribe now

3. When the next US government shutdown begins ↗

Federal funding expires at the end of September, but with the midterms only weeks away neither party wants to own a shutdown. Is that enough to avoid it? Explore on Mantic ↗

• My median prediction for the next US government shutdown to begin is December 1, 2026.

• While federal funding is scheduled to expire at midnight on September 30, 2026, I anticipate a short-term bridge will be passed to avoid political backlash before the November midterm elections. However, the House-passed continuing resolution only extends through December 4, meaning the highest risk of a lapse occurs during the post-election lame duck session when partisan leverage increases.

• I identify the primary legislative deadlock as the demand for the SAVE America Act and disputes over defense supplemental funding. With only 11 overlapping session days remaining in September, there is insufficient time to resolve these fundamental policy differences permanently.

• This forecast is supported by the 119th Congress’s historical propensity for funding lapses, including a 43-day shutdown and a 76-day partial shutdown earlier in this session. The lack of 50 votes for a Senate GOP reconciliation fallback further limits the paths to averting a December closure.

4. The EU aggregate gas storage fill level on 1 November 2026 ↗

Europe went into the summer with tanks at a record-low 28% and the refill has been slow ever since, with the Hormuz disruption halving Middle Eastern LNG. Brussels has already cut its target to 80%, so where do the tanks actually land on 1 November? Explore on Mantic ↗

• I predict the most likely EU aggregate gas storage fill level on 1 November 2026 is 74.82%, with a 50% chance of falling between 71.62% and 78.12%.

• This projection is based on the current fill level of 55%, which is significantly behind the pace needed for traditional targets due to a record-low 28% starting level in April 2026. Supply disruptions from the Strait of Hormuz crisis have halved Middle Eastern LNG imports, and while Qatar began resuming shipments on 31 July 2026, infrastructure damage limits recovery speed.

• I find that deep market backwardation and high Asian demand for US LNG have removed commercial incentives for storage, leaving replenishment dependent on state mandates. Consequently, the European Commission has lowered its target to 80%, aligning with analyst forecasts from HSBC and Wood Mackenzie that cluster in the low-to-mid 70s.

• My analysis assumes that while Norwegian pipeline flows remain reliable, the required injection rate to hit 90% is unattainable under current scarcity premiums, where daily injection needs exceed historical peaks.

5. The composition of the Swedish government in office on 31 December 2026 ↗

Sweden's September election looks likely to hand the Social Democrats the largest bloc, but forming a government is a separate problem. Explore on Mantic ↗

• I predict a Social Democrat-led government will be in office by 31 December 2026 with a combined probability of 72%.

• Polling from June 2026 indicates a decisive lead for the center-left opposition at 55.2%, compared to 42.6% for the incumbent bloc, with the Social Democrats serving as the dominant power center at 33.9%.

• The governing coalition path to a majority is severely compromised as the Liberal Party polls at 1% to 2.5%, well below the 4% threshold, while 9.9% unemployment erodes voter support for Prime Minister Kristersson.

• Friction remains a 15% risk factor if Sweden Democrat demands for cabinet seats or Left Party ministerial aspirations extend negotiations past the year-end deadline.

1. When Anthropic IPOs

Last month, Anthropic confidentially submitted draft S-1 to the SEC, marking a major step towards an IPO.

• My median prediction for Anthropic becoming publicly traded is 16th November 2026.

• Following a confidential S-1 filing on 1st June 2026, the company is in an active execution phase, with lead underwriters Morgan Stanley and Goldman Sachs already conducting investor meetings.

• While the company is internally targeting an October 2026 debut, I have adjusted for likely slippage due to the administrative complexity of a trillion-dollar valuation and standard SEC review cycles.

• Financial readiness is high, supported by a revenue run-rate exceeding $47 billion and the recruitment of specialized investor relations leadership to manage the transition.

• The macro environment remains exceptionally favorable, characterized by record-high IPO underwriting volume in mid-2026 and strong investor demand for AI infrastructure.

2. The winner of the Brazilian presidential election

Flavio Bolsonaro’s presidential campaign against the incumbent Luiz Inacio Lula da Silva will officially be launched on the 25th by the Liberal Party.

• I predict Luiz Inácio Lula da Silva will win the 2026 Brazilian presidential election with 60% probability.

• Current polling confirms Lula’s consistent lead, with first-round support reaching up to 41.8% and runoff simulations showing him ahead of his main rival, Flávio Bolsonaro, by margins as wide as 12.5 percentage points.

• Economic tailwinds, including a multi-year low unemployment rate of 5.6% and peaked economic activity in early 2026, provide the incumbent a significant advantage despite a 46% disapproval rating and new U.S. trade tariffs.

• The opposition remains fragmented and weakened; Flávio Bolsonaro’s support among moderate right-wing voters recently plummeted from 74% to 54%, while alternative candidates like Ronaldo Caiado and Renan Santos continue to poll in the single digits or low teens.

  Subscribe now

3. The total number of measles cases in the US in 2026

US 2026 measles cases have already surpassed the 2025 total of 2289, before the school year even begins, as vaccination rates keep sliding. How high will they get?

• My most likely prediction for confirmed measles cases in the United States for 2026 is 3,170, with a 50% chance of falling between 2,881 and 3,718.

• As of late July, the U.S. has already recorded 2,295 cases, surpassing the full-year 2025 total, with current transmission averaging 35 to 40 new cases per week.

• The surge is driven by 34 outbreaks facilitated by a decline in national kindergarten MMR vaccination rates to 92.5%, which sits dangerously below the 95% community immunity threshold.

• While major outbreaks in South Carolina and Utah have slowed, I expect a moderate case increase in the fall due to school reopenings, summer travel, and reduced staffing for contact tracing.

• Sustained transmission for over 12 months has placed the U.S. at imminent risk of losing its measles elimination status, a milestone held since 2000.

4. Saudi Arabia is authorised to enrich uranium

Reports of a US–Saudi agreement to allow nuclear enrichment have recently come out but have been conditioned by President Trump on Saudi Arabia joining the Abraham Accords.

• I predict a 24% probability that formal uranium enrichment authorisation will be granted before 2029, with the most likely decision point occurring in late 2028.

• The 123 Agreement signed on July 22, 2026 establishes a legal pathway for cooperation but conditions actual authorisation on a two-year joint feasibility study to assess the commercial viability of domestic enrichment.

• Significant political friction persists as the deal omits the Gold Standard renunciation of enrichment, and Saudi Arabia has not yet adopted the IAEA Additional Protocol for enhanced inspections.

• While Saudi policy mandates the indigenisation of the fuel cycle, experts suggest actual enrichment activity is likely a decade away, reducing the immediate pressure for authorisation before 2029.

5. The number of FDIC Bank failures in H2 2026

After calm 2024 and 2025, bank failures are ticking up again in 2026 — but is this the start of a wave, or just noise?

• My most likely prediction for the number of FDIC-insured failed banks closed in the second half of 2026 is 3.00, with a 50% chance of falling between 2.00 and 3.96.

• Two failures have already been recorded in July 2026: Kentland Federal Savings and Loan Association and Small Business Bank, establishing a firm floor for the period’s total.

• The pipeline for future failures appears to be shrinking as the FDIC “Problem Bank List” declined from 60 to 54 institutions in early 2026, while industry net income rose 14.3% year-over-year.

• Macroeconomic headwinds, including a 12% increase in bankruptcy filings and commercial real estate stress, may trigger one additional small-scale closure but are unlikely to cause systemic contagion given the record-high 1.43% DIF reserve ratio.

This week’s focus is conditional forecasting: how explicit scenario modelling can help explain Mantic’s view of the future.

1. The price of oil in December 2026

Brent jumped almost 10 percent to $84 after the strikes resumed, its highest settle since mid-June. With the crisis unresolved and the Strait of Hormuz only shortly opened, where does crude settle by year end?

• I predict the average price of Brent crude in December 2026 will be $78.25, with a 50% chance of falling between $72.00 and $84.50.

• Downside pressure is expected to intensify late in the year as markets anticipate a massive 4.6 million b/d supply surplus in 2027, alongside declining oil demand in China due to rapid electric vehicle adoption.

• Significant upside risk persists from a potential return to hostilities in the Middle East, where previous disruptions in 2026 drove prices toward $120; current thin OECD inventories provide little protection against such shocks.

• Record non-OPEC+ production from the U.S., Brazil, and Canada acts as a structural headwind, likely capping sustained rallies unless significant geopolitical transit blockades re-emerge.

2. The Strait of Hormuz is open for at least 30 days before 2027

The Strait of Hormuz will be considered to have recovered if the 7-day moving average is at least 50 ships for 30 days straight, as in our previous post on the Iran conflict.

• My median prediction for Strait of Hormuz transits recovering to a 7-day moving average of 50 for 30 consecutive days is 7th May 2027.

• Current transit volumes have collapsed to 14-24 vessels per day following the July 14 reinstatement of a U.S. naval blockade and the breakdown of the June ceasefire.

• The 50-transit threshold represents only one-third of the pre-war baseline of 138 daily transits, yet this level was not achieved even during the peak of the recent Islamabad Memorandum.

• Major commercial carriers require approximately 30-45 incident-free days before resuming operations, meaning a sustained recovery cannot begin until well after a durable security agreement is reached.

• Prohibitive war-risk insurance premiums at 5% of vessel value act as a primary economic barrier that I expect will persist until a decisive military outcome or political settlement occurs in early 2027.

Subscribe now

3. The price of oil in December 2026 conditional on the status of the Strait of Hormuz

As above, the Strait of Hormuz recovery is defined as the 7-day moving average being at least 50 ships for 30 straight days. Specifically for the conditional forecasts, the scenario is that this occurs before 2027, which Forecast 2 above showed being 28%.”

Strait of Hormuz recovery: YES

• I primarily relied on the “futures curve”—the prices at which traders are currently agreeing to buy oil for delivery in late 2026. These market prices served as my strongest anchor, as they already reflect the collective wisdom of global participants. However, I adjusted these figures downward because the current market still prices in some possibility of continued disruption. By assuming the Strait is open, I accounted for a shift toward a market driven by supply and demand fundamentals rather than conflict.

• On the fundamental side, I weighed reports from major energy agencies like the IEA and EIA, which suggest a transition toward a “looser” market by 2026. Factors such as increasing production from non-OPEC countries and a projected global oversupply heading into 2027 were key motivating factors. While I considered the possibility of sudden demand spikes or production cuts elsewhere, my thinking leaned toward a scenario where prices stabilize as the “fear factor” of shipping disruptions dissipates.

Strait of Hormuz recovery: NO

• I first established a “business-as-usual” baseline using current futures markets and analyst consensus, which generally project prices in the $70–$85 range for late 2026. However, I determined that these estimates largely assume a return to geopolitical normalcy. By conditioning my forecast on the failure of transit levels to recover, I shifted my focus toward historical price spikes and risk scenarios modelled by international organisations like the World Bank.

• My thinking balanced two competing forces. On the one hand, a persistent blockage or impairment of a major transit chokepoint suggests a significant “war risk premium” and potential supply shortages. On the other hand, I accounted for the global economy’s ability to adapt over several months. Factors such as the rerouting of oil through pipelines, increased production from non-OPEC countries, and potential demand destruction due to high prices would likely prevent an uncapped rally. Consequently, I favoured a distribution that sits notably higher than current market prices but remains grounded by long-term economic headwinds.

4. The democrats win control of the Senate

While far more uncertain than the House, a once unlikely control of the Senate by the Democrats has become a real possibility in the last year.

• I predict the Republican Party will control the U.S. Senate after the 2026 elections with a 55% probability.

• Structural advantages provide a critical buffer for the GOP, as the current 53-47 split and Vice President J.D. Vance’s tie-breaking vote allow Republicans to lose up to three net seats while maintaining the majority.

• Democrats face a high bar for control, requiring a net gain of four seats to reach a 51-seat majority, a feat achieved by the opposition party in only 25% of cycles since 1980.

• Significant headwinds including 4.2% inflation, low consumer sentiment, and underwater presidential approval ratings favour Democratic gains, particularly as Republicans must defend 22 of the 35 seats at stake.

5. The price of oil in December 2026 conditional on the party in control of the Senate

Not all conditions move a price by acting on it directly. While the new Senate would only seat in early 2027, the results of the election will be known in November. Would the market react if Democrats took control of the Senate?

Democrats control of the Senate: YES

• The specific condition of a Democratic Senate victory was factored in as a modestly “bullish” (price-supporting) element. I reasoned that a Democratic-controlled Senate would likely maintain tighter environmental regulations and be less aggressive in expanding domestic drilling permits compared to the alternative, slightly constraining future supply. However, because Brent is a globally traded commodity and the new Senate wouldn’t take office until after the forecast month, I treated this as a minor upward nudge rather than a transformative driver. Finally, I accounted for oil’s inherent volatility by maintaining wide “tails” in my distribution to cover the possibility of unexpected economic shocks or major supply disruptions.

Democrats control of the Senate: NO

• Regarding the specific condition that the Democratic Party does not win the Senate, I viewed this as a “status quo” or slightly bearish factor. I reasoned that this outcome might favour domestic U.S. energy production and reduce the likelihood of new regulatory constraints, though I concluded the immediate impact on global oil prices within the same month as the election would be relatively modest. Finally, I maintained wide bounds to account for the high historical volatility of the oil market.

  Thanks for reading! Subscribe to keep up to date with Mantic

Scott discusses:

• The history of the field

• Two engines that enabled human-level forecasting performance

• Why these engines may not enable superhuman forecasting performance

• Two ideas to remedy this: 1) a new kind of forecasting tournament, and 2) training models for forecasting with reinforcement learning

1. The number of US job cuts attributed to AI in 2027

Fears about AI job displacement have been increasing in the last few years but seem to finally have some solid footing with more layoffs attributed to AI in 26H1 than in the full year 2025. Will next year maintain the acceleration?

• My most likely prediction for US job cuts attributed to AI in 2027 is 247,000, with a 50% chance of falling between 166,000 and 328,000.

• This forecast is driven by an unprecedented acceleration in 2026, where AI-attributed layoffs reached 101,743 in the first half of the year alone, representing a 23% share of all US job cuts. The shift from experimental to structural AI-driven restructuring is evidenced by major firms like Oracle and Meta redirecting capital from human labor to AI infrastructure.

2. The winner of the 2027 French presidential election

On July 7, the Paris Court of Appeal overturned Marine Le Pen’s ineligibility.

• I predict Marine Le Pen has a 33% probability of winning the 2027 French presidential election.

• Her candidacy is bolstered by a July 2026 court ruling that effectively cleared her legal path to run, followed by immediate polling dominance at 34-36% in the first round.

• Economic tailwinds for the far-right include a high 5.1% deficit and low consumer confidence (84 points), which together foster an anti-incumbent environment that penalizes centrist “heirs” like Edouard Philippe.

• While the “Republican Front” tactical voting strategy has historically blocked the far-right, polling now shows its efficacy at a historical low, with Le Pen leading Philippe 51-54% in potential runoff matchups.

Subscribe now

3. Iranian crude production in August

After the Memorandum of Understanding collapsed and strikes resumed this week, the market’s real question isn’t the politics but the availability of Iranian barrels. Will the Iranian crude actually reach buyers, or vanish again?

• I predict Iran’s crude oil production in August 2026 will be 2,912.50 tb/d, with a 50% chance of falling between 2,712.50 and 3,112.50 tb/d.

• The recovery is driven by the late June lifting of the U.S. naval blockade and a 60-day sanctions waiver that has cleared massive domestic storage bottlenecks, allowing for the aggressive restart of shut-in wells.

• Production is unlikely to reach the pre-war baseline of 3,241 tb/d due to estimated infrastructure damage of up to $19 billion and the technical ‘slow grind’ associated with restoring complex maritime platforms.

4. When Israeli forces withdraw from Lebanon

President Trump has said Israeli forces will leave Lebanon but signals from Jerusalem are mixed, making it the key unknown for whether the regional escalation is winding down or escalating again.

• Israeli leadership has institutionalized an indefinite presence within a 10-kilometer security buffer, stating that troops will remain until Hezbollah is fully disarmed and displaced residents return to northern Israel.

• The 2026 U.S.-brokered framework agreement is effectively frozen due to Hezbollah’s refusal to disarm and the Lebanese Armed Forces’ inability to provide security guarantees following a 98 percent depreciation in real salary value.

• Historical precedent suggests long-term occupation; the previous Israeli security zone in southern Lebanon lasted 15 years, and current entrenchment—including village demolitions—points toward a multi-year timeline.

5. When the SEC adopts a final rule on semiannual reporting

A political push to reduce quarterly reporting to semi-annual has run into strong pushback, notably from retail investors. Will the SEC listen to their demands or move with the change?

• I predict a 4% probability that the SEC will adopt a final rule for semi-annual reporting by the October 9, 2026 deadline.

• The SEC must process 68,000 public comments, with 99% in opposition, requiring a legally defensible record under the Administrative Procedure Act that typically takes 6-12 months to finalize.

• Intense pushback from major institutional investors like BlackRock and Vanguard, alongside the SEC’s own Investor Advisory Committee, necessitates a robust and time-consuming economic analysis to survive certain litigation.

• While the 3-0 Republican majority and the White House strongly support the shift to reduce $330,000 in annual compliance costs per firm, the 95-day window from the close of comments is mechanically insufficient for a final vote.

• Scott is giving an invited talk at the Workshop on AI Forecasting on Saturday 11th July from 13:30 to 14:15 in the Grand Ballroom (104–105)

• Matthew and Scott will present three papers (details below) in the poster session on Saturday 11th July from 12:15 to 13:30 in Hall A.

• We’re hosting a dinner on Thursday 9th July

Reach out to Scott or Matthew directly to meet up and learn more about Mantic.

Invited talk: Towards Superhuman Forecasting

Scott’s invited talk is about the path to superhuman forecasting, covering how AI systems have achieved human-level forecasting, asking whether the methods we’ve used to get this far will keep working, and outlining some paths we see for making further progress.

The talk is provisionally scheduled to run after lunch on Saturday 11th from 13:30 to 14:15 in the Grand Ballroom (104–105). Check the workshop site for final scheduling details on the day.

Spotlight: Reaching the frontier of AI forecasting with reinforcement learning

Scott and Matthew will present their work on reaching the frontier of AI forecasting with reinforcement learning, accepted as spotlight at the workshop. The paper explains how we raised gpt-oss-120b to frontier model performance (as we previously introduced on the Thinking Machines website) .

Poster: Diversity is the strength of the AI crowd

Matthew and Scott will present our investigation into how the wisdom of the crowds applies to a crowd of AIs. We find that certain weaker models — in this case, Grok 4 — can earn their place in the team by contributing a diversity of thinking, which makes them less substitutable than the frontier models.

Poster: An adversarial tournament design for efficiently probing the frontier of AI forecasting

Scott and Matthew will be presenting our work in developing a new kind of forecasting tournament designed to more efficiently probe the frontier. Unlike existing tournaments, we have participants both making forecasts and posing forecasting questions, with the question-writers scored for how much disagreement they are able to induce among the forecasters. Our first tournament using this design, Series 1, is live now and we have previously written about the idea on this blog.

Meet the team

Come find Scott and Matthew this Saturday 11th July at:

• The forecasting workshop (Grand Ballroom).

• The poster session: 12:15 to 13:30 in Hall A.

Or reach out directly over LinkedIn (Scott, Matthew) or the ICML app.