The Signal Ledger: 2026, Read Backwards

At the start of 2026, a litre of Nairobi super petrol had a regulated maximum price of KSh 226.96. From 15 February, the next monthly schedule lowered that ceiling to KSh 178.28. By March, the landed cost of imported petrol had risen sharply. The year did not send one clear message. It sent separate numbers through different systems, each with its own date, unit and delay.

Read backwards from October, those numbers can be arranged into a story of oil, rates, disease, climate and investment. The danger is to make the arrangement look like prediction. A later event can make an early clue feel obvious even when it was not. This account follows the records as they changed and asks what each one could support at the time.

The answer is not that every risk was hidden. Some were measurable; some were only possibilities; some were not comparable at all. The disciplined task is to keep those categories separate.

Price is a signal with several clocks

The February global benchmark for Brent crude averaged about US$70.89 a barrel, according to the IMF commodity-price series distributed through FRED. The U.S. Energy Information Administration (EIA) later reported that Brent front-month futures began 2026 around US$61 a barrel and ended the first quarter at US$118, after military action in the Middle East on 28 February and a de facto closure of the Strait of Hormuz. Those are different measures: a monthly global benchmark average and a front-month contract at quarter-end. Neither is a Kenyan pump price. EIA’s first-quarter account makes the contract and date explicit.

Kenya’s pump price moved on a different schedule. The Energy and Petroleum Regulatory Authority’s January price list set Nairobi super petrol’s maximum at KSh 226.96 a litre from 15 January to 14 February. The next EPRA schedule, effective 15 February, set it at KSh 178.28. In its March release, EPRA reported that the average landed cost of imported super petrol had risen 41.53 per cent, from US$582.11 per cubic metre in February to US$823.87 in March.

These observations trace a sequence, not a one-to-one causal coefficient. EPRA’s retail formula includes landed product cost, transport, storage, financing, margins and taxes; the price-setting period also follows an import-cost window. In April, a tax change lowered the Nairobi petrol pump price even as landed costs had risen. A world crude price can surge before an importer’s cargo is priced, and a pump price can be held down by a government choice. EPRA’s formula explains why “Brent rose” and “fuel at the pump rose” are related statements, not identical ones.

The distinction matters when the next shock arrives. A falling Brent quote can coexist with expensive cargoes already contracted. A price cap can cushion a family’s fare today while moving part of the bill to a levy or public account. The household sees one number; the regulator calculates several.

Inflation and rates did not move in unison

Kenya’s policy rate also had its own clock. The Central Bank of Kenya lowered the Central Bank Rate to 8.75 per cent on 10 February, then the published rate history shows it unchanged in the listed April, June, August and October decisions, through 7 October. Meanwhile, Kenya’s year-on-year inflation rate moved from 4.3 per cent in February to 6.8 per cent in September. In the September CPI release, food inflation was 9.5 per cent and transport inflation 15.6 per cent.

Those indicators measure different things. The CBR is a monetary-policy instrument; the CPI reports changes in the price of a weighted basket. A steady rate is not proof that price pressure has ended. A rising CPI does not prove that a central bank must raise rates at its next meeting. Policymakers weigh expectations, output, credit, exchange rates and the source of inflation as well as its latest reading. Monetary policy can restrain broad demand; it cannot make a harvest, reopen a sea lane or lower the price of a single imported input by decree.

The June Federal Reserve minutes offer an unusually clear example of expectations moving before the physical economy. Participants noted that optimism over a possible Middle East resolution and the announcement of a U.S.–Iran memorandum pushed oil futures and near-term inflation compensation materially lower than in April. The minutes document market expectations; they do not show that the Strait had returned to normal. In the same month, the European Central Bank raised rates by a quarter point, citing elevated energy-related inflation pressure. The ECB’s 11 June decision and the Fed’s market account describe different institutions responding to related external pressures.

Calling 2026 “a year of rate hikes” would flatten those differences. Central banks did not share one timetable or one mandate, and rates were not a referendum on a single oil price. The more useful signal was the transmission: energy costs and financial expectations reached policymakers through different paths and with different delays.

A case count is not the whole outbreak

The health record is another example of why a headline needs its denominator. On 20 May, the World Health Organization reported 53 confirmed Bundibugyo Ebola cases in the Democratic Republic of the Congo and Uganda, alongside about 600 suspected cases and 139 suspected deaths. Its briefing described a serious regional emergency, but confirmed and suspected totals were not interchangeable.

By 20 October, WHO reported 9,704 confirmed cases and 4,687 deaths across 66 health zones in eight DRC provinces. Kenya had reported nine cases, including one death; WHO said the eight cases after the imported index case were known contacts and that there was no evidence of wider community transmission at that cutoff. The 23 October update dates each figure and limits the Kenyan finding to the information then available.

A confirmed count can rise because transmission increases, because more people are tested, or because delayed laboratory results arrive. WHO’s July situation report explicitly warned that new reports could include backlogged samples and not represent infections acquired on the report date. The count is still important: it records diagnosed cases under a defined system. But it is not a real-time map of the day’s transmission, and a crude ratio of reported deaths to confirmed cases is not a patient’s prognosis.

The signal for public-health planning was not only the total. It was the combination of continued spread, shifting geography, diagnostic delays and the capacity of clinics and laboratories to find, test and care for people. A figure becomes more useful when paired with the reporting date, location, case definition and known gaps.

Climate forecasts are not rainfall reports

By October, climate agencies were warning about a strong El Niño. NOAA’s 8 October diagnostic discussion gave a 75 per cent chance that the October–December three-month Relative Oceanic Niño Index (RONI) would reach at least +2.5°C, a threshold that would exceed recorded events back to 1950. That is a probability about a Pacific index, not a 75 per cent chance of floods, crop loss or disease in Kenya. NOAA’s discussion separates the index outlook from local impacts.

Local forecasts add the geography. The Kenya Meteorological Service’s October–December outlook expected above-average rainfall in several named counties and provided onset and cessation periods. It also directed users to follow shorter-range forecasts and updates. That qualification is central. Seasonal probabilities help agencies prepare water systems, roads, clinics and emergency supplies; they do not say which village will flood or when.

The same separation applies to health risks. Heavy rainfall can alter mosquito habitat, water quality, road access and disease surveillance. Those are mechanisms that may increase certain hazards. They are not proof that a disease outbreak will occur in a particular county. The signal ledger’s job is to identify a condition worth monitoring, then keep the possible consequence distinct from the forecast itself.

Capital announcements are not cash flows

The technology market added a different kind of number. Anthropic announced a US$30 billion Series G round in February at a US$380 billion post-money valuation. The company’s announcement reports both amounts, but they answer different questions: the company raised US$30 billion; the post-money valuation is an implied value based on the new round’s share price, not cash raised or a public price for every share.

Nvidia’s second-quarter fiscal 2027 results were another sort of evidence. The company reported US$96.2 billion in total revenue and US$89.0 billion in Data Center revenue, up 117 per cent from a year earlier. Those were realized company-reported quarterly results, not a valuation rumor. Nvidia’s release does not by itself show how concentrated its customers are, how much of their spending was financed, or whether every buyer will earn a return.

Partnerships complicate the picture. A cloud provider may invest in an AI developer that later buys compute from the provider. That arrangement could finance useful infrastructure, or it could make demand appear more independent than it is. The Federal Trade Commission’s staff study of partnerships between leading AI firms and cloud providers examined access to compute, switching costs, exclusivity and information rights. Its report identifies competition questions; it does not find that every partnership is unlawful or that a sector-wide financial bubble has been proven.

What “read backwards” can and cannot do

After the year’s outcomes are known, it is easy to choose the early signals that fit. A better review sorts the evidence into three categories. An observation has a date and a source: a measured CPI, a policy vote, a confirmed case count. A mechanism explains how one condition could affect another: oil can raise transport costs; a higher policy rate can change borrowing conditions. A forecast assigns a probability to a defined event by a deadline. One category cannot be quietly swapped for another.

Suppose, hypothetically, a forecaster assigns a 70 per cent probability to an event. Under the Brier score, a yes outcome gives (0.70 − 1)² = 0.09; a no outcome gives (0.70 − 0)² = 0.49. The score rewards probabilities that are both confident and correct, and penalizes confident misses. But one result cannot establish skill. Forecasts need explicit resolution rules and a record across many comparable questions.

The point of reading 2026 backwards is not to claim it was predictable. It is to identify what the evidence said, when it said it, and what it did not say. Oil price series changed before every Kenyan fuel schedule did. Central banks faced common pressures without making identical decisions. WHO’s case totals recorded diagnosed disease without showing every infection. NOAA forecast a Pacific index, while national meteorologists translated the climate outlook into county-level probabilities. A financing valuation conveyed a transaction price, not a cash balance.

Those distinctions do not make the signals less useful. They make them harder to misuse. A ledger is valuable when it preserves the date, unit, source and uncertainty alongside the number. The year only looks like one clear story when those labels are erased.

John Carlin’s Account Book Ledger, dated 1835–1856 and held by The Met. Historical object used as a visual metaphor for keeping records; it is not a record of 2026 events or forecasts. Credit: John Carlin, Account Book Ledger (1835–1856), Metropolitan Museum of Art, via Wikimedia Commons. Image source · CC0 1.0 Universal Public Domain Dedication