How to evaluate a future hypothesis
A life bar describes a research page’s follow-up; it does not prove that its projected future will occur. This guide explains SAGE’s indicators and offers a reusable reading checklist: examine facts, compare explanations, then check the stated observations. It is intended for curious readers and monitoring professionals.
Examine a research page: solar and farming · French
1. Start with a testable question
Identify the exact claim: who might change what, where and under which conditions? An analysis explains a mechanism; a scenario explores a possibility; a measurable hypothesis specifies an observation that could distinguish it from another explanation. These formats support different kinds of conclusions.
Record the initial wording, date, criteria and intended outcome source. A time horizon organises monitoring; its passage does not turn missing data into a refutation. A qualitative or incomplete research page can remain useful without being assigned a forced binary outcome.
2. Examine what the sources actually contribute
Distinguish publication date, event date and collection date for every reference. A company or official statement remains a statement: its existence can be established without proving its conclusion. Look for primary data, measurement scope and information that challenges the hypothesis.
Several newspapers repeating one wire report do not provide several independent confirmations. Trace the original information and ask whether each reference describes a genuinely new observation. Media repetition helps SAGE prioritise topics; it does not measure their truth.
3. Read documented support and probability separately
On SAGE cards, support is the share of examined references classified as favourable: 100 × favourable / (favourable + opposing + neutral), rounded to an integer. It is displayed after three examinations. This display threshold guarantees neither quality nor independence nor sufficient evidence.
Neutral references remain in the denominator. A change in points describes a change in this share; it is not an increase in the probability of the event. The initial probability is a separate estimate preserved in the register. A numerical revision requires its own dated record and justification.
Read the arguments as well: one decisive primary reference may matter more to the reasoning than several uninformative repetitions, although the bar does not measure that difference. No new point may reflect missing discriminating observations rather than a favourable or unfavourable conclusion.
4. Check outcomes without rewriting the past
Compare the observation with criteria stated before the event. News consistent with a scenario may also fit its alternatives: look for what actually distinguishes them. Preserve contradictions and record an indeterminate outcome when the available evidence cannot settle the question.
For resolved binary events, mean Brier score averages (p − y)², where p is between 0 and 1 and y is 0 or 1. Lower is better. Calibration compares groups of estimates with observed frequencies and requires a documented sample. See the ECMWF guide below.
SAGE’s tracking calculation uses numerical initial estimates and binary outcomes accompanied by a public evidence URL. This is not independent certification. An error on a continuous indicator is a different measure, and an open research page is neither a success nor a failure.
5. Use examples without treating them as successes
The solar and farming page helps distinguish land income, farming income and land use. The AI and harvest page highlights data availability and adoption. The diesel page illustrates why a change in crude oil alone cannot explain pump prices. These are methodological starting points; examine their sources and criteria.
This is an AI-assisted editorial resource. Illustrative arithmetic, language links and display rules have been checked automatically; no human validation or superior forecasting performance is claimed. Generated analysis does not replace reading the original documents. The linked research pages are in French; this guide is a separate, complete English edition.
6. Understand traceability and its limits
An OpenTimestamps proof allows verification that a document digest existed before a given point in time. It helps check a version’s integrity; it does not certify a source’s truth or a scenario’s accuracy. A pending anchor remains pending.
This method provides no personalised recommendation. Before an important decision, compare the research with independent data, alternatives and measurement limits. Earlier studies retain their original wording: this guide changes neither their estimates nor their history.
Worked examples · illustrative data
These figures are invented solely to explain the arithmetic. They are not SAGE results.
| Measure | Calculation | Value | Interpretation |
|---|---|---|---|
| Documented support | 2 / (2 + 1 + 0) × 100 | 67 % | 2 favourable, 1 opposing, 0 neutral; not 67% likelihood. |
| Support after one neutral examination | 2 / (2 + 1 + 1) × 100 | 50 % | −17 points; no changed probability implied. |
| One Brier contribution | (0.70 − 1)² | 0.09 | Illustrative 70% estimate, event occurs; not a calibration assessment. |
Six checks for a research note
- Original claim and date
- Primary source and event date
- Competing explanation
- Discriminating observation
- Revision and reason
- Outcome source or unresolved status
Method references
- ECMWF · Statistical Concepts — Probabilistic Data · version dated 3 July 2022: Brier score and calibration; weather forecasting, used here only for the statistical method.
- OpenTimestamps · How it works · document timestamping; not an endorsement of SAGE.
Guide version: 9 October 2026 · references consulted on that date.