How to read a back-test result before a decision
The essentials in 30 seconds
- Read a back-test in five steps: check input dates, replay the computation, open the dimension drivers, note the limits, and compare the score with the threshold.
- Every input must have been public on the decision date, as in the 2007, 2013, and 2017 EQOS Asia cases.
- The EQOS Asia method places the STOP zone below 0.618 (EQOS Asia method page, as of October 2026).
- One case is a sample of one, so several cases give a fairer reading.
- The result informs method choice and carries no investment advice.
A back-test result looks like a verdict, and a careful reader treats it as evidence about a method. This guide gives five steps for reading one before a decision, with examples from the EQOS Asia case studies as of October 2026. It belongs to the lessons in what past failures teach about building a decision model.
On this page
Compare the EQOS Asia offers and request a 30-minute pre-diagnostic.
What is a back-test result, and what does it show?
A back-test applies a model to a past case, using only the data that existed on the date of the decision, and compares the output with what happened afterwards. The result shows whether the signals of the model pointed in the right direction on that case. The EQOS Asia pages call these studies retrospective educational exercises on public data (EQOS Asia case-study pages, as of October 2026).
A reader gets three things from the result: a score, the dimensions that drove it, and the inputs behind each dimension. Reading all three matters. A score alone is a headline, while the drivers show the mechanism, and the inputs let anyone recompute the answer. The five steps below follow that order.
Which question about input dates comes first in a back-test?
Ask first whether every input was public on the decision date. The EQOS Asia pages state this for each case: 2006 and 2007 for the financial crisis, 2013 for Theranos, and 2017 for WeWork (EQOS Asia case-study pages, as of October 2026). A later input quietly turns a back-test into a story told with hindsight.
Check also that the study gives sources and dates for its inputs. A reader can then open each source, confirm the figure, and confirm that it predates the event. Where a source is missing, the sensible reading is to treat the related dimension as unproven and to look for the gap in the study before relying on the score.
Can a third party replay a back-test result?
A deterministic model returns the same score for the same inputs, so a third party can recompute the result. The EQOS Asia method page describes the Codex Universalis core as deterministic, and it states that scores come only from the Python engine and never from a generative model (EQOS Asia method page, as of October 2026).
Replay is the strongest test of a back-test. The case pages publish dimension scores and outputs, while weights and coefficients stay proprietary and are available under NDA. A reader therefore replays at the level that fits the need: reading the scores, questioning the inputs, or requesting a computation review under NDA.
Which dimensions drove the score in a back-test?
Open the dimension table before the headline. In the 2008 and WeWork cases, scores for integrity, transparency, and stability sit low while energy and resonance sit higher: 0.85 and 0.65 for the 2008 system, 0.72 and 0.42 for WeWork (EQOS Asia case-study pages, as of October 2026).
This pattern teaches a reading habit. When the dimensions that look healthy are those most visible to outsiders, the score deserves a closer look at the hidden ones. The driver table tells you where to ask questions and which documents to request before a decision, and it turns a single number into a short list of tasks.
A worked example applies these questions to the 2008 case. The back-test case page lines up a score of 0.234 with the dated events that followed, reads the dimension drivers and the leverage threshold, and states the limits of a single case, so a reader can see each step of the method on real figures.
Which limits should a reader keep in mind with a back-test?
A back-test on one case is a sample of one. It shows that the method read that case well, and it leaves open how the method reads the next one. Selection also matters: known failures are easier to study than cases that went well. A fair reading asks for several cases and mixed outcomes. Taken together, they show whether the method holds up across sectors and periods.
The EQOS Asia pages say this plainly: they describe themselves as retrospective educational exercises and state that they do not constitute investment advice (EQOS Asia case-study pages, as of October 2026). Treat the result as evidence about the method, then judge the method on the quality of its inputs, its replay, and its range of cases.
How do you read a threshold such as the 0.618 STOP line?
A threshold turns a score into an action. The EQOS Asia method places the STOP zone below 0.618 and the OK zone above it (EQOS Asia method page, as of October 2026). The back-test cases returned 0.234, 0.085, and 0.189, so each fell in the STOP zone by a wide margin.
Read the distance as well as the side. A score just under the line asks for a few targeted corrections, while a score far below it signals a structural problem. The case pages pair each STOP with named corrections and a recomputed score, which gives the reader a path forward rather than a verdict alone.
How do you use a back-test result before a decision?
Turn the result into three actions. Run the same dimensions on your own case, list the corrections that would raise the weakest dimension, and set a date to recompute. The free simulator gives a first estimate in about 10 minutes, and the method page explains how the Audit pillar goes deeper (EQOS Asia simulator page, as of October 2026).
A 30-minute pre-diagnostic by video, scheduled in Singapore time, applies the equation to a real case. Back-tests build confidence in the method, and the pre-diagnostic shows what the method says about your decision. Together they give a decision maker evidence about the tool and evidence about the case.
See also in this guide
Frequently asked questions
What is the first question to ask about a back-test?
Ask whether every input was public on the decision date. The EQOS Asia case pages state the data period for each study, and they list source families such as filings and market series, so a reader can confirm dates before trusting the result (EQOS Asia case-study pages, as of October 2026).
Does a good back-test prove a model will work next time?
A good back-test shows that the method read one past case well. Confidence grows with several cases, mixed outcomes, and a replayable computation. The EQOS Asia pages describe their studies as retrospective exercises with no promise about future results (EQOS Asia case-study pages, as of October 2026).
Why do dimension scores matter more than the headline score?
The headline gives a direction, and the dimension scores give the mechanism. They show which inputs drove the result, where to request documents, and which corrections would raise the total. In the EQOS Asia cases, each STOP comes with named corrections tied to specific dimensions.
How can I apply a back-test reading to my own case?
Run the nine dimensions on your own decision in the free simulator, note the weakest ones, and list the corrections that would raise them. A 30-minute video call by Singapore time applies the EQOS Asia equation to a real case (EQOS Asia simulator and contact pages, as of October 2026).
Test your knowledge
1. What should a reader check first in a back-test?
- Whether inputs were public on the decision date
- The font of the report
- The length of the study
Show the answer
Whether inputs were public on the decision date.
2. What does a deterministic model return for the same inputs?
- A random ranking
- A range of scores
- The same score
Show the answer
The same score.
3. Below which score does the EQOS Asia method place the STOP zone?
- 0.382
- 0.618
- 0.9
Show the answer
0.618.
4. What does one back-test case represent?
- A sample of one
- A full guarantee
- A market forecast
Show the answer
A sample of one.
5. How should a reader treat a score far below the threshold?
- Ignore the corrections
- As a structural problem
- Wait for hindsight
Show the answer
As a structural problem.
About the author
Stéphane Jambu
President, EQOS Forecast Asia
Stéphane Jambu leads EQOS Forecast Asia and works as a consultant in e-commerce, search visibility, and AI. LinkedIn profile
Sources: EQOS Asia 2008, Theranos, and WeWork back-test pages (eqos-asia.com/case-studies/crise-2008.html, theranos.html, wework.html), method page (eqos-asia.com/method.html), simulator and contact pages, consulted on 21 October 2026. Page updated on 21 October 2026.