By Cree Lawson. Research reviewed November 6, 2026.
A losing week creates two different questions: did the market become less friendly to the pattern your algorithm trades, or did something change in the algorithm itself? The Algo Consistency Index (ACI), published by Prismatic Algos, provides a structured way to investigate that distinction. It describes recent conditions and account results. It has not demonstrated that it can forecast your next profitable trading period.
1. Start with the mechanism, not the product name
An Expert Advisor (EA) is software that implements trading rules. ACI groups its research around three mechanisms: trend/breakout, mean reversion and session-clock. A breakout entry seeks continuation beyond a price boundary; a mean-reversion entry seeks movement back toward a reference; a session-clock entry depends on time-of-day structure. A robot can combine mechanisms, so a marketing label alone is not enough to choose a comparison.
Write down the actual entry, exit, holding period, instruments and exposure rules of your EA. If these differ materially from the reference recipes, record that mismatch. In particular, the published mean-reversion reading carries a caveat: it does not represent the project's library mean-reversion family.
2. Check the dates before interpreting the reading
The publication date and the date covered by a reading are different. Check the dates shown on the cards and, when using the downloadable data, the readings_through and scoreboard_through fields. A newly published page can contain older observations. Different layers can also stop on different dates.
The November 5 methodology update illustrates this: refreshed tick data advanced the probe readings through October 30, while the wider 25-symbol market layer remained at October 1 because two symbol files were missing. Those October observations were published late; they were not information available in real time on each historical day.
3. Read the layers separately
The probe index replays twelve frozen textbook recipes across seventeen symbols, with stated trading costs and a matched random-entry comparison. These results are hypothetical. The Scoreboard describes recent account-panel results and is lagging. Pattern Integrity describes market-state features and drift. The external panel contains hypothetical model-account records, not verified funded-account returns.
Read a low percentile as a position within that measure's own history. It is not the probability that tomorrow will lose. A drift alert identifies a change in measured conditions; it does not prove that an EA has stopped working.
4. Turn disagreement into an investigation
Illustrative example, not an observed ACI result: suppose a trend EA loses money while its reference reading improves. First align the dates and instruments. Then inspect missed orders, spread and slippage, holding periods, changes in position size, and whether the EA actually follows the same mechanism. An improving reference does not prove that the EA is broken; it gives you a specific comparison to examine.
If both weaken, record that coincidence. It still does not establish that the environment caused the EA's losses, or that either series will recover next week. Check enough observations to avoid building a story around one trade.
5. Keep a dated review sheet
- Observation time and methodology version.
- Each card's as-of date, sample count and freshness warnings.
- EA mechanism, instruments and known benchmark mismatches.
- Account outcome over the same window, including costs.
- Operational checks performed and unresolved explanations.
- Next review date and the evidence that would change your interpretation.
Archive the original reading when you record it. Recent external-panel rows can change as strategies report late. Looking back only at today's revised history can give a misleading impression of what you knew at the time.
What this workflow does not authorize
ACI's published tests did not establish forecasting, timing-overlay or position-sizing value. The proposed sizing multiplier was dropped on October 28. Do not turn a percentile into a new leverage, halt or restart rule without independently testing that rule out of sample. The useful output of this review is a better-documented question about performance, not an automatic trading instruction.
Research attribution: the ACI methodology and dated validation record, version 0.2, reviewed November 6, 2026. The author is the founder of Prismatic Algos, which publishes ACI. This educational case study reports the project’s own findings, not independent validation. Trading involves risk of loss.
