You find a strategy description with a clear entry signal. What else do you need before testing it? At minimum, you need to understand when positions close, how exposure is sized and which assumptions connect the description to an implementation.
We examined the Compendium export generated on September 21, 2026 at 15:09 UTC to organize those questions. This guide moves from trading families to rule completeness, then shows how to use the findings in your own research.
Scope: the export contains 3,582 research records, including strategy descriptions, indicators, utilities, overlays and versions. Family percentages use 2,904 classified records; field coverage uses all 3,582. These are catalog observations, not results from running or independently validating the algorithms.
1. Use the family label to choose your first questions
Trend-following descriptions look for continuation; mean-reversion descriptions look for a return toward a reference. Other families organize ideas around volatility, timing, market structure or statistical relationships. Execution tools and portfolio overlays may manage trades rather than supply a complete strategy.
In the classified subset, 1,054 records carry trend/directional labels and 789 carry mean-reversion/counter-trend labels. Together, that is 1,843 records, or 63.5% of the 2,904 classified records. Trade/execution management accounts for 376 records and risk/portfolio overlays for 101.
Percentages of classified records only. Labels as indexed; see the article methodology for scope.
Exhibit 1. Categories as indexed. The denominator is 2,904 classified records; 678 unlabeled records are excluded. Versions, utilities and overlays remain included.
Use the chart as a map of this catalog. During a targeted source review, we found a trend-template description classified under mean reversion and a reversal description under trend/directional. A system outside the grid category may still include grid logic. We retained the indexed labels rather than implying the population had been corrected and verified.
2. Check entry, exit and sizing separately
After removing defined placeholders, entry-rule text appears in 2,909 records (81.2%), exit-rule text in 2,208 (61.6%) and position-sizing text in 2,097 (58.5%). Stop-loss text appears in 1,981 (55.3%). All records have a source URL; 1,292 (36.1%) have a code-capture flag.
Text presence after placeholder filtering, or a metadata flag. Presence does not establish accuracy, completeness or performance.
Exhibit 2. Text presence or a recorded metadata flag across all 3,582 records. A code-capture flag does not establish that code compiled or passed a test.
These counts tell you where to investigate. They do not establish that the rule is precise enough to implement. An empty field can also reflect extraction failure: our source review found an entry trigger in captured source text that had not been preserved in its structured entry field.
3. Turn a description into a testable specification
Illustrative description: “Buy after a breakout and use a stop.” This is not a rule set we tested. It leaves several decisions open. Which boundary defines the breakout? Does the signal use a completed bar? Where is the stop placed? How much exposure is allowed?
| Part of the rule | What to resolve |
|---|---|
| Entry | Exact condition, data inputs and decision timing |
| Exit | Price, time or condition that closes the trade |
| Sizing | Volume calculation and limits on concurrent exposure |
| Implementation | Instrument, version, dependencies, costs and execution assumptions |
Record unresolved details before selecting a backtest setup. Otherwise, your implementation can become a test of assumptions you supplied rather than the source idea. See the field-audit guide for a repeatable way to track those gaps.
4. Check the source and the population
In this export, 2,170 source URLs point to ProRealCode and 824 to MQL5. Those domains account for 83.6% of records, so the collection is not a representative survey of the whole industry. The 3,437 existing version-group keys are also not an independently verified count of unique strategies.
No return estimate or performance ranking follows from either chart. Their useful output is a better set of questions about source material and implementation. For example, MQL5's beginner EA tutorial starts by specifying what the program should do before moving into code. Treat its platform-specific examples as dated instructional material.
Method and limitations
Source: Prismatic's private strategy-compendium export, September 21, 2026 at 15:09 UTC, retrieved September 23 from commit 375327a. A subsequent repository update did not change that export. Family percentages exclude the 678 unlabeled records; field coverage retains them. Defined placeholders were filtered before counting text; labels and code-capture flags were not independently validated.
The source review examined selected passages in eight captured documents across eight category strata. This purposive sample cannot estimate a database-wide error rate. Source websites were not rechecked for freshness in that audit. No algorithms were compiled, traded or backtested for this analysis. A larger current inventory does not change the denominator of this dated research.
Apply the checks
Choose one description, trace its source, and complete the four specification rows above. Keep uncertainties visible. Explore the full family guide or use the educational Algo Matchmaker to identify approaches to investigate. A color profile is not a performance grade or suitability assessment.
Prismatic develops algorithmic portfolios. This educational article does not evaluate or promote a Prismatic product. It is not investment advice or a suitability assessment. Trading can result in substantial loss.