A strategy record has a source link, entry rules and a code flag. Is it ready to backtest? You still need to check whether those fields describe reproducible decisions. Counting filled fields cannot answer that question.
In this guide, we will use the Compendium's field-coverage audit to build a four-step review. The output is a record of what you know, what remains ambiguous and what needs to be checked against the source.
What the field count measures
We examined all 3,582 records in the export generated September 21, 2026 at 15:09 UTC. The check asked whether selected fields contained text beyond defined placeholders. It did not test code, reproduce trades or measure returns.
Entry-rule text appears in 2,909 records, exit-rule text in 2,208, position-sizing text in 2,097 and stop-loss text in 1,981. Every record has a source URL. A code-capture flag appears in 1,292 records.
Text presence after placeholder filtering, or a metadata flag. Presence does not establish accuracy, completeness or performance.
Field-coverage denominator: all 3,582 records. Recorded presence does not establish accuracy, completeness or trading performance.
1. Trace each important field to its source
Save the source URL, capture date and version you are inspecting. Compare the structured entry and exit fields with the relevant source passage. A recorded URL is a reference to check; it does not establish that the page still exists or remains unchanged.
In our targeted review of eight captured source documents, an entry trigger appeared in the source but was missing from the structured entry field. That is evidence of an extraction issue in that example, not a measured error rate for the whole collection.
2. Separate missing text from ambiguous text
A missing exit field could mean the original description omitted the exit, or that extraction missed it. Until you inspect the source, leave the reason unresolved. A populated field also needs interpretation.
Illustrative wording: “Use appropriate position sizing” fills a text field but does not specify a volume calculation. “Exit when the trend changes” leaves the reference, timing and decision rule open. Neither phrase is a reproducible instruction on its own.
Ask whether another researcher could implement the same decision without inventing a missing condition. Check units, price references and the information available at the moment the rule acts.
3. Record the review status explicitly
The following worksheet is a proposed research aid, not a claim that the Compendium already stores these reviewed statuses. Keep separate rows for entry, exit, sizing, stops and implementation dependencies.
| Status | Meaning | Next action |
|---|---|---|
| Not yet reviewed | No source comparison completed | Inspect the relevant passage |
| Not extracted | A source rule exists but the structured field missed it | Record the omission and its source |
| Not described in reviewed source | The inspected material did not specify the rule | Seek further documentation |
| Ambiguous | Text exists but leaves an implementation decision open | List the missing condition or definition |
| Specified for implementation | The reviewed text supports a definite rule | Verify the implementation separately |
The final status is still not a performance assessment. Correctly transcribing a rule and establishing that it behaves as intended are separate tasks.
4. Write the test brief before running the test
State the instrument, data period, code version, settings, dependencies and cost assumptions. Specify concurrent-position and account-level limits. Make clear which choices came from the source and which are your own research assumptions.
- Can the source and version be traced?
- Are entry and exit conditions explicit, including decision timing?
- Are sizing and overlapping exposure defined?
- Are the implementation and test assumptions recorded?
If the brief still contains an important unknown, resolve it or explicitly design the study around that uncertainty. Do not quietly promote documentation completeness into an algorithm ranking.
Method and what remains untested
This uses the same September 21 export and eight-document source review as Catalog Notes 01, retrieved September 23 from commit 375327a. Counts include versions, utilities and overlays. Defined placeholders were excluded before measuring text presence; the code measure is a flag. Source coverage is concentrated and is not an industry-wide sample.
The purposive source review cannot estimate extraction accuracy across the collection. A future random sample with explicit review rules and independent assessment could address that question. It has not been completed here. No algorithms were compiled, traded or backtested for this analysis.
Continue your research
Read the companion guide to algorithm families and rules, then check how trading results are produced. If you want a starting point for exploring your preferences, try the educational Algo Matchmaker.
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.