Research integrity
Every conclusion should lead back to its evidence.
Tradour keeps the source, each change and every research decision visible. A later result cannot quietly rewrite what happened before.
Directional feature
Rise and Drop are source labels—not automatic advice.
Historical BSP display direction was produced by inverting the original raw model direction across the dataset. Tradour therefore models direction neutrally and stores the transformation and tested interpretation explicitly wherever source information permits.
No direction is silently transformed, and no label is assumed to be inherently bullish or bearish.
Make it clickWhy is a direction label not a trade instruction?
A Tradour directional feature records how a source labelled an expected move. It is research data, not an instruction to trade.
Read the full guideData architecture
Boundaries that protect the record
Each layer has a distinct purpose and promotion path. Presentation consumes safe, validated outputs rather than reaching into source or exploratory stores.
- 01
Immutable source
Preserved byte-for-byte with identity, checksum, capture time and access controls.
- 02
Normalised records
Reproducibly parsed under explicit, versioned and tested ingestion rules.
- 03
Derived analysis
Features and outcomes calculated by deterministic code with declared methodology.
- 04
Research evidence
Exploratory and candidate outputs remain distinct from reviewed conclusions.
- 05
Validated Playbook
Versioned findings linked to evidence, limitations and review history.
- 06
Prospective record
Timestamped observations preserved before outcomes and linked later without invisible rewriting.
Outcome definitions
The canonical Day 1 rule is fixed. Its inputs are not here yet.
Tradour measures from the last legitimate close before earnings information became public to the close of the first trading session in which it was public. BMO, AMC and non-trading-day rules are explicit; ambiguous or intraday publication is not guessed, and a zero move is Neutral. The supplied files lack the auditable market prices, calendars and publication timestamps needed to calculate that outcome.
Evidence lifecycle
Interesting does not mean validated.
Candidate rules must be frozen before later data is inspected. Multiple-testing provenance and data exposure are recorded, and an exposed replay cannot be promoted as out-of-sample evidence. There is no invented universal sample threshold or p-value shortcut.
- 01Exploratory
- 02Candidate
- 03Under review
- 04Validated
- 05Playbook
Also recorded: Rejected · Retired · Superseded
