Research Integrity
Evidence before narrative
Experimental interpretation follows the evidence rather than the desired story. When the record and the story disagree, the record wins and the story is revised.
Negative and null results
Negative, contradictory, and null results are not removed because they are inconvenient. An experiment in which nothing of interest happened is reported as such, with the same identifiers, provenance and care as any other.
Observation versus interpretation
Measurements and observations are kept distinguishable from interpretations of them. On this site, and in the laboratory's documents, each claim is labelled Known, Observed, Inferred, Hypothesized or Unknown so that a reader can tell which is which.
Falsifiability
Where practical, a research question identifies in advance the evidence that would weaken or reject the favored explanation. A question that no evidence could count against is not treated as a scientific question.
Reproducibility
Scientific claims should be reproducible or independently testable where practical, even when the complete laboratory implementation is not publicly released. Where a claim cannot be independently tested, that limitation is stated with the claim rather than left for the reader to discover.
Provenance
Important experimental artifacts carry enough provenance to establish what existed, when it existed, and which version of the apparatus produced a result. Released documents commit to their evidence with cryptographic hashes where appropriate.
Researcher intervention
Deliberate interventions in an experiment are documented and kept distinguishable from the behavior of the experimental subject. A change the laboratory made is never reported as something the subject did.
Uncertainty
When the evidence is insufficient, the laboratory says Unknown. Unknown is a legitimate result, not a gap to be papered over with a confident-sounding guess.
No predetermined intelligence claim
Project Second World does not assume that its developmental systems will produce intelligence, consciousness, artificial general intelligence, or any other extraordinary outcome. Whether anything of the kind emerges is the question under investigation, and the laboratory would treat its own eagerness to announce such a result as a reason for additional scrutiny.
Revision
Scientific conclusions are revisable when better evidence appears. Revising a conclusion is not a failure of the program; concealing the revision would be.
Errors are corrected visibly
Material errors are corrected visibly rather than silently rewritten when the change affects interpretation.
A released document is part of the record. When it contains a material error — one that affects how a result should be interpreted — the laboratory does not quietly replace the file. It records the correction on the document, revises the version number and revision date, and leaves the correction log visible to every subsequent reader.
Documents in the library may carry one of three lifecycle labels:
- Corrected
- The document has been revised. The correction log states what changed, when, and in which version.
- Superseded
- A newer document replaces this one. The earlier document is preserved as a historical artifact and links to its replacement; the replacement links back.
- Withdrawn
- The document should no longer be relied on. It remains available, labelled, with the reason recorded, so that the record is not silently altered.
An older report is not deleted merely because a newer one exists. Typographical corrections that do not affect interpretation may be applied without a lifecycle label but are still logged.