Scientific Discovery
The Psionics Hypothesis Ledger: Keep the Impression, Test the Claim
2026-09-10
A useful psionics hypothesis ledger keeps three records separate: what was experienced, what scientific claim was inferred, and what external evidence later showed. Each candidate receives an identifier, a measurable prediction, an ordinary alternative and an outcome status. The ledger proposed here treats Andrei Ursachi's applied psionics as an exploratory idea-generation framework, not as proof of access to hidden facts.
A research asset, not a collection of memorable moments
The value of an unusual impression is not that it sounds impressive in retrospect. Its value is that it may suggest a route worth checking. Without a record of the transition, a precise result can become attached to an originally vague impression, and the original uncertainty disappears.
Give the raw note its own immutable field. Write the candidate explanation in another field, and attach data results only after the computational test. Corrections become dated additions rather than silent replacements. A buyer should be able to follow how a question became a recommendation.
Preregistration literature distinguishes predictions specified before outcomes from explanations developed afterwards. The proposed ledger borrows that separation. It does not turn a private notebook into an independently validated method, and it does not establish that the initial impression had a special source.
Sources: Nosek et al. (2018), The preregistration revolution.
A worked entry about an overlooked delay
Consider a hypothetical review of an existing public dataset on building electricity use. An intuitive note says that the response feels delayed rather than weak. That phrase is not yet an energy-science finding. It could mean delayed occupancy effects, delayed weather effects or delayed logging.
The first candidate might predict that a preselected one-hour outdoor-temperature lag reduces held-out prediction error relative to a contemporaneous-temperature baseline. The ordinary rival is a clock-alignment error. Both possibilities should be visible before anyone sees the comparison results.
Do not insert a winning error reduction into the entry after exploring twenty lags. The one-hour proposal can fail, while a later exploratory sweep suggests something useful. Those are two research events with different evidential status.
| Record layer | Illustrative entry |
|---|---|
| Raw impression | The response seems delayed |
| Candidate H-017 | A fixed one-hour temperature lag improves forecast error |
| Ordinary alternative | Meter and weather timestamps are misaligned |
| Outcome | Pending until the frozen comparison is run |
Give the candidate a real promotion rule
A ledger needs states that change for a reason. Draft means the interpretation is not fixed. Testable means the variables, comparison and rejection rule are written. Supported on this dataset means that the specified comparison passed here. It does not mean universally true.
Use separate fields for feasibility and performance. A promising claim with unavailable timestamps is blocked, not supported. A failed script is not a failed hypothesis. A correct script whose target contrast is absent is an adverse result worth preserving.
Before promotion, check that preprocessing has not learned from the evaluation partition. The scikit-learn documentation illustrates this common leakage problem. The ledger should point to the data version and analysis revision, so the reported outcome refers to something another analyst can actually rerun.
Sources: scikit-learn, Common pitfalls and recommended practices.
Record cost and uncertainty alongside the science
Add a rough test budget, a maximum number of planned variations and the reason the question matters. These fields stop an interesting but unbounded idea from absorbing the entire engagement. They also make an early negative result commercially useful: the team has retired a route.
A confidence field belongs to the candidate at the time of prediction. Keep subjective conviction separate from evidence strength. A vivid impression can receive high personal confidence while still having no external support. The ledger should make that distinction ordinary rather than embarrassing.
Attach unresolved ambiguities explicitly. If a question has three plausible translations and no defensible way to choose, it remains ambiguous. A polished narrative should never be the mechanism by which ambiguity disappears.
Bring one decision and its existing evidence trail
For an initial research discussion, a useful starting packet is a non-confidential question, the available data description and one example of a decision your team needs to make. Existing dated notes can help identify which parts are hypotheses and which are already known.
The practical takeaway is simple: fund the next check, not the aura of the first impression. A Direction Preview can help turn one candidate into a bounded evidence challenge and a proposal for further investigation, without pretending that the entire research programme is already complete.
Questions this raises
Does documenting an impression make it scientific evidence?
No. It documents provenance. Evidence comes from a relevant external comparison whose assumptions and limitations can be inspected.
Can a failed candidate still be useful?
Yes. A well-tested failure can stop a weak route from consuming further time, provided the test actually addressed the candidate rather than a different question.
Sources and their limits
- Nosek et al. (2018), The preregistration revolution. Separating prediction from post-outcome explanation is the methodological context, not evidence for psionics.
- scikit-learn, Common pitfalls and recommended practices. Preprocessing must be learned on training data without leaking test information.
Prepared with AI assistance. The linked sources support the specified technical points; they do not validate applied psionics as a whole or guarantee a result for a client.
Read the editorial and evidence standard.
Continue reading
- Where hypothesis generation ends and validation begins
- How to read the resulting decision memo
- Explore the topic library
Explore Scientific Oracle consultingfor a scoped review of an existing-data research decision. Start with a non-confidential outline of the question, available evidence and the decision it needs to inform.