Negative results are published.
A positive result is informative only if the controls could have refuted it. Controls are specified first.
- 01
Preregister the gate
Thresholds, magnitudes and pass conditions are fixed before any data is collected.PREVENTS // fitting the threshold to the result. - 02
Run the destructive controls
Destructive controls, for example zero, random, shuffled or text-handoff arms, run beside each correct condition. Each entry lists the arms that ran.PREVENTS // mistaking an artifact for a signal. - 03
Audit independently
Hashes, doses and confidence intervals are re-derived by a second implementation.PREVENTS // a single implementation bug becoming a finding. - 04
Publish negative results
Failed gates and retracted numbers stay on the record, with the reason they failed.PREVENTS // survivorship: the reader sees what did not work.
A // Definitions
- Gate
- A preregistered pass condition for one stage of a claim. Gates are numbered and cannot be edited after data collection starts.
- Binding gate
- The gate that tests variable binding. The receiver must name the ID that goes with a queried color, including IDs that training never showed.
- Destructive control
- An arm that removes the thing being claimed. If the result survives it, the result was not about that thing.
- No-go
- A line of work whose gate failed or whose claim was falsified. Published in full.
- Rerun
- A second inference run of fixed jobs, with the same code, weights and hardware. Outputs must match exactly. It is not an independent replication. Counted and reported, including mismatches.
B // No-go ledger
| Code | Line of work | Result | Why it failed | Link |
|---|---|---|---|---|
| R.02 | Receiver-native controllability | [ NO-GO ] | The receiver-native gate failed at all five doses: masculine shifts did not beat norm-matched random directions (at dose 2.0: +0.007 nats, 99% CI −0.084 to 0.097). | |
| R.03 | Residual-stream injection | [ NO-GO ] | After a sign correction, Phase C passed 4 of 6 sub-gates; Phase D moved only the feminine margin; Phase E applied alpha twice, at the wrong token rows. | |
| R.06 | Binding failure mechanism | [ NO-GO ] | The trained bridge learns a separate code for each training ID, not a code built from letter and digit parts, and more training lowers the novel IDs further. |
C // Reporting standard
What every entry has to show.
- Thresholds are recorded before data collection.
- Destructive controls run beside every positive condition.
- A second implementation re-derives hashes, doses and intervals.
- Reruns are counted, and mismatches are reported.
- Negative results are listed alongside positive ones.