Old relations retained
—
After competing experience
Acquisition & retention · issue 110
Learn. Keep.
Answer freely.
Can the same acquired brain learn new relations and keep the old ones? Inspect teacher-assisted learning, then recall with the teacher absent.
Loading the local experiment file…
Recorded campaigns
Compare the full founder census. Recipes and budgets differ; these results are not an efficiency comparison.
A viewer, not a simulated result
No experiment loaded
Open an exported microscope JSON file, or serve this directory with a real demo-data.json. Results, failures and missing measurements appear exactly as recorded.
Expected schema: cadence.acquisition-demo.v1
The complete founder census
New relations acquired
—
Same acquired brain
Held-out recall
—
Unused relation instances
Recorded teaching refusals
—
Refused attempts remain charged
The learning trajectory
Free recall at recorded checkpoints. No labels during evaluation.
Training family recallDevelopment family recall
Training—
Development—
Old at this checkpoint—
New at this checkpoint—
Replay navigates existing measurements; it does not run learning in your browser. A family credit requires every instance in that panel to be correct.
What do the controls remember?
Read the recorded comparator, information and scoring boundary.
One repair, four competing cues
Separate native evidence · local interference and bounded follow-up controls
At the historical two-module checkpoint, the original local update improves its taught cue while making the three other cue losses worse. A useful local repair can interfere with other relations in the same graph.
—of the taught cue’s loss reduction is offset by increases on the other three cues.
Local neuronal agreement leading to global equilibrium remains the common System 1 rule. Follow-up research chooses a local repair of this existing implementation through settings, wiring, context or memory readout. The other cues were not all acquired skills. Issue #110’s native gate remains unpassed.
Source, limits & separate diagnostic work
These historical summaries cover separately pinned native and actual-outcome studies, independently of the selected confirmation founder or uploaded file. Their raw archives are external. A dense reference checks a local direction; it never teaches or becomes the answer head.
The scope of this microscope
This task isolates acquired graph relations and old-skill retention. It is a learning mechanism demonstration; task accuracy and numerical qualification are distinct results.
Teacher labels assist acquisition. The answers shown here are free recall, not clamped teaching targets. This viewer makes no animal competence, transformer superiority or physical energy claim.
Issue #110 remains open. Native movie-action generalization is a separate pending gate; those labels summarize future actions rather than identify cue families. Inspect the research ledger.