Experiment record / developmental-computation
Developmental computation: independent assessment
Original assessment and bounded engineering witness; both comparative protocols were unexecuted.
Question
Which narrow developmental-computation claim would survive conventional construction and ordinary growth controls?
Hypothesis
Contextual construction might improve transfer or growth utility; the assessment supplies discriminating protocols but no such comparative result.
Registration
Retrospective assessment publication. The two draft comparative protocols were unexecuted and not publicly preregistered.
Methods
Primary-source inspection is previously reported by the original assessment. A two-cell Scale/LeakyAccumulator witness accompanies the specification. Machine-local paths are replaced in the two Markdown downloads. The original HTML remains preserved locally and is excluded from this release; the four companion files listed here were found and inspected.
Baselines
Proposed parameterized conventional builder, ordinary function-preserving growth and matched controls; comparative protocols unexecuted.
Evaluation
The original check_results.json previously reported 17 engineering checks. A separate unchanged-source rerun in the continuation passed 17 checks; it is an engineering rerun, not independent reproduction of cited papers.
Resource ceilings
Original witness had one CPU worker, 30-second wall/CPU bounds, finite two-cell structures and a 512 MiB RSS ceiling. This assessment export launches no experiment.
Stopping conditions
The original witness and assessment closed before publication. No protocol execution is inferred from this release.
Executing code
Original witness SHA-256 0982ffb13fd18bbd5125bf4d0969e0e6aa7244730a57208462f141632ffb722a
Configuration
Original R-v1 engineering witness. No trained model or developmental-computation comparative campaign.
Complete outcomes
Assessment complete. Original witness: previously reported 17 passed checks with two recorded startup failures. Both comparative protocols unexecuted. Continuation witness rerun: 17 passed, original source unchanged.
Realized resources
Original record: process CPU 0.040172 seconds, peak RSS 15417344 bytes. New engineering rerun is accounted separately in the dated baseline report. Original address-space enforcement was unavailable and this was disclosed.
Limitations
No learning advantage, held-out-size transfer, biological equivalence, independent paper reproduction or public prior registration is established. Source inspections are retained historical assessments rather than a fresh exhaustive literature audit.
Reproducibility
Download checks.py and run python3 checks.py in a disposable directory; it writes check_results.json beside itself. Read research.md and handoff.md for original scope and limitations. Public Markdown replaces private local paths; hashes of originals label provenance rather than the redacted bytes.
Revision record
First public release on 14 October 2026. Original local assessment preserved. No earlier public availability is claimed.
Reviewed evidence & downloads
- research.md — derived. Selected project-authored material authorized for this report; cited papers remain linked, not redistributed.
- handoff.md — original. Selected project-authored material authorized for this report; cited papers remain linked, not redistributed.
- checks.py — original. Selected project-authored material authorized for this report; cited papers remain linked, not redistributed.
- check_results.json — original. Selected project-authored material authorized for this report; cited papers remain linked, not redistributed.
These are selected public derivatives; raw transcripts, internal configuration and private source are withheld. Hashes describe the actual downloads, not unmodified original evidence.
Artifact manifest and source provenance · Permanent revision URL