Lilienfeld 1926 Package Evidence for Independent Replication
# Package Evidence for Independent Replication: Three Items This Construction's Package Cannot Omit
The 1925 series closed by separating reproduction (re-analyzing the project's own data) from replication (building new specimens and collecting new data), and packaging protocols as executable checklists with every fixed parameter explicit. This construction's package needs that same discipline, plus three items specific to its own geometry that a package for the 1925 flat-electrode device never had to include: the transverse fracture procedure's narrow re-contamination window (Step 2), the Path B dwell-time methodology that separates capacitive decay from fracture-margin leakage (Step 8), and the glass-sheet correlation tracking needed for a valid reproducibility claim (Step 11). Omitting any of these lets a replicator build a device that looks complete but cannot be properly diagnosed.
The fracture timing window must be a stated number, not an implied best practice. A replicator unfamiliar with why Step 2 and Step 3 must be tightly sequenced could easily treat the fracture as a one-time preparatory cut, performed well in advance of electrode bonding, with no apparent harm — the fracture still looks the same to the eye hours later. Only an explicit statement of the re-contamination timescale (established in Step 2's own characterization) prevents a replicator from inadvertently defeating the entire point of fracture-mounting the electrode.
The Path B methodology must be packaged as a runnable procedure, not a description. It is not enough to tell a replicator that "capacitive and conduction leakage must be distinguished" — the package must include the exact dwell-time protocol, the fitting function, and worked examples of what a clean decay-to-zero looks like versus a plateau, so an independent party can actually execute the discrimination rather than reconstruct it from a prose description.
The sheet correlation log template is what makes Step 11's reproducibility claim auditable by a third party. Without a standardized template for recording which specimen came from which glass sheet, a replicator has no structured way to compute their own effective sample size, and a reviewer has no way to check that the original project's independence adjustment was done correctly. This is the one item on the list that affects a statistical claim rather than a measurement procedure, and it must be packaged as a literal template file, not a described practice.
| Step | Process operation | Input | Output | Specification | Constraint |
|---|---|---|---|---|---|
| 15.1 | Package Steps 1-14 as executable checklists, following the 1925 format | All step records from this construction | Checklist documents | Every fixed quantitative parameter (bias magnitudes, settling times, thresholds) stated explicitly | A checklist that cross-references prose explanation instead of stating the number forces a replicator to re-derive it |
| 15.2 | Write the fracture-timing-window document as a standalone, explicit specification | Step 2's characterized re-contamination timescale | Fracture timing specification | States the numeric window and what happens if it is exceeded | Burying this inside the general Step 2 checklist risks it being read as advisory rather than load-bearing |
| 15.3 | Write the Path B dwell-fit procedure as a runnable method, with worked examples | Step 8's dwell-time methodology | Path B procedure document | Includes the fitting function, at least one clean example, and one example showing a plateau | A description without worked examples leaves ambiguity in how to classify an ambiguous real dataset |
| 15.4 | Create the sheet-correlation-log template as a literal, fillable file | Step 11's specimen-to-sheet assignment structure | Sheet correlation template | Column structure matches what Step 11.6's effective-N calculation requires as input | A free-form log instead of a structured template risks a replicator recording insufficient detail to compute the correction |
| 15.5 | Attach provenance to every raw data file: specimen ID, instrument calibration reference, timestamp, checksum | Raw data from Steps 7-13 | Provenance-attached raw data | Carried forward unchanged from the original records; not recomputed at packaging time | Recomputing checksums at packaging time risks silently validating a file that was altered after original collection |
| 15.6 | Package analysis code for Steps 7 through 14's statistical and fitting procedures | Analysis scripts used in characterization and synthesis | Packaged analysis code | Code matches exactly what produced the published results, version-locked | Divergent code at packaging time means the package's reproduction path tests different logic than what generated the original verdict |
| 15.7 | Version the complete package against a specific project commit or state reference | All packaged materials from 15.1-15.6 | Versioned package, v1 | Any future correction produces v2, never a silent edit to v1 | Silent edits defeat the audit trail a versioned package exists to provide |
| 15.8 | Define the reviewer audit procedure explicitly, including re-running the Path B fit on provided raw data | Versioned package from 15.7 | Reviewer audit checklist | Audit explicitly includes recomputing at least one specimen's Path B plateau from raw I₁₃(t) data | An audit that skips this step cannot verify the one methodology most load-bearing to this construction's dependency graph |
Step 15 closes this construction's arc the same way Step 28 closed the 1925 series: by making every claim this project made checkable by someone who was not in the room. The three items specific to this construction's geometry — fracture timing, Path B methodology, sheet correlation — are not extra caution; they are the packaging equivalent of the dependency graph drawn in Step 14. Each one corresponds to a load-bearing result that, if a replicator or reviewer cannot independently re-derive it, leaves the construction's verdict resting on trust rather than on evidence anyone else can check.