The project begins with an uncomfortable asymmetry. The standard transistor story has had decades of textbooks, commemorations, patent histories and technical reconstruction working in its favor. A serious audit therefore spends disproportionate effort looking for its weak joints. That is a search strategy, not a license to grade alternatives more gently.
First: did the detail actually matter?
The existence of the transistor does not make every feature of the successful apparatus necessary. Gold, two-mil spacing, a particular germanium sample or a particular sequence of accidents may have been one workable realization among many. Before a feature carries provenance weight, physics has to show that changing it would have changed the declared outcome—or at least that the success basin was narrow.
The winning path does not get to define its own requirements after the fact.
Then: where did the information come from?
For a consequential choice we separate five things: the raw observation; the contemporary interpretation; the next information-bearing action; the documented source of that action; and whatever information remains unexplained. Modern correctness is not historical provenance. “Could have derived” is not the same as “did derive.”
Simulation is an argument compiler.
A claim such as “material X was not practically available and was needed” is split into two independently built sets. Historical work reconstructs what qualities were actually available by the relevant date. Physics maps the parameter region in which the chosen observable exists. Only afterward do we ask whether those sets overlap.
This is why later mobility or lifetime measurements are allowed as sensitivity bounds but cannot silently enter a 1947 feasible set. A beautiful TCAD result is otherwise an efficient way to launder hindsight.
Reporting is evidence too.
Bell publicity, patent filings, Nobel narratives, Purdue institutional history, oral histories and textbooks are not interchangeable witnesses. Each has a purpose and therefore a predictable direction of compression. We track those bias vectors without converting “biased” into “false.” Ten later accounts repeating one Bell publicity lineage remain one information lineage.
What counts as progress
A dissolved anomaly is a result. A failed simulation stays visible. A source that cannot be acquired is recorded as a bounded failure rather than promoted to “missing.” The investigation succeeds if it leaves fewer but better questions.