Why this part exists

Parts 1 to 4 describe institutions selected for having succeeded. Any feature common to them will look causal, including features that were also common to institutions that failed and are therefore not remembered. This is survivorship bias in its textbook form, and research-policy argument is unusually prone to it because the failures are less documented, less narratively satisfying, and often quietly dismantled rather than publicly concluded.

The correction is not to conclude that nothing can be learned. It is to check each candidate lesson against cases where the same feature was present and the outcome was different.

The cases

Established The Superconducting Super Collider was cancelled by the US Congress in 1993 after roughly $2 billion had been spent and substantial tunnel had been excavated in Texas. It had scientific consensus, an unambiguous physical objective, and elite backing — the features Parts 3 and 4 identify as decisive. It failed on cost escalation and political support, which are not technical variables. The physics it was built for was eventually done elsewhere, later.

Established Japan's Fifth Generation Computer Systems project ran for about a decade from 1982 with substantial state coordination and industrial participation, aiming at massively parallel logic-programming machines for artificial intelligence. It produced working hardware and did not produce the capability it was organised around. The field moved in a different direction — statistical methods and general-purpose hardware — which no amount of coordination would have revealed in advance.

Established Concorde is the instructive commercial failure: a genuine engineering triumph, in service for 27 years, that never repaid its development cost and was ultimately withdrawn. Technical success and programme success are different things, and conflating them is a recurring error in speculative technology assessment.

Frontier Corporate research laboratories are the largest unexamined category. Many large firms operated central research laboratories during the Bell Labs era; most produced nothing comparable. Since we do not study them, we cannot tell whether Bell Labs' distinguishing feature was its structure, its funding, its recruitment, or the accident of solid-state physics arriving when it did. Speculative That uncertainty is the honest state of the evidence, and it is uncomfortable for everyone who cites Bell Labs as a design.

Frontier What survives the correction is thinner than the usual list. Across successes and failures alike, the features that appear to matter are a well-posed objective, authority to terminate work, and a downstream customer. The features that do not discriminate include scale, urgency, national coordination and elite endorsement — all of which the Super Collider and the Fifth Generation project had in abundance.

Handwave The strongest and least supportable claim in this area is that failures were foreseeable and avoidable. Most were not, at the time, by the people best placed to judge — and a research portfolio with no failures is a portfolio that was not taking risk.

What makes failure hard to learn from

Frontier The binding constraint is documentation asymmetry: successes are written up and failures are dispersed. Frontier Second is attribution — cancelled programmes fail for tangled technical, political and financial reasons, and disentangling them after the fact is often impossible. Frontier Third is that the counterfactual is unavailable: we cannot know whether the Super Collider would have delivered had it been funded. Established Fourth is incentive: nobody's career is advanced by a thorough account of a project that did not work.

Institutional lessons and hazards

Frontier There is a real institutional argument for documenting failure deliberately — registered reports, published negative results, and post-mortems on cancelled programmes — and the reason it does not happen is incentives rather than difficulty. Established For a project like this Institute's map, the hazard is direct: a catalogue of speculative technologies is a catalogue of candidate failures, and treating each entry with the confidence that hindsight lends to the successes would misrepresent all of them. Frontier That is part of why the four-flag system exists.

What survives the correction

  • Survives: Frontier a well-posed objective; authority to terminate; an identified downstream customer.
  • Does not discriminate: Established scale, urgency, national coordination, elite endorsement — present in the clearest failures too.
  • Unknown: Speculative whether the Bell Labs model was structural or circumstantial, because the comparison set was never studied.
  • Method: Frontier check every proposed lesson against a case where the feature was present and the outcome was not.