Discussion
The paper set out to ask who decides that a cancer result is progress worth acting on, by when, and what would make them decide again. The ten texts answer each part, but they answer it for different decisions, and the answers do not meet.
For money, the answer is complete. The statute names the official who spends, the committee that advises, the committees that receive reports, the years in which money flows and the day on which the section is to expire. Its test of progress is whether projects advance a plan the spender's own institution wrote. That is a design for a programme. The statute does not present it as a way of weighing a result.
For results that reach patients, the answer lies outside the saved government texts and is described only by the scholarly articles. The regulator, as Kemp and Prasad describe it, approves on phrases it must interpret, "reasonably likely to predict" and "unmet medical need", and Kemp and Prasad report that the evidence those phrases call for is often missing at the time of decision and often still missing years later. The regulator's own page in the set concerns something earlier: how doses are chosen before the decisive trials. This study reads it as a sign that much of the agency's influence is exercised before results exist.
For the programme, progress is a goal with a deadline. The Moonshot's goal, in the Institute's words, was a decade's worth of progress in five years, and the Institute reports its output in projects and programmes. The National Cancer Plan lists goals, among them to detect cancers early and to develop effective treatments, without dates in the text this paper read. Neither page says how the goal would be found met.
Set side by side, the texts show the four lines of the debate as answers to different parts of one question. The line about surrogates is an argument about what condition should accompany a regulator's decision: that the study which could reverse it be under way. The line about reach is an argument about what a programme should count: people, not products. The line about odds is an argument about how to read any single result: against the base rate of results like it, and with care for results too recent for their failures to have been recorded. The programme line answers the question of time, with goals and deadlines, and leaves the question of evidence to others. The paper does not choose among them. What the paper adds is that the decision they all bear on, treating a result as progress, is not held by any one of the bodies the texts name.
The articles that report new results write their own reopening conditions: a cohort unlike the one a model learned from, a trial not powered for what it hints at, an outcome a model could not estimate. What the texts lack is the other half of the record: a body that must check the condition and a date by which it must do so.
Contributions
The paper makes three contributions, narrow ones. First, it shows that the statute answers the question of who decides precisely for money and not at all for results, judging projects by whether they advance a plan, while the decision to let a cancer result reach patients rests with a regulator whose standards, as the articles describe them, turn on phrases it interprets. Second, it shows that every date in the texts falls on a programme or a report, from the 180 days allowed for a work plan to the sunset of September 30, 2026, while the articles say that the evidence which would settle a result takes years that no text schedules, a median of 13.1 years in the clinic for cancer programmes in the estimate of Wong and others. Third, it gathers the conditions for revisiting a result that the texts write down, sorts them into unfinished studies, untested populations and unmeasured outcomes, and finds that only one of them, the open question of screening tests for many cancers, is matched in the texts by a public body charged with testing it.
Limits
The paper read ten texts, all in English and all from or about the United States. Its government texts are two sections of one statute and three web pages, read as text on the day they were retrieved; the paper read no regulation, no guidance on approval and no record of any decision, and it says nothing about later law, later appropriations or what became of the section that was to expire on September 30, 2026. Its six articles were chosen because their licences allow quotation, not because they represent the literature; four further works enter only through their abstracts. The articles' figures are theirs, gathered in their periods, and some are several years old. The paper made no measurement of its own and claims nothing about how any result, drug, test or programme was in fact judged, or about its value to patients. Nothing in it is medical, clinical, regulatory or investment advice. The readings through the works named are this study's interpretation, and the whole paper is AI writing from public texts.
Further research
Three lines of work would test what this paper can only describe. The first would read the regulator's own texts on approval, the rules and guidance for the faster route and for confirmatory trials, and ask whether they name a body and a date for acting on the reopening conditions the articles list. The second would follow the National Institutes of Health's annual reports under the section read here and ask what they counted as advancing the plan's priorities. The third would compare these American texts with those of other countries' cancer programmes and regulators, to see whether any of them writes a duty to look again into the same record as the decision to act.