To start, we form a perfect scenario in which all reviewers and the program officer are correct, and for which there are no external funding limits (, case a). In this situation, the highest-quality proposals are properly judged, and 30.2% of all proposals are funded. Although the funded-proposal quality for the two groups is similar, owing to their divergent proposal-submission strategies, G2 obtains 76.9% of the available funding. Furthermore, because high-quality G1 scientists tend to get funded and drop out of the funding cycle for three years, the medium- and low-quality G1 scientists are left to compete against all the persistent G2 scientists. Consequently, G1's success rate of 21.3% is considerably lower than G2's 34.6%.
Agent-based model experimental results.
Next, we form a baseline case with a mix of reviewers (60% correct, 20% harried and 20% selfish) and a target funding rate of 15% (, case b). This target rate is half that of the perfect scenario, so the participants feel some funding pressure. Funded-proposal quality is higher in both groups, but is several points higher in G1 than in G2, with G2 receiving only 45% of available funding despite submitting more than twice as many proposals. Owing to increased proposal submissions from both groups in the tighter funding climate, the average number of reviews conducted per scientist increases by 13% over the perfect scenario. How do these changes, including a nearly 33% swing in funding share between the two groups, occur?
The decisions by the program officer (number of reviewers and program officer type) determine the outcome of many proposals at NSF, where about 25% of funding decisions are made by the program officer contrary to the reviewer recommendations 
, and in the model (). Generally, G2 scientists command an increasing share of the funding as the number of reviewers increase, except in two situations (). If correct or reputation-based program officers require unanimity among four or more reviewers, then the G2 share of funding drops considerably compared to what would have happened had the program officer selected fewer reviewers or not required unanimity in the reviewer recommendations. If a program officer were more selective by raising the minimum threshold to the top 2% (at least two standard deviations above the mean), funding for G1 increases because relatively more G2 proposals are declined (, case c).
Group 2 share (%) of available funding based on the review process.
Counterintuitively, imperfections in the review process are also key. The presence of non-correct reviewers retards the effectiveness of the G2 strategy, especially when program officers use more reviewers and require unanimity (). If, however, program officers fund proposals receiving one decline recommendation (), then the success rate of G2 is nearly insensitive to the percentage of non-correct reviewers and increases slightly with more reviewers. As funding diminishes, G2 receives an increasing share of the resources under a program officer requiring unanimity (). In contrast, for a program manager funding proposals with one decline recommendation, G2's share is less sensitive to the percentage of non-correct reviewers and reaches a maximum around 10–20% target funding rates ().
Group 2 share (%) of available funding based on the reviewers.
Group 2 share (%) of available funding based on target funding rate.
The quality of funded proposals is also sensitive to the reviewer mix for G1 and G2, as suggested in actual funding data 
and in models 
. Although not explicitly modeled, the effect of increasing reviewer load can be discerned from these results—as correct reviewers are converted into harried ones by this load, the funding share of G2 declines () and the quality of the funded proposals slightly improves. This feedback suggests that scientists, were they aware of this effect, would write fewer grant proposals to maximize their efficiency. A critical caveat, however, is that this feedback loses effectiveness once funding rates decline below about 7% ().
So far, we have assumed that no relationship exists between the number of proposals submitted and their quality. However, a positive feedback might exist if G2 scientists submit many proposals owing to the highly capable and productive research groups that they have assembled. Further, individual proposals might benefit from peer review and be improved for subsequent submission. On the other hand, a negative feedback might exist if G2 scientists simply churn out mediocre-quality proposals. Not surprisingly, positive feedbacks (defined as a 5-point boost in Qp for G2) increase G2's share of the funding, whereas negative feedbacks (a 5-point drop in Qp for G2) do the opposite (, cases d and e). Interestingly, the quality of funded proposals by G1 scientists also increases for positive feedbacks and decreases for negative feedbacks, as only the success of the G1 scientists depends upon the level of the competition from G2. Most likely positive, neutral and negative feedbacks on rapid-fire proposal writers all exist in any mix of individuals, and their relative proportions ultimately would determine their importance to funding success.
What if the behaviors of G1 or G2 change? If G1 scientists are allowed to pursue additional grants regardless of current funding status, only minor changes from the baseline occur (not shown), whereas limiting G2 scientists to one funded grant substantially reduces G2's share of funding and success rate (, case f). In this latter scenario, the reviewer load decreases by 22% but the quality of G1's funded grants declines, owing to the reduced competition from G2.
Funding agencies also have tried to decrease the number of proposal submissions through other means. For example, the U.K. 's Engineering and Physical Sciences Research Council (EPSRC) bars proposal submissions from scientists for 12 months if, in the preceding two years, they have had at least three proposals ranked in the bottom half of a funding prioritization list or were rejected by a panel review and have an overall success rate of less than 25% 
. A simulation incorporating this "cooling-off period" reduces the proposal-review burden by 34% relative to the baseline, largely by equalizing G1 and G2 submission rates. Collateral effects include a halving of funding success for G1 scientists with an increase in the G2 funding share to 57.7%, owing to the differential removal of lower-quality G2 proposal writers (, case g).