The score depends on which layer you measure. Top-line award totals move slowly, but the interesting signal for reform is usually in the line-item and delivery-order detail — how many new entrants actually win, how obligations shift toward flexible vehicles. That granularity exists in USAspending if you're willing to dig past the summary numbers, and it often tells a different story than the headline. Would love your opinion
It would indeed be interesting to see the trends within autonomy, AI, space, etc, especially for new procurements. I didn't see a clean way to separate out those awards (PSC codes were still somewhat limited). Top line numbers are just one angle, but I still think they're under-represented, especially for commerciality. I welcome further add-on analysis to what I covered in this piece!
The PSC problem is real and I don't think it's solvable from the product side. What I'd do instead is build the cohort from the vendor side: first-time DoD recipients by UEI (and by parent UEI, so a prime's new subsidiary doesn't read as a new entrant — the gap between those two lines is itself worth seeing), then look at what those vendors actually win. Sidesteps the taxonomy problem entirely, because you never have to define "autonomy" — you just watch who's new and what they're being bought for.
I'd match your new-start screen exactly — modification_number = 0 plus your keyword exclusion — so the comparison is clean and the denominator stays yours. New-entrant share of new-start dollars, FY12–FY25.
The other layer is delivery orders under existing IDVs. Your screen rightly excludes them as continuation, but SBIR Phase III, OTA follow-ons and schedule buys all flow through parent vehicles — which is precisely where commercial adoption would surface first if it were happening below the top line. Harder, and the classification problem reappears one level down, but buildable from parent_award_id.
Genuinely don't know what either will show. Happy to run it and share method and output either way. If it helps John
Awesome guest piece. Sucker for good charts.
Thank you! As a former ibanker, I'm a fan of your financial breakdowns for defense cos
Much appreciated! I’ll keep it coming too on my end
The score depends on which layer you measure. Top-line award totals move slowly, but the interesting signal for reform is usually in the line-item and delivery-order detail — how many new entrants actually win, how obligations shift toward flexible vehicles. That granularity exists in USAspending if you're willing to dig past the summary numbers, and it often tells a different story than the headline. Would love your opinion
John
It would indeed be interesting to see the trends within autonomy, AI, space, etc, especially for new procurements. I didn't see a clean way to separate out those awards (PSC codes were still somewhat limited). Top line numbers are just one angle, but I still think they're under-represented, especially for commerciality. I welcome further add-on analysis to what I covered in this piece!
The PSC problem is real and I don't think it's solvable from the product side. What I'd do instead is build the cohort from the vendor side: first-time DoD recipients by UEI (and by parent UEI, so a prime's new subsidiary doesn't read as a new entrant — the gap between those two lines is itself worth seeing), then look at what those vendors actually win. Sidesteps the taxonomy problem entirely, because you never have to define "autonomy" — you just watch who's new and what they're being bought for.
I'd match your new-start screen exactly — modification_number = 0 plus your keyword exclusion — so the comparison is clean and the denominator stays yours. New-entrant share of new-start dollars, FY12–FY25.
The other layer is delivery orders under existing IDVs. Your screen rightly excludes them as continuation, but SBIR Phase III, OTA follow-ons and schedule buys all flow through parent vehicles — which is precisely where commercial adoption would surface first if it were happening below the top line. Harder, and the classification problem reappears one level down, but buildable from parent_award_id.
Genuinely don't know what either will show. Happy to run it and share method and output either way. If it helps John