What we count,
and how far we will claim it.
Capital stage and commercialization stage are separate axes. Merge them and a company that merely raised money looks identical to one with paying customers. This page states how we keep them apart, and which sources we actually use.
An announcement is not a payment.
One award produces a new headline at each step. Summing those headlines counts the same money several times. We store each step in its own column and use only the appropriate one per metric.
Contract ceilings are never added to confirmed inflow — nothing guarantees the ceiling gets ordered. An undisclosed amount is stored as undisclosed, never as zero.
Every figure says where it came from.
Every figure carries how far it was checked. Collapsing them into one number would claim verification that never happened.
Pulled automatically from a government transaction or a filing. The source is the issuing record itself, not a press report. Nearly all capital events sit here
Based on an announcement or press report, not on the issuing record
An estimate, with the method stated
Inferred from context, never included in totals
Funding tells you who raised.
LCI tells you who deploys.
Raising money and deploying to customers are not the same thing. LCI measures the difference using only verifiable facts.
1. Commercialization Stage (0–6)
Customer adoption, not technical maturity
TRL measures whether it works. Stage measures whether customers pay for it.
Partnership alone never advances the stage. A “pilot” with no confirmed payment stays at 3. Stage 4+ requires contract or purchase evidence.
2. LCI (0–100) breakdown
Weights are public
Eight dimensions, weighted. Deployment carries the most weight; funding alone cannot move the score much.
3. Evidence Grade (A–D)
Shown separately from the score. Grade D alone never advances a stage.
- APrimary (multiple)Independent primary sources: customer announcements, government contracts, regulatory filings×1.00
- BPrimary (the company)The company’s own announcements, annual report and job postings×0.85
- CSecondaryTrade press, major media, research reports×0.60
- DUnverifiedSocial posts, anonymous claims, AI inference×0.25
Contracts are confirmed from the customer side where possible. Absence of information is never treated as evidence of absence.
4. Momentum (90/180 days)
LCI is position. Momentum is velocity — and timing shows up here, not in the total.
decay = exp(−days / 180)
LCI v0.1 is an experimental classification model. Its correlation with investment outcomes has not been tested, and it guarantees nothing about future performance. It exists to order which primary records to read first — it is not the ranking axis. We will backtest it against 100 North American robotics companies and update the weights from the result.
Cadence, lag, and how things are counted.
Cadence
Collection runs four times a day; capital-flow aggregation and report generation run daily. Outlays and recipient resolution run weekly.
Typical lag
NASA and DOE publish within about a month. The Department of Defense lags roughly three months and cannot be followed weekly, so it is kept as a separate retrospective. The lag is measured every run, not assumed.
Entity resolution
Recipient names are normalised, then matched on the SAM UEI. Similar names alone are never merged. Parent-child links come from USAspending recipient records, and rolling up to the parent is left as the reader's choice.
Deduplication
The same money recorded at different stages or from different sources is collapsed before counting. When one award is tagged to both a parent theme and a subtheme, only the subtheme is counted — counting both would double the total.
Kinds of amount
This-period obligation, lifetime obligation, contract ceiling, outlay, and Form D amount sold are five different numbers. Where they appear together the kind is always stated, and they are never summed.
Unknown stays unknown
An unknown amount is not zero. When an entity's type cannot be determined we do not call it an operating company; it is separated out as unknown. Filling the gap would put a guess next to verified facts.
Corrections
No published figure has been corrected so far. When one is, we will not simply replace the number — what changed, when, and why will be recorded here. A silent fix is invisible to the reader.
What runs today, and what does not.
Figures come from USAspending transaction records and SEC EDGAR Form D filings. Planned sources stay labelled as planned until they run.
- LIVEUSAspending.gov (federal award transactions)
- LIVESBIR / STTR (extracted from USAspending descriptions)
- PLANNEDSBIR.gov official API (unavailable upstream)
- BETASEC EDGAR (Form D)
- PLANNEDSEC EDGAR (8-K, 10-Q, 10-K)
- PLANNEDCompany newsrooms
- PLANNEDCustomer and partner announcements
- PLANNEDCompany job boards
- LIVEThe Robot Report
- LIVEIEEE Spectrum
- LIVETechCrunch
- LIVERobotics & Automation News
- LIVEJapanese-language web research
- PLANNEDGitHub / research papers
- PLANNEDPatents, regulation, certification
LIVE = in use for aggregation and reports. BETA = ingested but only partly used. PLANNED = not implemented. We never show a planned source as live, and sources we have decided against are removed rather than left listed as planned.
What we do not track
- Real-time equity market flows
- Position estimates from 13F
- Fund flows from N-PORT
- Full coverage of every sector and geography
- Estimates we cannot show the basis for