Every grade in the Israeli Defence-AI Deployment Index has exactly one threshold, and each threshold is crossed by one kind of published document. Nothing else moves an entry: not funding, not valuation, not a partnership, not the age of the company, not how the product is described. This is what defence technology evidence grading means in practice, stated as a set of artefacts a reader can go and look for.
L0 to L1: a named source describing a trial or evaluation
L0 records that a company claims a defence application and that no independent public source corroborates it. The document that moves it is an account, published by someone other than the company, of a trial, pilot, military evaluation or exhibition demonstration.
Three L1 entries show what that looks like. Israel Hayom reported in November 2025 on Wonder Robotics technology letting drones land autonomously in rough sea conditions. TechCrunch reported in December 2025 on Skana Robotics and fleets of underwater robots communicating with each other. The Defense Post reported on Althimis's Guardian drone defences in July 2026.
None of those reports describes a purchase. That is precisely why the entries sit at L1 rather than L2. The threshold is publication by an identifiable outlet of something that actually happened, not the significance of what happened.
L1 to L2: a publicly announced contract or order
L2 requires a publicly announced procurement contract or order. The announcement may come from the buyer, the seller or a regulated disclosure, and the index records which.
Steadicopter's entry rests on Naval Today reporting the Israeli Navy selecting its Black Eagle electric unmanned helicopter. Cognata's rests on The Times of Israel reporting the Defense Ministry engaging the firm for autonomous driving simulation with the IDF. Simlat's rests on Shephard Media reporting a contract award. Odysight.ai's rests on a Nasdaq-carried release announcing a purchase order from an unnamed international defence contractor. Cyabra's rests on a GlobeNewswire release describing a multi-year contract with an unnamed APAC intelligence agency.
Two things follow. An order of $3.4 million and an order of an undisclosed multimillion-dollar value produce the same grade, because the index does not weight by value. And an anonymous customer is still an L2, because the threshold is a public announcement rather than an identified buyer. Whether the announcement came from a company or from a reporter is recorded on the entry and examined in press release or reporting.
L2 to L3: a named source reporting operational use
This is the threshold that separates 13 companies from the other 37, and it is the strictest. It requires a named public source reporting that a military has used the system in actual operations. Exercises, trials and training do not qualify.
The distinction is not theoretical. D-Fend Solutions' index source is a feature describing EnforceAir2 deployed by the US Army during Combined Resolve 25-02. The word deployed appears in the source. The index still grades the entry L2, because the setting described is an exercise rather than operations. Compare SMARTSHOOTER, graded L3 on a Breaking Defense report from June 2025 that the US Army was expanding its use of the fire-control system, and Axon Vision, graded L3 on a Globes report of the IDF using its AI against drones.
The grading definitions are published in full. Their practical effect is that the top grade is largely outside a company's control: it depends on a customer being willing to be described, and a journalist choosing to describe it.
What does not move a grade
Money does not. Dream Security is graded L0. The Times of Israel reported it as a cyber unicorn valued at $3 billion aiming to help governments own and control their AI. The valuation is a real, dated, checkable fact and it does not corroborate a defence deployment. The general case is set out in funding stage is not deployment evidence.
Partnerships do not. AeroNous Solutions is graded L0 on an Innoviz investor release announcing a lidar partnership. A supply relationship between two vendors establishes that two vendors have a relationship.
Company self-description does not. Sightec is graded L0. Its index source is a release published through the Starburst accelerator describing over 3,000 "battle-proven deployments" of its GPS-free visual navigation technology. That is a larger claim than most L3 entries make, published without an independent source that a reader can check it against.
Age does not. Cogniteam, founded in 2007, is graded L0. Combatica, founded in 2021, is graded L3. TSG IT Advanced Systems, founded in 1969, is graded L3, and so is XTEND, founded in 2018.
And the same document can produce different grades for different companies. eyesAtop at L3 and Kela Technologies at L2 cite the same Calcalist report. What the article says about each company differs, so the grade differs.
Grades move down as well as up
The index is versioned. An entry can fall if its supporting source is retracted, if a report is corrected, or if a company sends a published source that contradicts the entry, in which case the entry is amended or removed. This is the same discipline the SIPRI Arms Industry Database describes in its sources and methods, where coverage is bounded by what companies and governments disclose and figures are revised as disclosure changes.
Falling grades are rarer than rising ones for a mechanical reason: published reporting accumulates and is seldom withdrawn. The index treats that asymmetry as a known bias rather than a property of the sector.
How to use this if you are a buyer or a company
For a buyer, the useful question is not what grade a supplier holds but which specific document is missing. An L2 supplier is one report of operational use away from L3, and asking who could publish that report is a more productive line of enquiry than asking the vendor to self-assess. The procedure is set out in how to verify a defence AI vendor's deployment claims.
For a listed company, the route up is a published source, not an argument. In the densest part of the field, the Israeli counter-UAS category, four companies sit at L0 and would move on a single independent report of a defence order. The full index and the wider distribution are set out in the deployment evidence gap.
Limits of this reading
The thresholds are bright lines applied to sources that are not always bright. Trade press sometimes uses "deployed," "fielded" and "in service" interchangeably, and the index has to read intent from context. Different editors would place some entries a band higher or lower, and D-Fend is the clearest example of a judgement call rather than a mechanical result.
The scale measures publication, not capability, and it is asymmetric by construction. A company can hold a classified contract, an undisclosed customer or Hebrew-only coverage and remain at L0 indefinitely while fielding systems. The index records that gap; it cannot close it.
Frequently asked questions
What is the difference between an L2 and an L3 grade?
L2 means a procurement contract or order has been publicly announced, with nothing published about the system in use. L3 means a named public source reports a military using the system in actual operations. Exercises, trials and training do not qualify for L3.
Can a company get its grade changed?
Yes, by pointing to a published source. The index amends or removes an entry when a company supplies a public source that contradicts it, and raises a grade when a qualifying document appears. Arguments about capability do not move a grade.
Does the size of a contract affect the grade?
No. A $3.4 million announced order and an undisclosed multimillion-dollar order both produce an L2. The index records whether a public announcement exists, not what it was worth or who the customer was.
Sources
- idf.ai, The Israeli Defense-AI Deployment Index v1.0, 24 August 2026
- Breaking Defense, Israel's SMARTSHOOTER says US Army expanding use of high-tech fire control system, June 2025
- Unmanned Systems Technology, D-Fend Solutions EnforceAir2 deployed by U.S. Army during Combined Resolve 25-02
- Naval Today, Israeli Navy picks Steadicopter's Black Eagle electric unmanned helicopter, 28 July 2022
- Israel Hayom, Israeli technology lets drones land autonomously even in rough sea, 13 November 2025
- The Times of Israel, Israeli cyber unicorn valued at $3 billion aims to help governments own and control their AI
- SIPRI, Arms Industry Database: sources and methods
Independent publication of idf.ai. Not affiliated with, endorsed by, or connected to the Israel Defense Forces, the Israeli Ministry of Defense, or any government body. Compiled entirely from publicly published sources. No classified, restricted or non-public information. Listed companies may dispute any entry: send the published source that contradicts it and the entry will be amended or removed.