The coding rubric

Axis A: instrument preference

LBI-BAN Binding instrument with prohibitions

The state supports a legally binding international instrument that includes prohibitions. This covers both a full ban and the two-tier approach: prohibit some systems, regulate the rest.

LBI-OPEN Binding instrument, form open

The state supports negotiating a legally binding instrument but has not committed to prohibitions. The form of the instrument stays open in its statements.

REG-SOFT New non-binding measures

The state supports new international measures that are not legally binding: political declarations, codes of conduct, or guidelines.

CCW-ONLY CCW process only

The state supports continuing the consensus process in the Convention on Certain Conventional Weapons and its Group of Governmental Experts. It leaves open what that process should produce.

OPPOSE Opposes new instruments

The state opposes new international instruments on autonomous weapons. Its stated view is that existing international humanitarian law suffices.

AMBIG Ambiguous or evolving

The record supports more than one reading, or the position is moving. A first-class category. The evidence entries show the tension rather than hiding it.

NONE No substantive position on record

This project reviewed the record and found no substantive stated position. This is a coverage finding about the record, not a claim that the state holds no view.

Axis B: confidence

Confidence describes how directly the evidence supports the coding. It modulates visual saturation and a dotted-underline convention on the site. It never changes the category hue.

EXPLICIT

The state says it directly, in an official statement or document, in words a careful reader cannot mistake. Authoring a draft instrument counts, on the terms set out under "Authoring a draft instrument" below.

INFERRED

The coding rests on interpretation: voting behavior, sponsorship, alignment statements, or untranslated sources described rather than quoted.

AMBIGUOUS

The record points in more than one direction and the coding marks the best-supported reading of an unclear record.

PROVISIONAL

The coding rests on secondary reporting of a primary record this project has not yet retrieved and verified. A provisional coding carries a note saying exactly what is pending, and it is upgraded or corrected once the primary record is reviewed.

Authoring a draft instrument

When a state is a named author of a substantive draft treaty text, that authorship is read as an explicit statement of the position the text takes, and the coding is EXPLICIT. The text itself is the quoted evidence.

A draft instrument is the most committing written form available to a state in a negotiation. It is not a speech about what should happen; it is the operative language the state is asking other states to sign. A government that puts its name on an article beginning "It is prohibited to develop, produce, possess, acquire, deploy, transfer, or use under any circumstances autonomous weapons systems if:" has stated a position on prohibition in terms no careful reader can mistake, and has done so more concretely than most floor statements manage.

Read the conditions with the verbs, always. The article above prohibits conditionally: the conditions that follow its "if" are cumulative, so the prohibition reaches systems designed to attack outside meaningful human control AND whose use does not comply with international humanitarian law, while the rest of the instrument regulates. Quoting the verb list without the "if" converts a two-tier instrument into a blanket ban and overstates the author's position. This site quotes the trigger with the prohibition, or quotes the condition as its own evidence entry beside it.

This rule is narrow on purpose, and it is applied symmetrically: it reads authorship the same way for a draft that prohibits and for a draft that declines to. Three things it does not cover. Co-sponsoring a resolution is not authorship, and sponsorship never produces a coding by itself. Signing a joint statement is not authorship of an instrument. An author who publicly departs from the text it co-authored is coded on the later record, and the divergence is shown rather than resolved.

Authorship evidences the position the text actually states, and nothing further. Where a draft states an instrument preference on its face, its authors are coded to that preference at EXPLICIT. Where a draft states no instrument preference at all, this rule yields no Axis A category, and the honest coding is AMBIG with both readings shown, or NONE where the rest of the state's record is empty. It is not open to this site to supply the missing preference by reading the venue the text was filed in or what the text declines to ask for. Those are readings of a silence, and a silence can be read in either direction, which is the definition of AMBIG rather than a licence to pick one. The same reason the site refuses to treat a missing endorsement as opposition applies here.

A statement in a state's own voice is not overturned by an inference drawn from another document's silence. Directness ranks before recency: an EXPLICIT record governs an INFERRED one whatever their dates. The later-record rule decides between records of equal directness, and a later record can only displace an earlier one by saying something.

Every coding made under this rule shows the draft text, the document symbol, the date, and the state's name among the authors, so a reader who thinks authorship should count for less can see exactly what the coding rests on and discount it themselves.

Worked examples

Each of these is a coding you can check against its own evidence. Follow the link and read the documents the coding rests on.

United States

An explicit-statement case, and the site's clearest coded shift. The recorded vote arc runs Yes on 78/241, Yes on 79/62, No on 80/57, and behind it the coding moves from support for non-binding measures to a stated preference for keeping the work inside the CCW, dated to the explanation of vote that says so.

Austria

A coded position that moved, and the clearest illustration of why sponsorship is not evidence. Austria co-authored a 2018 paper calling for negotiation of a binding instrument without committing to prohibitions, which is LBI-OPEN, and by 2024 was stating in its own submission that it would promote negotiations for an instrument consisting of prohibitions and regulations, which is LBI-BAN. Austria also sponsored all three resolutions, and none of that sponsorship carries any weight in either coding. The statements do the work.

China

The hard case, and the reason AMBIG exists. China's own written papers hold two things at once: that general prohibitions would undermine legitimate defence capabilities, and that a legal instrument prohibiting a narrow class of systems could follow the precedent of the Protocol on Blinding Laser Weapons. That narrow class is defined by five characteristics a system must have in combination, so the prohibition China entertains is far smaller than the word suggests. China abstained on all three resolutions. The record supports two readings, and the evidence entries show both rather than choosing.