top of page
Search

Reform as a Corpus

  • Writer: Aişe Gül Akkoyun
    Aişe Gül Akkoyun
  • Oct 17, 2025
  • 4 min read

Updated: 3 days ago

What UNCITRAL Working Group III looks like when you read it as data rather than as doctrine.

Background

UNCITRAL Working Group III met in Vienna from 22 to 26 September 2025 for its fifty-second session on investor-State dispute settlement reform. Two strands of work were on the agenda: the draft statute of a standing mechanism for the resolution of international investment disputes, and draft provisions on procedural and cross-cutting issues.

Two working papers give a sense of where the difficulty now lies. A/CN.9/WG.III/WP.253 contains draft provisions on procedural questions, including security for costs. A/CN.9/WG.III/WP.256 addresses the structure and design of a standing mechanism, and it is the more consequential of the two: whether the mechanism would be a two-tier body or two separate bodies, the nature of its jurisdiction, how States would consent to that jurisdiction, and how the whole thing would relate to the ICSID Convention.

Those are not abstract questions. They are the questions that determine whether a standing mechanism can function alongside the existing system or would have to replace it.

From principle to plumbing

The reform debate began with legitimacy concerns: inconsistent decisions, arbitrator independence, cost and duration, the absence of appeal. Those framings are now largely settled as diagnoses. What remains is institutional design, and design work is harder to write about because it is not obviously about anything.

Consent is the clearest example. A standing mechanism only has cases if States agree to be bound by it, and the mechanism for that agreement determines everything downstream. If consent is opt-in by treaty amendment, adoption will be slow and partial. If it is expressed through a multilateral instrument that modifies existing treaties, adoption could be broader but the legal engineering is considerably more delicate. The choice looks technical. It decides whether the mechanism has a caseload.

Security for costs is a smaller version of the same pattern. Whether a tribunal can require a claimant to post security shapes who can realistically bring a claim, and therefore what the caseload looks like. A procedural provision is a filter on access.

Why this is an unusually good corpus

Here is what makes WGIII interesting from a computational legal studies perspective rather than a doctrinal one.

The process generates a complete, dated, publicly available textual record: working papers, revised drafts, session reports, and State interventions across more than a decade. Each document supersedes an earlier one. Positions are attributed. Language changes between drafts are traceable.

That is a rare object in international law. Most treaty negotiation happens behind closed doors and surfaces only as a final text. Here the drafting history is the public record, and it is long enough to support longitudinal analysis rather than anecdote.

What could actually be measured

  • Provision survival: which draft provisions persist unchanged across successive working papers, and which are rewritten every session? Stability is a proxy for consensus that does not depend on anyone characterising the mood of the room.

  • Hedging language: track the ratio of mandatory to permissive formulations over time. A draft that moves from shall to may is recording a negotiation outcome in its grammar.

  • Intervention patterns: which States speak on which agenda items, and how has the composition of speakers changed as the work moved from principle to design?

  • Bracket density: the number of bracketed alternatives per provision is a direct measure of unresolved disagreement, and it can be counted.

  • Convergence timing: how long, on average, does a provision take to move from first draft to settled text, and does that duration correlate with subject matter?

None of this requires access to anything confidential. It requires treating the documents as a corpus rather than as a series of announcements, and it produces claims about the reform process that can be checked.

Why it matters beyond WGIII

The broader point is about method. International legal scholarship tends to read negotiation records for what they say. Reading them for how they change is a different exercise, and one that is now technically straightforward.

If reform is a decade-long process of incremental textual revision, then the object of study is the revision history, not any single draft. Institutional learning, if it exists, should be visible in the language before it is visible in the outcome.

The open question

WGIII has been working since 2017. That is long enough for a legitimate question: is the slowness a sign that the process is failing, or a sign that it is doing something genuinely difficult and doing it carefully?

The honest answer is that nobody knows, because the metric that would settle it does not exist. We have no baseline for how long multilateral institutional design should take, and no agreed way to distinguish productive iteration from drift.

A corpus-level analysis would not resolve the political disagreement. But it would let the argument be conducted over evidence rather than impressions, which is not nothing.

Sources

Comments

bottom of page