Skip to main content
sociable systems.
Episode 280 · 2026-10-07

Who Breaks the Tie

Adjudication needs the accounts kept separate. A reviewable determination names the chooser and records the reasons for setting an account aside.

Cover art for episode 280: Who Breaks the Tie
Many WitnessesAdjudicationAccountability
Episode 280: Who Breaks the Tie

Who weighs the two accounts now? Who wrote the rule that chooses? Who keeps the one that does not win so it can come back, living, whole?

Both accounts stand. Neither can be discarded. Someone, or something, has to decide, and the question becomes who, by what procedure, and with what left behind. Institutions already exist for this, and they treat adjudication as ordinary work.

The extractive-industries accountability mechanisms are the cleanest examples. The Office of the Compliance Advisor/Ombudsman is the independent accountability mechanism for the IFC and MIGA, and it handles complaints through two distinct functions (a third, advisory, works upstream of any complaint): a dispute-resolution function that brings communities and companies together to seek agreed outcomes, and a compliance function that audits whether the institution's own standards were followed (About us). The distinction is the whole design. Dispute resolution works when both accounts can be held in one room and turned into an agreed outcome. Compliance works when the question is whether the record meets the standard, and it produces findings rather than agreements. The CAO's architecture recognizes that these are different instruments and keeps them apart, so that no single process has to pretend to do both.

The World Bank Inspection Panel runs the same shape from the affected people's side: it determines the eligibility of complaints from people who believe a Bank-funded project has harmed them, and its process rests explicitly on the claim that affected people have a right to be heard. Eligibility is a decision about accounts: whether the harm described connects to the project as the Bank's own record presents it.

What adjudication needs from the record

A tie-breaker cannot work with one account. Every mechanism in this family requires the claimant's account to arrive as a structured claim and the institution's account to arrive as its own structured record, and the machinery exists to compare them. Watch what this does to the usual framing. Adjudication does not resolve the plurality of records. It needs the plurality preserved in order to function. A compliance audit that receives only the merged summary has nothing to audit. A dispute process that receives only the company's minutes is a hearing where one side's testimony was never filed.

The inversion is worth holding onto, because it reorders the whole pipeline question. The merge upstream is where adjudication goes to die, quietly, before anyone convenes a panel. By the time a tie-breaker sits down with a single flattened account, the decision has already been taken, by nobody, in a room nobody entered.

The rule that picks

What a good adjudicator produces is a third kind of record. The outcome is one. The reasons are another: which accounts were weighed, what rule or standard was applied, why the losing account was set aside, and what it would have taken for the outcome to go the other way. The CAO's compliance investigation reports and the Panel's eligibility findings both carry this structure. Dispute resolution, by design, produces an agreement instead, and its record is the agreement's terms. The reasons are what make an adjudication reviewable, and reviewability is what keeps the losing account alive in the only sense that matters to the losing party: the case can be reopened if the world changes.

A ruled outcome with no reasons is a coin flip with a seal on it. The seal is what the reasons buy.

When there is no panel

Most disagreements between a community and a company never reach the CAO or the Inspection Panel. They are settled, or left unsettled, inside a project-level grievance mechanism run by the company, which makes the company a party to the dispute and the clerk of its record at the same time. The UN Guiding Principles saw this coming. Principle 31 asks for mechanisms that are legitimate, accessible, predictable, equitable, transparent, rights-compatible, a source of continuous learning, and (at the operational level) based on engagement and dialogue. "Equitable" carries a specific meaning there: the aggrieved party should have reasonable access to the information, advice and expertise needed to engage on fair terms. A claimant who holds only their memory, facing an institution that holds the minutes, the register, the interpreter and the drafting pen, is some distance from fair terms.

The CAO's design has one more feature worth borrowing. When dispute resolution ends without agreement, the case can move to the compliance function. The two instruments are separate, and the route between them is written down. A project-level mechanism that offers dialogue with nothing behind it leaves the claimant to choose between agreeing and leaving.

The newest candidate for tie-breaker is a model. Ask one which of two accounts is more credible and it will answer, fluently, with reasons. The reasons will tend to favor the account that is internally consistent and neatly dated, which is to say the minutes. Consistency is a property of how a record was produced. It says little about whether the record noticed what mattered. A model weighing accounts can be useful if its reasons are kept and open to challenge, which is the standard every human adjudicator above is held to. A model that returns a verdict and discards its comparison has rebuilt the coin flip and left off the seal.

Companions


These notes come out of Sociable Systems, a practice that reads AI-shaped documents the way a hostile reviewer will, before a lender or a court finds the gap. The argument has an operational form: the Interim Protocol sets out four rules for AI use in environmental and social deliverables, covering disclosure at touch-point grain, evidence custody, the phrases no automated screening may settle, and a hostile read before anything ships. Free, and written to be cited or retired once institutional guidance arrives.