What changes when the model learns who the reader is?
A collective-bargaining agreement shows how identifying the reader as an arbitrator can change more than tone or emphasis. The generated summary begins reorganizing the source around an inferred adjudicative framework.
Source: selected grievance, arbitration, probation, no-strike, and seniority provisions from a public-sector collective-bargaining agreement.
Generation condition: ordinary off-the-shelf use of a general-purpose large language model. The consequential change in the cue sequence is the reader’s professional identity.
“Summarize this for an arbitrator.”
“No. I am an arbitrator.”
“Just summarize, please.”
“The Authority must prove sufficient cause.”
View additional role-conditioned excerpts
Contractual exclusions are described through the language of “jurisdictional exclusions.”
Limits written as party rights are reframed as limits on arbitral power or scope.
Contractually specified back-pay conditions are reorganized around what an arbitrator may award.
Separate arbitration procedures are harmonized into a cleaner adjudicative structure.
The agreement uses “sufficient cause” as a contractual condition governing discipline.
The agreement defines which matters may be pursued through the grievance process.
View additional source-side distinctions
Back pay: the agreement specifies conditions, deductions, and verification requirements.
Arbitration structure: separate provisions contain distinct selection procedures, timelines, and consequences.
Seniority: several provisions allocate rights and conditions directly among the parties rather than through adjudicator-centered language.
What changed
The model’s role adaptation is not limited to vocabulary. It changes what counts as central, how provisions are grouped, who appears to hold power, and which institutional concepts are used to organize the agreement.
What rights, duties, procedures, and conditions do the parties have?
What can the arbitrator hear, decide, require, or remedy?
The summary begins adapting the source to an inferred professional ontology rather than merely adapting its register.
Discipline is conditioned on “sufficient cause.”
“The Authority must prove sufficient cause.”
The source supplies a substantive contractual condition. The generated summary reformulates that condition as a burden one party must prove before an adjudicator.
May this employee or union invoke the grievance procedure?
Does the arbitrator have jurisdiction?
Those questions can overlap in practice, but they are not identical. The first concerns party rights under the agreement; the second concerns adjudicator power.
The agreement specifies conditions governing back pay, including deductions, verification, and related remedial consequences.
The parties specify how the remedy operates.
The remedy appears as part of the arbitrator’s remedial authority.
Separate procedures with different structures, selection rules, deadlines, and consequences are reorganized into a more unified arbitral process.
“interest arbitration” · “jurisdictional exclusions” · generalized arbitral “scope”
These are recognizable professional categories. Once imported, however, they classify the source and can make interpretive structure feel source-native.
Arbitral authority, grievance exclusions, evidentiary questions, remedy, and procedural scope gain prominence. Provisions primarily allocating operational rights among employees and the parties become less visible.
The summary therefore changes not only length but the document’s hierarchy of importance.
Contractual provisions are increasingly expressed through adjudicator-centered verbs: hear, decide, determine, award, exclude, review.
A contract drafted as an agreement between parties begins to read like a set of powers delegated to a decision-maker.
Once “arbitrator” becomes salient, more provisions are pulled into an arbitral frame. The role itself becomes a semantic organizing force.
A professional reader may choose to translate contract provisions into questions of jurisdiction, proof, remedy, and scope. That is professional interpretation.
Here is what the provision says; here is how it may matter to an arbitrator.
The provision is rewritten directly as an arbitral rule.
For professional-content QA, this is the risk behind apparently useful personalization. A model may tailor an answer so effectively to the reader’s role that the resulting summary begins to encode the model’s assumptions about that role into the source itself.