Part of Temporal Constructs Appendix, which contains details behind the research: dates, quotations, descriptions of instruments, and sources.

General definition

In practice-based research, a record made from the working sessions of a practice. It recovers what was decided, by whom and on what basis from the transcript of the work.

Relation to the practice

The practice has kept every session it worked with its agents. The sessions begin on 22 April 2026, the earliest in the record, and six months of daily work make a corpus of hundreds of sessions and a complete record of a collaboration.

Turning the corpus into evidence means coding decisions. This page describes how that is proposed, what the corpus can and cannot support, and how far the work has reached. It is written as a method. The research it describes has not been carried out.

Session corpus

These sessions are held outside the archive, in the environment the agents run in. They were not recorded for research purposes. They cover the six months from 22 April to late October 2026, at eight hours a day, across CONDUCTOR and the works. They contain the exchanges themselves. Those exchanges make the corpus usable as evidence. They make the work difficult. Nothing in them is labelled as a decision.

Scope and limits

The corpus can support claims of the form this is what happened, here is the session, and here is the passage. That is the form the practice’s claims take, and the form artistic research can defend. The corpus cannot support rates, proportions or comparisons of frequency. It is one practitioner’s work, so a proportion computed over it describes these six months alone. The researcher would do the coding, a machine would search for terms the researcher chose, and no second coder would check the result. The sessions vary in length and content, so counts describe the corpus only, as in how many sessions contain work in KML, and do not support a claim about collaboration.

Material in hand

The archive already rests on cases that came out of these sessions. The masterclass of 4 March 2026, in which the sequencer M was voice-driven, came before the corpus begins, and it is documented on its own page. There is the run at MU in which the installation played between performance moments with no performer present. There is the transfer of agent-written code into Trip-A-Tron, which works and is not dependable. There is the navigation and verification layer built for the editor before an agent could compose in it. There are the tools built to read the Atari ST‘s screen. Every one of those is a session that can be dated and quoted.

Session record contents

The archive itself is a position. Five scheduled research tasks were set up in October 2026 and then set aside. The artist wanted a chronological and auditable retrospective that keeps sessions with no bearing on the project in the account. A session that looks irrelevant at the time can then be judged later against what turned out to matter. This entry and the Activity record are kept on that basis.

Coding outstanding

The codebook, the rule by which cases are selected, and the four groups of questions put to the corpus are on Coding the session record. That page also describes the safeguards that keep a search from only succeeding. It lists the files the scripts write and the work still outstanding.