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

General definition

The coding protocol for the session corpus sets out the unit a coded decision is and the rule by which cases are selected. It sets out the questions put to the corpus and the safeguards against a search that can only succeed.

Relation to the practice

The coding itself is set out here: what each decision notes, how cases are chosen, and the four groups of questions the corpus is asked. The page also sets out the reverse search that guards those questions and the files the scripts write. It states plainly how far the work has reached.

The corpus, what it can and cannot support, the material already in hand and the position behind the archive are on Session record.

Codebook

The unit is the decision at the scale of single exchanges, because who decided what is what the thesis turns on. Each coded decision notes four things. The first is what was asked and in whose words. The second is what the agent proposed or did and whether it acted unasked. The third is what was overridden and the reason stated at the time. The fourth is what failed and what was rebuilt in response.

Because the artist steers the distribution, the coding also looks for the moments when the balance is changed between delegation and collaboration, and for what caused each change.

Each coded decision also records which of four categories it falls under: delegated, withheld, initiated unprompted or overridden. It records whether the decision was thinking with or thinking for, and whether its result was kept, changed or discarded. The codebook is fixed before any searching begins, so the categories come from the research.

Case selection

Cases are selected by criteria announced in advance, so a reader can see the selection rule and the cases the current claims need. The first is the first time the agent acted unasked and what it did that was kept. The second is the run at MU, in which the installation played without a performer. The third is the most severe failure between an agent and an instrument and what was built in response. The fourth is an error that proved useful, and the fifth is the earliest session in which the system ran with nobody present. Any one of those would move a claim from its design to evidence.

Questions to the corpus

Four groups of questions are put to the corpus, one for each finding the research cannot yet support. The first group concerns what was decided. In which sessions did the agent propose something that was adopted unchanged, and what was it? In which sessions was what the agent produced rejected or rewritten, and what reason was given at the time?

In which sessions did the agent act unasked, and what became of that action? In which sessions was work done by hand that the agent could have done, and why? In which sessions can the origin of a decision no longer be assigned to either party, with the transcript showing the origin becoming unrecoverable?

What failed and was rebuilt forms the second group. Which sessions contain a failure between an agent and an instrument, what broke, and what was built in response? In which sessions did an error turn out to be useful, and is the moment legible in the transcript and not only in memory? What was rebuilt more than once, and what does the repetition show?

The third group concerns the instruments and the shared language. For each instrument in turn, which sessions show the agent learning to operate it, and what had to be written down for that to happen? In which sessions did the vocabulary of CUE change, and what prompted the change? In which sessions did the artist adapt their own method to the system?

The fourth group concerns the two runs. Which sessions fall in the period of the IFFR premiere, and which fall after it? What did the agent do at MU that it could not do in January, and where is that visible? What is the earliest session in which the system ran with nobody in the room, and what did it do?

Alongside those runs there is one automatic pass over everything. That pass records dates, hours, instruments present, work in progress and whether a failure or an override occurs. It produces the index of the corpus. The index shows which sessions are worth reading closely.

Safeguards

Every question is also asked in reverse. A search that can only succeed is not evidence. The reverse question asks whether delegation replaced learning or supported it, whether a proposal from the agent was better left unmade, and whether an override was a mistake. A reverse search that returns nothing is noted as a finding. Every claim has a session identifier, a date and a quotation, and the research does not count an untraced claim as evidence. The close reading will cover a named subset of the corpus. The page will state the subset and the rule by which it was chosen. An unstated subset is not a method.

Instrumentation

The codebook and the scripts that perform the index and the candidate search are held with the research materials. The archive is for pages only. The scripts read the codebook as a file, so the categories stated here and the categories the search uses cannot drift apart. Every term, every hit and every reverse search is written out, so the selection that produced the cases can be inspected. The scripts cannot decide. They find words, and whether a passage marks a delegated decision, a withheld one, an override or a failure is a reading only the researcher can make.

Work outstanding

The corpus has not been indexed. No decision has been coded and no case has been traced to a session identifier. The method described here has not been run. The page states that plainly as part of the account: the sessions exist and are complete, and the work that remains is the passage from transcript to evidence.

Two things are needed before that work can begin. Both are named here so that a proposal is not mistaken for the work done. The first is the sessions themselves, held in the environment the agents ran in and not yet brought within reach of the research. The second is the vocabulary.

The markers that find candidate decisions must be the phrases the artist uses. Until those are supplied, the searching runs on words chosen for it, which on the material tried so far return technical terms only. All of it is built and tested: the codebook, the scripts that perform the index and the candidate search, and the files they write. When the sessions arrive, the whole corpus can be indexed at once and the first cases taken from it.

Unforeseen

Here the unforeseen situations, sometimes difficult, sometimes unexpectedly interesting, will be described as the project runs. Many of these unforeseen situations have already occurred and have been recorded. They will be added in the time to come.