← Back to the README · Versión en español · Affective genealogy · Observer provenance · Bibliography
Lore Plugin is a product. LUS (Lore User System) is the research program from which it emerged. LUS studies what happens when a relationship accumulates criteria that can participate in later decisions. The software makes some of those questions operational; it does not prove the theory.
The name also carries two sounds that predate the acronym: luz in Spanish and us in English. The “user” in Lore User System is not a person standing in front of an assistant. The unit LUS examines is the relation.
LUS calls that unit the Between: a relationship in which each party's decisions change the other's space of possibilities, under an explicit purpose and through an iterable practice. A conversation is not a Between merely because two entities took part. Interdependence and purpose are part of the definition.
The human–AI Between is the program's primary and only evidence-rich instantiation today. Whether the structure holds across other disciplines or other compositions remains an open empirical question. LUS does not claim every relationship is a Between, and intimate relationships without an explicit discipline or iterable artifacts are outside its present boundary.
Three distinctions matter here:
- Experience is not everything that happened. It is an event that can still change a future decision.
- Criterion is the restriction left by that transformation. Writing it down does not make it causally effective.
- Lore is one operational architecture for making criteria visible, routable and revisable. It is not the whole of LUS.
Time in LUS is read as past, present and future: past experience is distilled or questioned in the present so it can restrict the future. The future may force the old criterion to be re-distilled, but it does not causally modify the past. LUS therefore does not posit a “temporal Between.”
LUS is currently in Phase VII — case studies, corpus version 1.24. Its records distinguish:
- a corpus, where the program's current conceptual claims live;
- case studies, which document situated observations and their limits;
- hypotheses, each with a state, an evidence count and a stated path to replication or rejection;
- a methodological Lore, which governs how evidence is admitted without turning method into theory;
- a registry of arbitrated external works, where a work enters by being arbitrated, not by being read, and every entry writes down where it loses against the program's purpose.
That last registry was completed on 2026-08-30 with the five dialogues the founding essay had listed since its first version and that had never been carried out. None entered confirming: three authors contribute one claim each and three remain as bibliography with their obstacle written down. See the bibliography.
Small samples remain a hard boundary. Most cases come from one researcher, one ecosystem and human–AI work. Repeated appearances inside that environment do not become independent replications by being counted separately.
These are open hypotheses, not product claims:
- H01 — LUS as an epistemological layer over disciplined development:
n=2, with two additional quantitative multidomain objects. It has not reached replication because longitudinal reduction of relearning in a real non-software practice has not been measured. - H02 — human–human Betweens:
n=0. LUS has no evidence for extending the framework there. - H13 — fertile effort and continuity:
n=1, situated in the same long-running Between. Fertility means recognizable movement after effort or correction; agreement, ease and frequency of use are not substitutes. The relation remains under study. - H14 — criteria without a point of application:
n=1. A criterion may be written and correct yet govern nothing because no procedure loads it, runs it or consumes its result. - H16 — the end of a Between is not its exit:
n=0. This conceptual distinction authorizes no product mechanism while it lacks a case. - H18 — the observer's prior repertoire conditions what they can distinguish:
n=0, with its blockage declared: testing it requires a Between with a different observer, which is what H02 has been waiting for since it opened. While it stays there it governs nothing. It was opened anyway because an unnumbered promissory note comes back as a fresh finding. - H17 — invoked criteria can still be too thin to produce:
n=1. A rule may load, run and affect the task yet remain sufficient for recognition and insufficient for generation.
The early problem of ephemeral experience remains historically important: learning can disappear between sessions even when facts survive. It is not the whole current research program. LUS now studies not only whether criterion exists, but whether it is housed, connected, invoked, consumed and sufficient for the work it claims to govern.
A program that studies a relation has a problem the others do not: whoever observes is part of the history that made it observable in that particular way. LUS treats this as an instrument and puts a hard limit on it, adopted on 2026-08-30 when von Foerster was arbitrated:
Situating the observer is a control procedure, never an epistemology.
Using “the observer participates in what they observe” as the explanation of a finding is forbidden. If it explained them, no finding could ever be wrong — and a program where nothing can be wrong has no hypothesis registry: it has a diary. Reflexivity enters to locate bias —declare the provenance, write down the coincidence, name the rival explanation— and stops there. The procedural half of second-order cybernetics is adopted and the constructivist half is rejected, because this program's apparatus of states and counters rests on something outside being able to knock a claim down.
For Lore Plugin this is a practical prohibition: the kit is not justified by saying that the observer participates. Its claims rest on its benchmark and its cases, with the distance to their evidence declared.
Each is a source that did not enter as criteria, with the exact question that would have to be resolved:
- Heidegger — how a relation can anticipate an ending that neither of its parts can anticipate on the other's behalf.
- Wegner — what a transactive directory is when one party does not persist.
- Maturana and Varela — how criteria can be transferable if only perturbation passes between structurally determined systems. The most expensive of the three: if it is resolved on their terms, the program stops saying that criteria travels.
The direction is controlled in both directions:
LUS observation → scientific arbitration → product decision, if warranted
Lore Plugin friction → product record → LUS evaluation, if scientifically relevant
Nothing crosses automatically. A product case does not become scientific evidence merely because it is public, and an open hypothesis does not become a skill rule because it is suggestive. Product usefulness and scientific validity are different judgments.
The benchmark in this repository measures whether the complete Lore Plugin 2.3.2 plus routed-Lore treatment changed Codex outputs under a frozen protocol. It does not measure the richness of the Between, professional development, general intelligence, or the validity of LUS as a whole. Lore Plugin records it as Product Case 19; LUS records the same event as Scientific Case 18. The ledgers are independent, so their ordinals are not expected to match.
The public LUS NotebookLM is an accessible introduction, not the source of record. The current scientific corpus remains local while the program is active. Public questions, external cases and refutations can begin in GitHub Discussions; appearing there does not make a contribution scientific evidence.
For the three kinds of provenance kept separate by the program, continue with the bibliography —who it argues with—, the affective genealogy —what sensibility made certain questions thinkable— and observer provenance —what distinctions the researcher already carried—. Of the third, the method is published; its content does not travel, and the reason is written there.