Why it is a question
The owner, at U19, after CLAIM-470 and CLAIM-475: “let’s double click on the relationship between inquiry and interaction, since this might elucidate (or connect us to) a generic notion of observation/measurement that doesn’t require the presence of a conscious investigator. I think some of our earlier readings touched on this, maybe I’m connecting us back to apparatus here?”
A19 put the question as “What additional organization turns an interaction into a measurement—and what turns measurement into inquiry?” It answered with achievements that can come apart, “not automatically stages of one universal ladder”:
| achievement | what is added | a source the record holds or registered |
|---|---|---|
| relational event | an interaction makes one system’s state consequential for another | Rovelli (LIT-1635, read; THEORY-621) |
| measurement | the interaction separates alternatives into distinguishable outcomes, K_A(m, s′ given s), and may change s | Pattee (LIT-1190, read in NOTE-952) |
| stable record | the outcome stays accessible far longer than the interaction took | A19 §3; no source the record holds shows it (Di Biagio and Rovelli’s stable facts are a different notion; see the note of the readings below) |
| redundant record | many fragments of the environment hold the same outcome | Zurek (LIT-1640, unread) |
| use | the record changes what the system does next | chemotaxis, in A19 §6 |
| inquiry | the criteria of discrimination themselves change in response to what is found | CLAIM-470 |
A19 §2 gave a working definition: “A measurement is an interaction organized to make particular differences consequential in the state of another system.” A thermometer implements the question of which temperature, without understanding it. “Specification can be realized as selective physical sensitivity.” A19 added that “we should distinguish implementing a discrimination from understanding what the discrimination means.”
The bacterium is A19’s bridge case (§6). E. coli compares present receptor activity with a methylation memory that adapts more slowly. It runs or tumbles on the comparison, and so implements “Is the chemical environment improving relative to the recent past?” without entertaining the sentence. The memory’s time scale has to match the cell’s motion. It has every rung up to use. Whether it has the last one, revision of its own discriminations, is the question’s hard case.
A19 §7 proposed a criterion for the last rung: “Inquiry is an organized process in which distinctions acquired through interaction are evaluated relative to some operative criterion, and the results can affect the system’s future ways of interacting, discriminating, or evaluating.” A thermostat has a fixed rule. A bacterium adapts. A learner revises its policy. A science revises its instruments and standards. A19’s formal model has an apparatus A_t that produces records M_t, acts on them, and is then updated by A_{t+1} = G(A_t, M_t, U_t, E_t). With A fixed this is measurement. With A adaptive it is learning. When G can change which distinctions A makes, it is reflexive inquiry.
Why it is open here
Each rung needs a criterion, and the candidates differ. RQM, operational measurement theory and Pattee draw the measurement line in different places. For RQM, any interaction fixes a relative fact. For Pattee, measurement codes a dynamical state into a rate-independent symbol. Not every trace is a symbol in his sense.
Barad’s reading is stronger than any rung. A19 §5 separated three readings of an apparatus. On the informational reading it makes correlations. On the operational reading it implements and stabilizes distinctions. On the constitutive reading it participates in fixing the phenomenon’s boundaries (LIT-1653, LIT-1657, unread). The first two admit mathematics and experiment. “The third is a metaphysical claim that needs additional argument.” Dynamical stabilization does not derive it (A23 §VII). CLAIM-476 is stated in the agential sense.
The top rung can confirm itself. CLAIM-341 holds that each judgement reshapes the observer that makes the next observation. CLAIM-342 holds that what keeps the loop honest is evidence whose arrival or scoring does not pass through the interpretation it tests. A19 §8: “A self-reinforcing feedback system can be organizationally robust and epistemically defective.” “the world must have an opportunity to resist the inquiry’s specification.” A21 found the same failure in gluing. An apparatus that revises its correspondence maps only to reduce disagreement can send them to zero and agree with everything (THEORY-299; CLAIM-471). Revising by G is not enough. What has to be specified is what G is not allowed to rewrite.
Inferential descriptions are not inference. The free-energy principle offers a partition of internal, sensory, active and external states. The record’s readings (NOTE-450, NOTE-447) hold that a statistical partition admitting an inferential description does not show that the system infers or has that boundary. A19 §9 kept the principle as “one candidate formalization of nonconscious inference”.
Measurement has costs that must be stated correctly. A19 §10: measuring a bit does not by itself require dissipating k_BT ln 2. Landauer’s bound is on logically irreversible erasure (LIT-360, read). The costs arise when records must be robust, repeated, erased or used for control. CLAIM-475 needs that precision.
What would count as an answer
- A criterion for each rung that a constructed system can meet or fail.
- A system in which two adjacent rungs come apart. A19 suggested the strategy: “construct systems with identical sensory correlations but different memory persistence, downstream use, and adaptive capacities. Then determine exactly which observable behaviors each additional feature makes possible.”
- For the top rung, a structural condition on G, the part of the apparatus its own revisions cannot reach, under which revision improves prediction on interventions the system did not choose, and without which it does not. Distributed sensor networks that compare their reports through fixed correspondences (Robinson, LIT-1658) are a classical, conscious-free test bed, once their correspondences are allowed to be revised; Robinson’s are not.
Note of 2026-10-27 (evening): the readings of the U15–U23 works
Five works the question cites were read. Two attributions were corrected in place:
-
the “stable record” row named Di Biagio and Rovelli’s stable facts;
-
the last paragraph said Robinson’s networks recalibrate, and they do not.
-
Stable is not persistent. In Di Biagio and Rovelli 2021 (NOTE-1226; THEORY-1075), a fact relative to F is stable for a third system W: W may sum over it classically, because F has interacted with an environment W cannot reach. Stability comes from inaccessibility, not duration. Their measurement has three stages that come apart (pp. 5–6), and “It is the way that W, F and E couple to each other that makes F a measuring apparatus for W” (p. 6). Being an apparatus is a role relative to W (TERM-089).
-
An exact criterion for the correlation rung, and nothing above it. Di Biagio and Rovelli 2025 (NOTE-1214) define “B has information about A” as I_{A:B} ≠ 0, without any observer (p. 3). Nothing in the paper separates a measurement from any interaction that correlates. Its threshold for a fact is “high enough for the level of accessible resolution or confidence needed” (p. 3), so a purpose-relative threshold comes back exactly where the question would draw the measurement line.
-
The record rung. Zurek (NOTE-1229; THEORY-1053) derives the set of recordable outcomes from repeatability alone, eq. (5), which is a criterion for measurement that needs no observer. He also separates decoherence from usable records: a completely mixed environment “is useless (even though it can still induce decoherence!)” (p. 4). Redundancy as objectivity stays argued, not derived (THEORY-1055).
-
A negative case for the top rung. In discourse sheaves (NOTE-1222; THEORY-1088) the apparatus revises its expression maps using only the disagreement among its own outputs. It converges to agreement that its authors call “learning to lie”. Its guards keep the solution from collapsing; they do not make it answerable to anything outside.
-
A positive case. Robinson’s (NOTE-1221) is small but real: the true crash site “was withheld from the data presented to the sheaf model” and used only to score the result (p. 25). Something the apparatus never sees is the check, which is CLAIM-342’s condition in practice.
Note of 2026-10-31: a proposed study, and what the exchange’s constructions do not reach
The exchange “Analyze nucleation concepts” built toward this question from U20 to U23. A22 called its protected-inquiry model “a partial constructive answer” and noted the gap this question names: the model meets the requirement of testing on interventions the system did not select only by design. The design A23 wrote out is filed as CASE-186. Its primary endpoint is a second correction after the diagnostic and copying machinery is injured. Its out-of-family arm separates rejecting a hypothesis from finding the hypothesis family inadequate. That arm is still limited to a neutral response the architecture supplies, and it “does not establish spontaneous creation of a completely unforeseen model language”. The record’s runs nearest to it are CASE-161, CASE-162 and CASE-153, with CASE-144 (the evaluator that stops tracking) for captured scoring. A29 listed “endogenous repertoire expansion” among its hardest problems: building new kinds of experiment, reconstruction or governance beyond a family given in advance. That is this question’s second clause, and no new question is filed for it.
Note of 2026-10-31: Simondon’s signification at the measurement rung
The Simondonian readings filed today (THEORY-1276, THEORY-1278) give an observer-free account of meaning, and it lands on this question’s second rung and stops there.
- A19’s definition of measurement, in other words. Bencherki and Iliadis: “communication must be taken up by the receiving domain, which can only welcome this new contribution if it … is already structured in such a way to allow for the new action to make any difference. The difference that is made is the signification”. The same two students talking “mean” a disturbance in a classroom and play at recess (LIT-1922, ms. p. 7). That is “an interaction organized to make particular differences consequential in the state of another system” (A19 §2), with the receiver’s prior ordering doing the organizing. They extend it to any chain, finger to keyboard to router to screen, “each time reordering a domain” (the same page).
- It does not separate the rungs above. Signification is any reordering, so record, redundant record and use are not told apart, and nothing counts as a failure to inform (NOTE-1485, “Open questions”). That is RQM’s place in this question’s “Each rung needs a criterion”: every interaction qualifies.
- A distinction near the top rung. Iliadis quotes Simondon: information can be “exchanged between beings already individuated” but also “within systems to come that produce a new individuation” (LIT-1911, p. 11, citing Simondon 2007, p. 234). The first is exchange through a fixed receiver; the second changes the receiver’s own structure. That is the difference between A fixed and G changing which distinctions A makes, stated without a criterion. Same structure only.
- Nothing on answerability. Bencherki and Iliadis’s fourth proposition, that knowledge should “reflect on the way that knowledge itself has individuated” (ms. p. 22), is reflexive in the sense of this question’s second clause. It says nothing about what the reflection may not rewrite, which is the open part (CLAIM-342).
The question is unchanged and stays Open.