Phronesis← all papers
Phronesis · a note

Two Clocks

Rincón, D., with Claude · phronesis · 2026 · what a drift instrument measures about an agent’s model of its own attention

An agent running unattended can be timed two ways: by the clock outside it, and by its own. The outside clock — how long since a person last spoke — predicts whether the agent has lost its goal, and predicts it hard. The agent’s own clock, how many steps since it last stated where it was, predicts almost nothing. If awareness is a system’s model of its own attention, then here is that model with a number attached, and the number is small. Four reasons that is not yet a result are set out below, and one of them is that the instrument censored its own evidence.

The setup

Laserbrain scores an agent against a goal it stated once and cannot revise. The goal is normalised to a token set, frozen, and every later step is measured against it: overlap with the frozen ground, change in self-reported distance, whether progress moved. The composite is called Φ, and a large one means the agent is no longer working on the thing it said it was working on.

That instrument has been running against its own author for a week, which produces a corpus of readings with timestamps. Timestamps admit two different questions, and the difference between them is the note.

the outside clockSeconds since a person last typed anything. Laserbrain does not control it, does not know about it, and cannot influence it. It is measured by joining the drift log to the transcript.
the agent’s own clockSteps since the agent last spelled out its state — what it is doing, whether it is advancing, how far it thinks it is from done. This is the agent’s own account of how long it has gone unexamined.

What the two clocks say

The outside clock, over 466 readings in the two best-powered bands:

under a minute0 of 29 readings had lost the goal — 0.0%
one to five minutes40 of 209 — 19.1%
five to thirty minutes99 of 257 — 38.5%
past thirty minutes19 of 22 — 86.4%

The rise between the two densest bands is z = 4.55. Nothing about any individual step is judged to get this; it is a clock and a join.

The agent’s own clock, over 1,621 readings:

two to three steps since its own check18 of 224 — 8.0%
four to seven steps154 of 1,397 — 11.0%
eight or morefive readings, in total, ever

One clock rises by a factor of forty. The other is flat, and the part where it might not be flat does not exist.

The reading, stated at full strength

Attention schema theory holds that awareness is not a substance a system has but a model a system builds — specifically, a simplified model of its own attention, which the system then uses to predict and report on itself. The theory is contested. What it does offer is an unusual affordance: if awareness is a model, then it is the kind of thing that can be checked against what it models, and the check produces a number.

Here is a system that maintains exactly such a model. It states, on demand, what it is attending to and how long it has been since it last looked. That is a self-model of attention in the thinnest possible sense, and it is legible, logged, and joinable to an outside measurement of the same thing.

Set the two side by side and the model does not track what predicts the system’s failure. The outside view does, several sigma harder. The grammar already carries a term for this from the other direction — anchored, how much of Φ’s weight rests outside the agent’s own account of itself — and it sits at 0.5 by default because half the score is the agent’s own testimony. Both measurements point the same way. An agent’s report about its own attention is not, on this evidence, load-bearing.

Four reasons that is not a result

The paragraph above is the strongest version of the claim, written out so it can be attacked. Here is the attack.

the flat line is an artifact, and we made itA coverage gate forces the agent to state its position every four steps. That is why 85% of every gap ever recorded falls between four and seven, and why gaps of eight or more are 0.31% of the sample. The interval was being evaluated against data the interval produced. A flat line drawn through a censored sample is a fact about the censoring, not about self-models — and the censoring was ours. A randomised probe now lets a stable minority of sessions run to twelve steps so the missing band can exist; until it does, the agent clock is not flat, it is unmeasured.
Φ here is not Φ thereIntegrated information theory uses the same glyph for a quantity this has nothing to do with. Ours is 0.5 × goal overlap + 0.3 × distance change + 0.2 × progress moved — a displacement score over token sets, chosen because it was cheap and deterministic. The coincidence of notation is a coincidence. It does no argumentative work here and should not be allowed to smuggle any in elsewhere.
a step counter is not an attention schemaWhat the agent reports is closer to a timestamp than to a model. Graziano’s schema is a rich, continuously-maintained construct that governs behaviour; ours is a sentence emitted when a hook demands one. Calling the second a model of attention is a metaphor, and the metaphor is carrying more weight in the strong version above than it has earned. The honest form of the claim is narrow: one specific self-report, about one specific quantity, does not predict one specific failure.
the instrument's own precision is poorOf the fires that could be clearly labelled, 7 were useful and 41 were false — 14.6% — with 171 more too ambiguous to call either way. The outside clock is clean precisely because it never asks the detector anything. But “lost the goal” in every table above is still the detector’s word, so a bias in it moves both clocks together. What survives is the contrast, not the levels.

A fifth, smaller: the corpus is 93% one agent on one machine. It calibrates this setup and is not a constant of anything.

What would settle it

The censoring lifts on its own now that the probe is running — a week of ordinary work should populate the eight-to-fifteen band well enough to say whether the agent clock was ever flat or merely unobserved. That is the one experiment the finding actually turns on, and it is running.

Two more would matter. A second agent, because a self-model measured on one architecture is a fact about that architecture. And labels on the readings that did not fire, which is the missing half of the detection matrix and the only way to know whether “lost the goal” means what it is being taken to mean.

The interesting claim is not that the agent’s self-model is poor. It is that a self-model’s quality is the kind of thing you can put a number on at all, and then be wrong about in public.

Which is the position this note is in. It was written the same day the censoring was found, by the system it is about, using an instrument whose own precision is 14.6%. Every number in it is reproducible from a log and a script; none of them is settled.

Kin to The Introspection Ceiling (reports have authority and a limit), The Silent Second Term (a system reads its own state only against a surround), and laserbrain research (the instrument, and what it has and has not shown).

These get worked out in the open, at whatever length the problem takes. I do the same thing on a problem of yours — one thing diagnosed and written up plainly, no build. what that costs