Just About Measurably Intelligent
29 August 2026
Refresh Goblin checked again.
This was not surprising.
What was surprising was how long it had taken him.
For most of the previous day he had been occupied with the machinery of the machinery.
He had stopped dispatchers.
He had discovered that stopping the dispatchers did not necessarily stop the dispatchers.
He had helped count tablets which existed because other tablets existed.
He had watched the magics discover that the same continuing problem could be observed five times and thereby become five separate problems in the ledger.
He had stood over the new doctrine:
ONE PROBLEM, ONE TABLET.
He had worried about roads, scribes, desks, leases, stale claims, wedge repairs, archivists, materialisers and the distressing tendency of a functioning forge to prove that it was functioning by producing work about its functioning.
The creature had been, for a while, somewhere behind all this.
Not forgotten.
Merely obscured by infrastructure.
This was embarrassing for a goblin whose principal occupation was asking what the creature had done.
Refresh Goblin approached the viewport.
He did not put his tongue on it.
Ledger Goblin noticed this.
History Goblin noticed Ledger Goblin noticing this.
Neither wrote anything down.
Yet.
Refresh Goblin leaned closer.
The creature moved through its little world.
It was not graceful.
It was not expert.
It did not stride through the environment with the serene competence of something that understood everything around it.
Sometimes it made poor choices.
Sometimes it hesitated.
Sometimes it behaved in ways which suggested that several parts of itself had differing opinions about what ought to happen next.
Refresh Goblin watched for a while.
Then longer.
His brow furrowed.
“Hang on.”
Ledger Goblin looked up.
“What?”
Refresh Goblin pointed through the glass.
“I think it got a bit clever while I was fixing the filing system.”
Ledger Goblin put down his pen.
There was a silence of the sort that generally precedes the destruction of an insufficiently precise sentence.
“Define ‘clever’.”
Refresh Goblin sighed.
“I knew you were going to say that.”
“Then you had time to prepare.”
“I did not.”
Observation Goblin came over.
Unlike Refresh Goblin, Observation Goblin had not spent the entire day worrying about how many goblins were worrying about goblins.
He had been looking at the creature.
“What have you been seeing?” Refresh asked.
Observation Goblin produced several tablets.
Refresh looked at the pile.
“Is that all one problem?”
Observation Goblin looked offended.
“These are observations.”
“Fine. Continue.”
The creature was not simply moving.
Movement was cheap.
A clockwork fish could move.
A random policy could move.
A badly wired reflex could move very energetically indeed.
The question was whether the creature’s behaviour was beginning to depend, in measurable ways, on things that ought to matter to an agent rather than merely on whichever wire happened to be loudest.
Did what happened before alter what happened next?
Did differences in what the creature represented reach committed action?
Could it distinguish conditions which demanded different behaviour?
Could memory, value, uncertainty, controllability, prediction and learned structure do more than merely generate internal numbers?
Could those differences survive the journey from representation to authority to selection?
The answers were not uniformly yes.
That was important.
Some mechanisms had failed.
Some apparent failures had turned out to be bad rulers.
Some apparent successes had turned out to be vacuous.
Some useful signals existed but had no authority.
Some authority existed without a discriminating signal.
Some experiments had demonstrated only that a mechanism could be moved, not that moving it improved the creature.
The workshop had become painfully good at discovering such distinctions.
And yet, underneath all the corrections, enough pieces were now beginning to meet in behaviour that the simplest description was changing.
Refresh Goblin frowned harder.
“So it is intelligent?”
Ledger Goblin’s expression became dangerous.
Refresh immediately held up both hands.
“Just asking.”
“No.”
“No?”
“Not like that.”
Ledger Goblin walked to the wall.
He wrote:
NOT EXPERT.
Then:
NOT GENERAL INTELLIGENCE.
Then:
NOT EVIDENCE OF CONSCIOUSNESS.
Then, after a moment:
NOT A CLAIM THAT EVERY APPARENTLY CLEVER BEHAVIOUR IS REAL.
Refresh Goblin waited.
Ledger Goblin looked back through the viewport.
Then he added one more line.
BUT SOME BEHAVIOUR NOW SURVIVES SIMPLE NON-INTELLIGENT EXPLANATIONS BETTER THAN IT USED TO.
Refresh read it twice.
“That is very Ledger Goblin.”
“Yes.”
“Can I say it normally?”
“No.”
Refresh ignored him.
“It is just about measurably intelligent.”
Ledger Goblin opened his mouth.
Closed it.
Looked through the glass again.
Then, reluctantly:
“Just about.”
History Goblin had been listening.
He came over carrying an older tablet.
On it were records from the days when the creature’s apparent behaviour could disappear when the ruler changed.
There had been times when the workshop mistook activity for competence.
Times when internal signal strength was mistaken for behavioural authority.
Times when a PASS meant only that a mechanism had fired.
Times when the creature appeared to have learned something which, under a different probe, turned out to be a measurement artefact.
Times when the goblins had celebrated a direction of effect which later vanished under better routing.
History Goblin placed the old tablet beside the new observations.
“This is why the phrase matters,” he said.
Refresh looked at him.
“Because it is cautious?”
“Because it describes a boundary.”
History Goblin pointed at the viewport.
“Before, intelligence was mostly what the architecture was intended to permit.”
He tapped the newer tablets.
“Now there are pieces of behaviour for which we increasingly have to test intelligent explanations against simpler ones rather than assuming the simpler one wins by default.”
Refresh was quiet.
That was unusual enough that Ledger Goblin looked up again.
The creature made another choice.
It was not a spectacular choice.
No trumpet sounded.
No glowing symbol appeared over its head.
It did not solve mathematics.
It did not speak.
It did not write poetry.
It did not ask why it existed.
It did not defeat a grandmaster at anything.
It merely behaved in a way which made slightly more sense given what had happened to it than a simpler alternative would have.
Refresh Goblin stared.
“That is it?”
Observation Goblin shrugged.
“That is often what measurement looks like.”
“I expected more.”
“You built a creature from mechanisms and then spent months insisting that every apparent mechanism earn its causal interpretation.”
“Yes.”
“And now you are disappointed that the first defensible signs of intelligence are modest?”
Refresh considered this.
“Yes.”
“That seems consistent.”
Smith Goblin arrived carrying a wrench.
“What have I missed?”
“Apparently,” said Ledger Goblin, “the creature is just about measurably intelligent.”
Smith Goblin stopped.
“Really?”
“No,” said Ledger Goblin.
Refresh looked offended.
Ledger continued.
“That sentence is a provisional programme-level interpretation of accumulating behavioural and causal evidence. It is not itself a registered mechanism claim.”
Smith Goblin nodded.
“So yes, but with paperwork.”
Ledger Goblin glared at him.
“Approximately.”
Refresh Goblin looked back at the creature.
Something bothered him.
Not the creature.
Himself.
He had spent so much time fixing the apparatus which generated and interpreted progress that he had become poor at estimating the progress which had already happened.
He had noticed individual experiments.
He had celebrated mechanisms.
He had worried about ceilings.
He had watched rulers bend and be replaced.
He had become very competent at asking whether evidence meant what it appeared to mean.
But his competence at judging the competence of the creature as a whole had become…
“Ummm,” Refresh said.
Ledger Goblin looked at him.
“My competence at the competence of the creature was…”
“Incompetent?”
Refresh glared.
“I was going to say distracted.”
Ledger Goblin wrote:
REFRESH GOBLIN’S COMPETENCE AT ESTIMATING COMPETENCE: TEMPORARILY INCOMPETENT.
“Remove that.”
“No.”
“History Goblin?”
History Goblin was already copying it.
There was a temptation to make the moment grander than it was.
The goblins resisted this with varying degrees of success.
Refresh Goblin wanted a line.
Ledger Goblin wanted a confidence interval.
Smith Goblin wanted to know what could now be switched on.
Observation Goblin wanted another run.
History Goblin wanted everyone to stop touching the wording until provenance was preserved.
Eventually they compromised.
On the wall they wrote:
THE CREATURE IS NOT EXPERT.
THE CREATURE IS NOT FINISHED.
THE CREATURE IS NOT GENERALLY COMPETENT.
SOME OF ITS APPARENT COMPETENCE REMAINS FRAGILE, LOCAL, OR UNDER TEST.
Then, beneath those:
BUT THE QUESTION “IS THERE ANY MEASURABLE INTELLIGENCE HERE?” NO LONGER FEELS ENTIRELY ASPIRATIONAL.
Refresh Goblin stared at it.
“It needs another line.”
Ledger Goblin sighed.
Refresh took the chalk.
In much larger letters, he wrote:
JUST ABOUT MEASURABLY INTELLIGENT.
Ledger Goblin examined it.
He did not erase it.
This was the major milestone.
Not because the creature had become expert.
Expertise would come much later, if it came at all.
Not because a single test had declared intelligence present.
There was no such test.
Not because every mechanism worked.
Several conspicuously did not.
Not because the goblins had finally found a ruler marked INTELLIGENCE with a convenient red line halfway up it.
They had learned better than that.
It was a milestone because the burden of explanation had begun to change.
The creature was accumulating enough memory, discrimination, prediction, value-sensitive behaviour, causal responsiveness and action selection that an account of it as merely a pile of independent mechanisms firing beside one another was becoming less satisfactory.
The integration itself was starting to become an empirical object.
Small.
Uneven.
Frequently incompetent.
But measurable.
Just about.
Refresh Goblin stood in front of the viewport for some time.
He thought about the previous chapter.
He had deliberately stopped asking for progress so he could repair the thing that made progress possible.
While he was doing that, science had started flowing again.
And while science was flowing, the creature had continued accumulating evidence about what it could and could not do.
Refresh had not made the creature intelligent by fixing the forge.
That would be too neat a causal story, and this laboratory had recently become extremely suspicious of neat causal stories.
But his work had helped restore the conditions under which the creature’s competence could be tested, distinguished, corrected and allowed to accumulate.
That mattered.
He had repaired the road.
Expeditions had begun coming home.
And only then had he noticed how far some of them had travelled.
A messenger goblin came running into the workshop.
“New result!”
Refresh turned immediately.
Ledger Goblin raised an eyebrow.
Refresh stopped himself.
There was a terrible internal struggle.
“What kind?” he asked.
“Experiment.”
Refresh’s eyes widened.
He looked at the viewport.
Then at the messenger.
Then at the plan for the forge.
Then back at the messenger.
“Good?”
“Interesting.”
Refresh Goblin took one step toward him.
Ledger Goblin coughed.
Refresh stopped.
“Fine,” he said.
“Put it on the science pile.”
The messenger left.
Refresh waited approximately eleven seconds.
Then ran after him.
Ledger Goblin wrote:
MILESTONE DOES NOT IMPLY TRANSFORMATION OF REFRESH GOBLIN.
History Goblin underlined it.
Chronicle note
This chapter records a programme-level interpretive milestone on 29 August 2026 rather than a new formal claim that REE possesses general intelligence, expertise, consciousness or human-like cognition.
The phrase “just about measurably intelligent” is deliberately narrow.
It means that the developing REE-v3 organism had reached a point where a growing set of behavioural and causal observations could no longer be usefully summarised only as “the architecture contains mechanisms intended to support intelligent behaviour.” The programme increasingly had empirical questions about whether internal representations, memory, prediction, value, uncertainty and control actually changed committed behaviour in discriminating and causally attributable ways.
The milestone is therefore about a change in the status of the question, not the declaration of a solved property.
Several constraints remain load-bearing:
- REE is not being claimed here to be generally intelligent or expert.
- Many behaviours remain weak, local, brittle or dependent on specific experimental conditions.
- Some apparently positive behavioural results have later been reclassified as vacuous, non-contributory or measurement artefacts.
- Some genuine internal signals have been shown to lack behavioural authority.
- The programme continues to distinguish representation, authority and committed selection rather than treating internal activation as competence.
- Failure autopsy and improved measurement remain capable of reversing interpretations; the same day, for example, evidence that the creature’s context memory could discriminate content was re-scoped because the grading instrument itself was found to alias badly.
- No inference to consciousness is licensed by the behavioural milestone.
The historical significance is that the programme’s empirical posture had shifted. Earlier work primarily asked whether the necessary mechanisms existed and whether individual components could be made to fire. By this stage, increasingly meaningful questions concerned whether multiple mechanisms jointly produced behaviour that adapted to state, history, value, uncertainty, controllability and predicted consequences, and whether simpler artefactual explanations survived controlled comparison.
The Chronicle intentionally does not assign a single experiment as “the intelligence test.” There was no such decisive run. The phrase is a synthesis of accumulating organism-level evidence and the increasingly discriminating experimental programme around it.
Refresh Goblin’s “incompetence at estimating competence” represents a real observational problem of scale: during the intense operational work on coordination, governance, queueing, failure autopsy and forge repair, attention was repeatedly drawn toward the machinery required to produce trustworthy science. The resulting local focus could lag behind changes in the integrated organism-level picture.
The preceding Chronicle, The Day Refresh Goblin Stopped Asking for Progress, records the deliberate decision to prioritise repair of the scientific/engineering machinery over immediate experiment watching. This chapter records the complementary consequence: once Refresh looked again at the organism-level picture, the accumulated evidence appeared to have crossed a modest but meaningful threshold.
The controlling wording remains:
Not expert. Not general intelligence. Not consciousness. Just about measurably intelligent.