Before March 2026 — The roots
Aviation communication, detection and recovery The framework did not begin from nowhere. Earlier work in aviation communication had already been concerned with detecting misunderstanding, checking, recovery and training people to recognise when communication was beginning to fail. At this stage there was no NL. But some of the problems that would eventually produce it were already present: How do we detect something early enough to do something useful about it?March 2026 — A Model for Good
The first substantial preserved representation We have a dated document called A Model for Good. Its core chain was: ACTION → DIFFERENCE → DETECTABLE DIFFERENCE → RECOGNITION → SUFFICIENCY → ACTION The threshold was central: when is the available signal good enough to justify action? The document already contained several ancestors of current NL: difference, detection, signals, thresholds, uncertainty, action and iteration. But it also made claims NL would later question, reduce or abandon. The framework had begun. It was nowhere near finished.Late March 2026 — Looking before moving
Attention moves upstream The original chain began with action. Very quickly a prior question appeared: What happens before the action? The record introduced a “zeroth step” — initially framed as the conscious decision to look and ask which direction was ‘good’. That formulation would not survive unchanged. But the move upstream would. This became an important recurring pattern in the development of NL: What has to happen before that can happen? Again and again, apparently fundamental starting points turned out to have something upstream of them.Spring 2026 — Signal, threshold and decision
The model is pulled apart During the following months, apparently simple concepts became less simple. What is a signal? What counts as recognition? What does confidence do? Is a decision selection — or elimination? What happens when neither option is sufficiently eliminated? What does WAIT mean? The model increasingly became less interested in prescribing the right answer and more interested in how something becomes available for judgment in the first place. The timeline becomes messy here. Because it was.June 2026 — The receiver becomes central
The architecture moves inside the receiver By June, the investigation had moved much further upstream. A major distinction emerged between processes requiring deliberate attention and processes that could operate before deliberate engagement. Expertise, habit, recognition, practice, improvisation and learned responses became ways of investigating this boundary. On 23 June, for example, the provenance records the working compression: Innate is hardwired. Pre-deliberate is learned. That exact formulation would itself require later qualification, but the important development was deeper: The receiver was no longer simply the person at the end of a signal. The receiver's existing architecture helped determine what could become available at all.Late June / early July 2026 — The snag moves upstream
Not everything reaches deliberate attention The developing architecture tried gates, turnstiles and eventually a router metaphor. Signals might effectively: PASS / ESCALATE / TRIGGER IMMEDIATE RESPONSE And the snag became increasingly important: something that did not simply disappear into normal processing. This was a major change. The question was no longer merely: What do we do with the signal? It became: Why did this become available for further processing when everything else didn't? That is recognisably the ancestor of today's Notice.July 2026 — Finite resources changes the picture
Why don't we investigate everything? A deeper organising constraint became explicit: Our resources are finite. Attention is finite. Time is finite. Effort is finite. Investigation has a cost. Judgment has a cost. Even opening a decision problem has a cost. This helped explain why noticing, ignoring, trusting, deciding, waiting and learning could not sensibly be considered independently. The provenance eventually captured the idea particularly neatly: The cheapest decision is the one you never open. NL was becoming less like a linear decision model and more like a framework concerned with how finite receivers allocate scarce resources in an information-rich world.July 2026 — Models of normal
Difference requires something to differ from Another question moved upstream: Why does one thing snag while another passes? Increasingly the answer involved the receiver's existing model of normal. What I notice isn't determined by the incoming signal alone. It depends partly on what I already expect. Experience can enrich that model. Expertise can increase its resolution. Something obvious to one receiver may be invisible to another. And learning can change the model against which tomorrow's signals are encountered. The emerging framework was becoming recognisably receiver-relative.Late July 2026 — A more stable architecture
A useful working sequence emerged: DETECTION → EVALUATION → COMMITMENT → ACTION → LEARNING But importantly, it wasn't intended to erase everything happening underneath. There were now pre-deliberate outcomes such as: IGNORE / PASS / ESCALATE / TOO FAINT and deliberate possibilities including investigation, deliberate ignoring and waiting. One particularly important distinction became clearer: Deciding whether something deserves a decision is itself consequential. NL was now concerned not merely with decisions, but with deciding to decide.August 2026 — Learning changes the receiver
The loop doesn't return to the same starting point The framework had often been represented cyclically. But a problem appeared: After learning, the receiver has changed. So: RECEIVER₀ → EXPERIENCE → LEARNING → RECEIVER₁ The next signal does not encounter precisely the receiver that encountered the previous one. Something that once required substantial resources may now be cheaply available. Something that once snagged may pass. And increased resolution may allow something finer to snag instead. The loop had become developmental. The process doesn't merely produce an outcome. It can help produce the receiver who encounters what comes next. That was a major step toward current NL.August 2026 — “A way, not THE way”
The claims become smaller as the framework becomes larger Earlier provenance contains remarkably strong language: universal mechanisms, laws of nature, claims about how decisions actually work. As the investigation became more sophisticated, the epistemic posture became more cautious. Functional resemblance wasn't mechanistic identity. A useful representation wasn't reality itself. The framework could describe something without claiming to have discovered the mechanism producing it. Eventually: A way, not THE way. NL didn't become weaker by giving up some of its strongest claims. It became more defensible.August–September 2026 — Units, rulers and perspective
Then came another deceptively small question: ‘unit?’ That opened a much larger investigation. A judgment depends partly on the unit, resolution, baseline, comparison group, threshold, purpose, timeframe and perspective used to construct it. 82% can be near the top of one reference frame and near the bottom of another. The fact hasn't changed. The representation has. The emerging diagnostic became: Relative to what? and: Tell me what unit you used to make the judgment. This strengthened an idea already running through NL: GOOD, BETTER, LARGE, SMALL, NORMAL, DIFFERENT, EXPENSIVE, SUCCESSFUL — relative to what ruler?September 2026 — NL turns onto itself
The framework was increasingly used to examine its own development. Why did this idea snag? Why allocate more resources to it? What did I treat as normal? What difference caused me to reopen something I'd banked? At what resolution does the claim hold? What ruler are we using to call this “better”? What changed in the receiver? The provenance itself became an NL application. Not: Look how long I've worked on this, therefore it must be right. But: Here is the trail. Inspect it yourself.September 2026 — Public release
This website becomes the next provenance entry. For roughly six months, NL has largely developed inside a very small system of receivers. Now I deliberately change that. More receivers → more models of normal → more differences → more opportunities for something to snag. Some people may find NL useful. Some won't. Some will misunderstand it. Some will identify earlier versions of these ideas elsewhere. Some may find contradictions. And somebody may see something I haven't. That's the point. The provenance doesn't end with publication. Publication changes what becomes available to the framework next.
Provenance Timeline
© The Noticing Loop - 2026
Before March 2026 — The roots
Aviation communication, detection and recovery The framework did not begin from nowhere. Earlier work in aviation communication had already been concerned with detecting misunderstanding, checking, recovery and training people to recognise when communication was beginning to fail. At this stage there was no NL. But some of the problems that would eventually produce it were already present: How do we detect something early enough to do something useful about it?March 2026 — A Model for Good
The first substantial preserved representation We have a dated document called A Model for Good. Its core chain was: ACTION → DIFFERENCE → DETECTABLE DIFFERENCE → RECOGNITION → SUFFICIENCY → ACTION The threshold was central: when is the available signal good enough to justify action? The document already contained several ancestors of current NL: difference, detection, signals, thresholds, uncertainty, action and iteration. But it also made claims NL would later question, reduce or abandon. The framework had begun. It was nowhere near finished.Late March 2026 — Looking before
moving
Attention moves upstream The original chain began with action. Very quickly a prior question appeared: What happens before the action? The record introduced a “zeroth step” — initially framed as the conscious decision to look and ask which direction was ‘good’. That formulation would not survive unchanged. But the move upstream would. This became an important recurring pattern in the development of NL: What has to happen before that can happen? Again and again, apparently fundamental starting points turned out to have something upstream of them.Spring 2026 — Signal, threshold and
decision
The model is pulled apart During the following months, apparently simple concepts became less simple. What is a signal? What counts as recognition? What does confidence do? Is a decision selection — or elimination? What happens when neither option is sufficiently eliminated? What does WAIT mean? The model increasingly became less interested in prescribing the right answer and more interested in how something becomes available for judgment in the first place. The timeline becomes messy here. Because it was.June 2026 — The receiver becomes
central
The architecture moves inside the receiver By June, the investigation had moved much further upstream. A major distinction emerged between processes requiring deliberate attention and processes that could operate before deliberate engagement. Expertise, habit, recognition, practice, improvisation and learned responses became ways of investigating this boundary. On 23 June, for example, the provenance records the working compression: Innate is hardwired. Pre- deliberate is learned. That exact formulation would itself require later qualification, but the important development was deeper: The receiver was no longer simply the person at the end of a signal. The receiver's existing architecture helped determine what could become available at all.Late June / early July 2026 — The snag
moves upstream
Not everything reaches deliberate attention The developing architecture tried gates, turnstiles and eventually a router metaphor. Signals might effectively: PASS / ESCALATE / TRIGGER IMMEDIATE RESPONSE And the snag became increasingly important: something that did not simply disappear into normal processing. This was a major change. The question was no longer merely: What do we do with the signal? It became: Why did this become available for further processing when everything else didn't? That is recognisably the ancestor of today's Notice.July 2026 — Finite resources changes the
picture
Why don't we investigate everything? A deeper organising constraint became explicit: Our resources are finite. Attention is finite. Time is finite. Effort is finite. Investigation has a cost. Judgment has a cost. Even opening a decision problem has a cost. This helped explain why noticing, ignoring, trusting, deciding, waiting and learning could not sensibly be considered independently. The provenance eventually captured the idea particularly neatly: The cheapest decision is the one you never open. NL was becoming less like a linear decision model and more like a framework concerned with how finite receivers allocate scarce resources in an information-rich world.July 2026 — Models of normal
Difference requires something to differ from Another question moved upstream: Why does one thing snag while another passes? Increasingly the answer involved the receiver's existing model of normal. What I notice isn't determined by the incoming signal alone. It depends partly on what I already expect. Experience can enrich that model. Expertise can increase its resolution. Something obvious to one receiver may be invisible to another. And learning can change the model against which tomorrow's signals are encountered. The emerging framework was becoming recognisably receiver- relative.Late July 2026 — A more stable
architecture
A useful working sequence emerged: DETECTION → EVALUATION → COMMITMENT → ACTION → LEARNING But importantly, it wasn't intended to erase everything happening underneath. There were now pre-deliberate outcomes such as: IGNORE / PASS / ESCALATE / TOO FAINT and deliberate possibilities including investigation, deliberate ignoring and waiting. One particularly important distinction became clearer: Deciding whether something deserves a decision is itself consequential. NL was now concerned not merely with decisions, but with deciding to decide.August 2026 — Learning changes the
receiver
The loop doesn't return to the same starting point The framework had often been represented cyclically. But a problem appeared: After learning, the receiver has changed. So: RECEIVER₀ → EXPERIENCE → LEARNING → RECEIVER₁ The next signal does not encounter precisely the receiver that encountered the previous one. Something that once required substantial resources may now be cheaply available. Something that once snagged may pass. And increased resolution may allow something finer to snag instead. The loop had become developmental. The process doesn't merely produce an outcome. It can help produce the receiver who encounters what comes next. That was a major step toward current NL.August 2026 — “A way, not THE way”
The claims become smaller as the framework becomes larger Earlier provenance contains remarkably strong language: universal mechanisms, laws of nature, claims about how decisions actually work. As the investigation became more sophisticated, the epistemic posture became more cautious. Functional resemblance wasn't mechanistic identity. A useful representation wasn't reality itself. The framework could describe something without claiming to have discovered the mechanism producing it. Eventually: A way, not THE way. NL didn't become weaker by giving up some of its strongest claims. It became more defensible.August–September 2026 — Units, rulers
and perspective
Then came another deceptively small question: ‘unit?’ That opened a much larger investigation. A judgment depends partly on the unit, resolution, baseline, comparison group, threshold, purpose, timeframe and perspective used to construct it. 82% can be near the top of one reference frame and near the bottom of another. The fact hasn't changed. The representation has. The emerging diagnostic became: Relative to what? and: Tell me what unit you used to make the judgment. This strengthened an idea already running through NL: GOOD, BETTER, LARGE, SMALL, NORMAL, DIFFERENT, EXPENSIVE, SUCCESSFUL — relative to what ruler?September 2026 — NL turns onto itself
The framework was increasingly used to examine its own development. Why did this idea snag? Why allocate more resources to it? What did I treat as normal? What difference caused me to reopen something I'd banked? At what resolution does the claim hold? What ruler are we using to call this “better”? What changed in the receiver? The provenance itself became an NL application. Not: Look how long I've worked on this, therefore it must be right. But: Here is the trail. Inspect it yourself.September 2026 — Public release
This website becomes the next provenance entry. For roughly six months, NL has largely developed inside a very small system of receivers. Now I deliberately change that. More receivers → more models of normal → more differences → more opportunities for something to snag. Some people may find NL useful. Some won't. Some will misunderstand it. Some will identify earlier versions of these ideas elsewhere. Some may find contradictions. And somebody may see something I haven't. That's the point. The provenance doesn't end with publication. Publication changes what becomes available to the framework next.
Provenance Timeline
