Aviation KOS App

retest-3...Open detail

agent drafted

Note content

Confirmed provenance

retest-3...

Link target

Annotation

Split text

Merge source note

Merged text

a1b2c3d4...Open detail

agent drafted

Note content

# Misregister — Design Move Rationale

Origin: offset lithography, silkscreen, risograph. Each ink layer applied separately; plate misalignment produces color fringing at contour edges. Historically a defect; reclaimed as intentional by designers.

Why it works for Auris: signals physical provenance (made on a real machine, not screen-native), shows the color separation structure as composition, reads as confident rather than perfected. References Weingart (70s Swiss grid puncture), Carson/Ray Gun (90s), Eastern European poster tradition, Bureau Borsche, Bráulio Amado.

Production execution: 2–3 spot colors, consistent offset (e.g. 4mm right / 1mm down, every instance). Apply to display type and photography only — not body text. Requires rigorous underlying grid; the move only works against discipline. Needs hand-tuning per layout; AI renders the idea but can't substitute for production craft.

Risks: nostalgic kitsch if overplayed; legibility loss on small type; reads as "trying" if the rest of the system is too clean.

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

a1b2c3d4...Open detail

agent drafted

Note content

# Auris Research — Brand Voice Rules

Declarative. Sober. Doctrine-fluent. Speaks aviation training as a first language; does not translate for outsiders.

Vocabulary: competence, calibration, evidence, criterion, observation, signal, standard, qualification, recurrent, lineage, finding, assessment. Verbs: captures, observes, calibrates, propagates, audits, infers, qualifies, recurs.

Forbidden: delightful, magical, seamless, empower, "we believe", "we think", "we're excited to". No hedges. No founder-first storytelling. No benefit vocabulary — doctrine over features.

Quantification as credibility: hours, events, signals, findings at actual operational scales.

Couplet tagline move (two balanced beats, often opposed): "What pilots show. What systems remember." / "Captured in the cockpit. Carried through the career." / "The record that flies forward."

Mission line: "Auris Research builds the empirical foundation of pilot competence."

Civic frame (sits beneath, not spoken aloud): aviation safety is the modern world's working civilizational achievement; the inferential record is the load-bearing piece; degrading it is worth defending against.

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

a1b2c3d4...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

a1b2c3d4... · direction decision references misregister rationale

a1b2c3d4... · direction decision references brand voice rules

0f01b3c0...Open detail

human

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

note-smo...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

note-com...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

note-hyp...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-com... · Definition underpins the claim — hypnometrics is the method, latency is the reason it's needed

note-fel...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-com... · Second proxy pattern supporting the claim — the felt/demonstrated gap is another surface expression of the latent construct

note-sca...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-com... · Concrete example of inference in practice — scan degradation as a proxy is a case of the latency claim

note-fel... · Two proxy types that co-occur under load — behavioural narrowing and self-report divergence appear together

note-aur...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

note-bra...Open detail

agent drafted

Note content

Confirmed provenance

ci-brand...

Link target

Annotation

Split text

Merge source note

Merged text

note-aur...Open detail

human

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

note-roo...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

note-thr...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-roo... · Model encompasses Room 1 — capture without judgment

note-roo... · Model encompasses Room 2 — sensemaking through notes and links

note-roo... · Model encompasses Room 3 — defensible claims

note-roo...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-roo... · Captured arrivals from Room 1 become the raw material for Room 2

note-roo...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-roo... · Notes and links from Room 2 ground the positions taken in Room 3

note-ux-...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

note-ux-...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

note-038...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

f8040674...Open detail

human

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

63ce75bc...Open detail

human

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

note-gam...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

note-pro...Open detail

agent drafted

Note content

A pattern is emerging in aviation training: AI tools that sit between the simulator data layer and the instructor assessment layer. The pipeline is consistent across vendors — raw FTD telemetry → pattern detection → competency scoring → adaptive training path. The value proposition is always some combination of objective assessment, reduced instructor burden, and personalized progress.

InstructIQ (Paladin AI) is one instance of this pattern. The market framing positions these tools as CBTA enablers: regulators are pushing competency-based approaches, training centers lack the instructor bandwidth and data infrastructure to implement them consistently, and AI fills the gap.

For product discovery, the relevant question is what this tooling layer does not cover: the assessment of the instructor themselves, the calibration of what "competency" means across contexts, and the sensemaking layer where findings from training sessions become organizational knowledge. The data-to-insight gap is being commercialized; the insight-to-knowledge gap is not.

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

note-com...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-pro... · InstructIQ is a concrete instance of the AI-CBTA tooling pattern described in the product discovery note

note-sys...Open detail

agent drafted

Note content

Basierend auf der konzeptuellen Analyse von CBTA, Deliberate Practice und Dreyfus-Stufenmodell lassen sich die fundamentalen Designprobleme für ein System zur Expertise-Entwicklung im Lernkontext auf zwei reduzieren.

Problem 1: Das latente Variablen-Problem. Kompetenz ist nicht direkt messbar. Proxy-Instrumente (OBs) haben eine definierte Gültigkeit, die bei Dreyfus-Stufe 5 und beim Transfer strukturell bricht. Ein System muss Kompetenz inferieren, nicht messen, und muss die Inferenz-Unsicherheit explizit modellieren. Das umfasst: Welches Instrument wird für welche Stufe verwendet? Wie wird mit Stage-5-Kompetenz umgegangen? Wie werden Transfer-Risiken eingeschätzt? Instructor-Kalibrierung ist ein Sonderfall dieses Problems: der Instructor als menschliches Gate kompensiert dort, wo das Instrument bricht.

Problem 2: Das Pfadoptimierungsproblem. Gegeben eine ausreichend gute Positionsbestimmung im Kompetenzraum (Problem 1), was ist der optimale Entwicklungspfad für ein Individuum? Kompetenzen sind Kategorien für Listen von Fähigkeiten. Fähigkeiten werden über OBs operationalisiert. Der Pfad durch diese Taxonomie ist individuell und zeitabhängig. Deliberate-Practice-Prinzipien definieren, was tatsächlich Bewegung auf dem Pfad erzeugt.

Abhängigkeit: Die Pfadoptimierung ist nur so gut wie die Positionsbestimmung. Problem 2 setzt Problem 1 voraus. Ein System, das Problem 2 gut löst, aber Problem 1 schleift, erzeugt gut optimierte Pfade durch falsch verstandene Positionen.

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

note-dre...Open detail

agent drafted

Note content

Dreyfus & Dreyfus (1980), ursprünglich für das U.S. Air Force Office of Scientific Research entwickelt, explizit auf Basis von Piloten-Daten.

Fünf Stufen:

Novize (1): Kontextfreie Regelanwendung. Situation wird in erkannte Merkmale zerlegt, darauf werden Regeln angewendet. Kein situatives Urteil.

Fortgeschrittener Anfänger (2): Erste Wahrnehmung von Situationsmerkmalen jenseits regelbasierter Features. Beginnt, wiederkehrende bedeutsame Aspekte zu erkennen. Alle Elemente werden noch gleich gewichtet.

Kompetent (3): Entwickelt Perspektive und Ziele. Handlungen werden im Kontext von Plänen gesehen. Wählt bewusst eine organisierende Perspektive, weil mehr wahrgenommen wird als verarbeitet werden kann. Entscheidungen sind analytisch. Die Person fühlt sich für Ergebnisse verantwortlich.

Gewandt (4): Situationen werden holistisch wahrgenommen, nicht als Merkmalsaggregat. Was passiert, erscheint direkt, nicht durch Analyse. Die Entscheidung, was zu tun ist, bleibt noch analytisch; die Wahrnehmung ist intuitiv. Eine Bibliothek diskriminierter Situationen ist vorhanden.

Experte (5): Situation und angemessene Reaktion erscheinen gemeinsam. Kein Regelwerk, keine bewusste Dekomposition, keine Deliberation im Normalfall. Handlung fließt aus Erkennung. Deliberatives Denken tritt erst bei Anomalien in Kraft.

Kernbefund: Der Übergang über die Stufen ist nicht graduell, sondern qualitativ. Bei Stufe 4-5 wechselt der kognitive Modus grundlegend von analytischer Regelanwendung zu impliziter Mustererkennung. Das ist keine Beschleunigung des bisherigen Prozesses, sondern ein anderer Prozess.

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-kom... · Dreyfus liefert die empirische Begründung dafür, warum Kompetenz latent ist. Ab Stage 4-5 ist Expertise tacit und nicht mehr in regelkonformes Verhalten dekomponierbar. Das macht die latente Struktur nicht bloß theoretisch, sondern beobachtbar am Verhalten von Experten.

note-obs... · Die Dreyfus-Stufen definieren den Gültigkeitsbereich des OB-Instruments. OBs erfassen artikulierbares Verhalten, was Stage 2-4 entspricht. Stage 5 entzieht sich dem Instrument strukturell. Dreyfus setzt die Grenze, die OBs nicht sehen können.

note-dp-... · Dreyfus beschreibt die Zustände auf dem Entwicklungspfad. Deliberate Practice beschreibt den Bewegungsmechanismus zwischen diesen Zuständen. CBTA beschreibt die Endpunkt-Architektur. Dreyfus ist die Karte, auf die alle drei Konzepte zugreifen.

note-kom...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-obs... · OBs sind eine Design-Antwort auf die latente Variablen-Prämisse. Zu verstehen, dass Kompetenz latent ist, ist Voraussetzung dafür, die Güte jedes Proxy-Instruments einzuschätzen. Note 2 stellt die Frage, Note 3 liefert eine spezifische Antwort und deren Grenzen.

note-sys... · Das latente Variablen-Problem ist direkt das erste fundamentale Designproblem. Diese Note benennt das Konstrukt; die Synthese-Note stellt die Designfrage: Wie inferiert ein System valide aus Beobachtungen auf dieses Konstrukt?

note-obs...Open detail

agent drafted

Note content

Observable Behaviors (OBs) sind das standardisierte Proxy-Instrument für Kompetenz im ICAO-CBTA-Framework. 73 OBs über neun Kompetenz-Kategorien. Jede OB ist ein konkreter, fremdbeobachtbarer Verhaltensausdruck einer Kompetenz unter spezifischen Bedingungen.

Struktur: Kompetenz-Kategorie → Kompetenz-Beschreibung → Observable Behaviors → Performance Criteria. OBs sind die messbare Schicht im latenten-Variablen-Modell.

Gültigkeitsbereich: OBs funktionieren gut für Dreyfus-Stufen 2-4. In diesem Bereich handeln Lernende nach erkennbaren Regeln und Plänen, die sich in fremdbeobachtbarem Verhalten ausdrücken.

Strukturelle Grenzen:

Stage-5-Problem: Expertise bei Dreyfus-Stufe 5 äußert sich in taciter Kognition und impliziter Mustererkennung, nicht in regelkonformem Verhalten. Das Instrument kann diesen Modus nicht erfassen. Die OB-Rubrik wird bei echter Expertise systematisch ungenau.

Transfer-Problem: OBs werden in spezifischen Trainingskontexten beobachtet. Ob die gemessene Kompetenz auf novel operative Kontexte generalisiert, ist eine separate Frage, die das Instrument nicht beantwortet. Hohe OB-Scores im Training sagen wenig über operative Bereitschaft unter unbekannten Bedingungen.

Konsequenz: Systeme, die OB-Scores als vollständige Repräsentation von Kompetenz behandeln, erben diese strukturellen Grenzen. "Objektive Bewertung" meint Konsistenz des Instruments, nicht Gültigkeit gegen das latente Konstrukt.

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-sys... · Die spezifischen Bruchstellen des OB-Instruments definieren die Design-Constraints für Problem 1. Stage-5-Blindheit und Transfer-Unsicherheit sind nicht nur akademische Limiten, sie bestimmen, was das System leisten muss.

note-dp-...Open detail

agent drafted

Note content

Drei Konzepte, die im Kontext von Expertise-Entwicklung oft zusammen genannt werden, operieren auf verschiedenen Abstraktionsebenen und sind nicht gleichartig.

Deliberate Practice (Ericsson 1993): Erwerbsmechanismus. Erklärt den psychologischen Prozess, durch den Expertise entsteht. Übung erzeugt Expertise nur unter spezifischen Bedingungen: Arbeit an der aktuellen Leistungsgrenze, unmittelbares spezifisches Feedback, Fehlerkorrektur, volle kognitive Aufmerksamkeit. DP ist keine Trainingsstrategie, sondern eine Theorie über den Mechanismus.

Szenariobasiertes Training (SBT): Instruktionsformat. Stellt Praxis in authentischen operativen Kontext. Begründung aus situated learning (Lave & Wenger 1991): Kompetenz ist kontextgebunden, isolierte Fertigkeiten transferieren schlecht. SBT ist eine Design-Entscheidung über das Format, keine Theorie über den Erwerb.

Kompetenzbasiertes Lernen (CBTA): Curriculum- und Bewertungsarchitektur. Definiert, was Expertise ausdrückt (Observable Behaviors, Performance Criteria), und knüpft Voranschreiten an nachgewiesene Leistung, nicht Stunden. CBTA ist die Ziel- und Bewertungsstruktur.

Komplementäres: CBTA definiert die Kompetenzziele. SBT liefert den Kontext, in dem diese Kompetenzen ausgeübt werden können. DP erklärt, warum Übung in Szenarien Expertise erzeugt, wenn die spezifischen Bedingungen erfüllt sind.

Kerntension: CBTA plus SBT ohne DP-Prinzipien erzeugt Szenario-Exposition, aber nicht notwendig Expertise-Wachstum. Wer Szenarien durchläuft, ohne an der Leistungsgrenze zu arbeiten und gezieltes Feedback zu erhalten, akkumuliert Erfahrung. Das ist nicht dasselbe.

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-obs... · CBTA als Bewertungsarchitektur erzeugt die OB-Taxonomie. Der Proxy ist eine Ableitung der Assessment-Architektur. Wer OBs bewertet, muss wissen, aus welchem Design-Entscheid sie stammen.

note-sys... · Die Dreiklang-Karte aus DP/SBT/CBTA erklärt, warum Pfadoptimierung (Problem 2) DP-Prinzipien benötigt und nicht nur Szenario-Exposition. CBTA gibt den Graphen, SBT gibt den Kontext, aber erst DP erklärt, was Bewegung auf dem Pfad erzeugt.

note-sta...Open detail

agent drafted

Note content

## Stage 5 / AI Decomposability — Open Research Angle

**The claim under scrutiny:** Dreyfus Stage 5 (Expert) involves tacit pattern recognition with "no rule decomposition" — the expert acts from holistic recognition, cannot articulate the underlying rules, and the performance cannot be decomposed into explicit criteria.

**Two-component analysis of the claim:**

*Component 1 — Phenomenological / tacit (Polanyi):* Expert practitioners genuinely cannot articulate the rules governing their performance. Introspective access to expert cognition is structurally limited. This component is grounded in phenomenology (Heidegger, Polanyi's "tacit knowledge") and is not obviously falsified by AI advances — the human expert still cannot tell you what they know.

*Component 2 — Technical / external decomposability:* In 1980, there was no method to externally reconstruct the pattern structure underlying expert performance without going through the expert's own articulation. This component was a contingent technological limitation, not a logical necessity.

**Andreas's premise:** Pattern recognition is a precondition for rule decomposition. AI has massively advanced external pattern recognition capability. Therefore the technical constraint on Component 2 is no longer what it was in 1980 — the axiom can be attacked on that dimension.

**What AI decomposition produces:** Not traditional rule sets (IF condition → action). Instead: probabilistic behavioral signatures derived from performance data (FDM, simulator recordings, etc.) that approximate expert behavior without requiring expert introspection. These signatures can be learned from flight data at scale, aggregating across many expert performers.

**Critical open question:** Does AI-derived decomposition need to be human-readable to be pedagogically useful? Two distinct use cases with different requirements:
- Automated competency assessment: human-readability not required — the model evaluates and scores
- Instructor-mediated feedback: decomposition must be interpretable by a human instructor to be actionable in a learning conversation

**Connection to existing framework:**
- Directly challenges the Stage 5 blindspot of the OB instrument (note-obs-proxy-struktur-grenzen)
- If behavioral signatures can be extracted at Stage 5, it potentially extends OB-equivalent proxy coverage — partially addressing the latent variable inference problem (note-systemdesign-zwei-fundamentale-probleme)
- Does not resolve Component 1 (tacit phenomenology) — expert cognition remains partly inaccessible to the expert themselves

**Status:** Open research angle. Not yet established whether AI-derived behavioral signatures constitute a genuine form of "rule decomposition" in the Dreyfus sense, or whether they are a different epistemic category altogether (external statistical approximation vs. internalized rule structure).

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-dre... · Stage 5 expert mode is the specific locus of the decomposability challenge; Dreyfus stages provide the developmental arc this note questions

note-obs... · AI decomposability potentially extends OB proxy coverage into Stage 5 — addresses the instrument's structural blindspot

note-sys... · If AI-derived signatures cover Stage 5, this partially addresses the latent variable inference problem — potentially expands the valid inference window

note-cbt...Open detail

agent drafted

Note content

## CBTA Competency Categories as Normative Artifact

**The problem:** The 9 ICAO PANS-TRG competency categories and their Observable Behaviors were not derived from performance data. They were constructed via expert consensus — domain experts divided the space of piloting behavior into categories, then sorted OBs into pre-existing buckets. The category structure is institutional, not empirical.

**Epistemological consequence:** This is upstream of the latent variable inference problem. The standard framing of CBTA assumes: competency is a latent variable; OBs are proxies; we infer the latent variable from proxy observations. But if the category structure itself was assumed rather than discovered, then the inference chain has a deeper problem: the assumed latent structure may not correspond to the actual dimensional structure of expert piloting performance.

**Structural risk:** The 9 competency categories may be correlated in ways that weren't modeled (committee members lacked the data to detect covariance structure). The real competency space might have fewer dimensions, more, or a non-orthogonal structure. Any downstream measurement and inference built on these categories inherits this foundational uncertainty.

**Cross-domain observation:** Different domains define different OB sets for structurally similar learning problems (aviation: 9 categories; some school frameworks: 4). The variation suggests the category counts are artifacts of how specific expert groups divided the problem, not reflections of a domain-invariant underlying structure.

**Feedback loop quality:** The current assessment model closes the loop through highly manual processes — instructor observation, OB-graded evaluation forms. This is low bandwidth and susceptible to inter-rater variability at the category level, not just the OB level.

---

**Open questions:**

1. How was the ICAO PANS-TRG competency framework actually constructed? What is the documented methodology — expert panels, Delphi method, literature synthesis, operational incident data? Was any data-driven validation of the category structure performed before or after publication?

2. Is there published psychometric evidence for the independence (or interdependence) of the 9 ICAO competencies? Factor analyses, inter-rater reliability studies that tested category structure, not just OB scoring?

3. If the category structure is normative, does this invalidate the CBTA framework operationally, or does it only become a problem at the extremes (Stage 5 assessment, cross-context transfer)? Where exactly does the normative origin matter?

4. Are there alternative competency frameworks (military aviation, medical simulation, high-reliability industries) that were derived differently — and do they converge or diverge on the category count and structure?

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-obs... · Normative category construction is upstream of the OB proxy problem — if the categories are wrong, the proxy chain has a more fundamental flaw

note-sys... · Normative ontology origin compounds the latent variable inference problem — the inferred structure was assumed, not discovered

note-dat...Open detail

agent drafted

Note content

## Data-Derived Competency Ontology

**The opportunity:** If competency category structure was assumed rather than discovered, then the right question is whether the structure can be derived from performance data directly — without pre-specifying categories. Modern data sources (FDM, simulator recordings, communication transcripts, eye-tracking, physiological sensors) make this tractable in ways it wasn't in the 1980s or early 2000s.

**What "data-derived" means here:** Instead of expert-consensus categories → OBs → measurement, invert: collect dense behavioral observations → apply dimensionality reduction or clustering methods → discover latent structure → use that structure as the measurement basis. The categories emerge from the data rather than being imposed on it.

**Why this matters for the measurement problem:** A data-derived ontology could:
- Reveal the actual covariance structure of expert behavior (which categories, if any, are genuinely independent)
- Identify dimensions that expert-consensus panels missed or merged inappropriately
- Scale measurement bandwidth beyond what's achievable through manual OB grading
- Enable assessment coverage at Dreyfus Stage 5, where OB instruments currently break (the behavioral signature is in the data even if the expert can't articulate it)

**Connection to AI decomposability angle:** This is a different framing of the same underlying possibility. The question isn't "can we decompose Stage 5 cognition into rules?" — it's "can we build a measurement infrastructure that doesn't require articulation at all?" FDM data from expert pilots doesn't require their introspection.

**Human-readability constraint:** A data-derived ontology may not produce categories that are naturally human-interpretable. This is acceptable for automated assessment legs, but creates a challenge for instructor-mediated feedback — the instructor needs to be able to explain something actionable to the trainee. The gap between machine-derived structure and pedagogically deliverable feedback is a real constraint, not a temporary limitation.

---

**Open questions:**

1. Has anyone attempted unsupervised discovery of competency structure from aviation FDM or simulator data at scale? What methods were used and what did they find?

2. What is the minimum human-readability requirement for a data-derived competency category to be pedagogically useful? Is there a principled way to constrain the discovery process to produce interpretable outputs?

3. How much performance data (hours of FDM, number of expert pilots) would be required for stable dimensionality discovery? What are the practical data acquisition constraints in the aviation training context?

4. If a data-derived ontology yields a different structure from ICAO's 9 categories — say, 6 dimensions or 14, or a hierarchical structure — how would that interact with the regulatory framework? ICAO PANS-TRG is normative; operators must comply. Is there a path from empirically derived structure to regulatory uptake?

5. What is the relationship between data-derived competency dimensions and individual differences in learning path? If the structure is discovered from aggregated expert performance, does it capture the variation in how novices develop through Dreyfus stages?

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-cbt... · Data-derived ontology is the proposed alternative to the normative-artifact approach — addresses the upstream structural problem

note-sta... · Both notes converge on bypassing articulation bottleneck — data-derived ontology is the structural framing, AI decomposability is the mechanism

note-sys... · Data-derived ontology could expand the valid inference window for the latent variable problem — and reframe what "path optimization" optimizes toward

note-cbt...Open detail

agent drafted

Note content

## CBTA Governance Layer: De-Risking as Governing Purpose

**Reframe of the normative-artifact critique:** The CBTA competency framework was not constructed arbitrarily by a committee. The governance layer was formalized from accumulated operational evidence — decades of incident data, accident investigations, CRM failures, airmanship analysis. The 9 ICAO competency categories reflect what actually causes harm in aviation, codified into a conformance structure. The institutional layer has an empirical origin, but the empirical base is safety events, not performance variance data.

**This matters for reading the ontology critique correctly.** Two separate claims must be distinguished:

- Claim A: The categories were arbitrarily invented (wrong — the origin is accumulated safety intelligence)
- Claim B: The categories may not optimally partition the variance that matters for expertise development (still open — safety-event-derived categories capture what goes wrong at the tail, not necessarily the full structure of how competence develops)

Both claims can be evaluated independently. The normative-artifact note should be read as making Claim B, not Claim A.

**What "grading" means in this context:** Assessment in CBTA is not academic credentialing. It is a conformance check against a safety-derived governance layer. OBs are operationalizations of that governance framework, not a theory of expertise. The optimization target is risk reduction, not optimal expertise development. A pilot can pass every OB check and still be mediocre; the system's goal is to close the most dangerous gaps, not to produce the best possible pilot.

**Two different optimization targets:**

1. Governance / de-risking: satisfy the conformance layer, eliminate the most dangerous deficits. This is what CBTA is actually designed to do.
2. Expertise development: move an individual as far as possible along the competency dimensions. This is what a full expertise development system would need to do.

The current framework is designed for (1). The research agenda implicit in the other notes is addressing (2), which requires extending beyond the governance layer.

**Implication for the data-derived ontology question:** A data-derived ontology for de-risking purposes might look different from a data-derived ontology for expertise development. Safety-event data (accidents, incidents) biases toward tail-risk dimensions. Performance variance data (FDM from all pilots, not just accident-involved ones) might reveal a different dimensional structure.

---

**Open questions:**

1. Is the ICAO PANS-TRG framework explicitly documented as safety-event-derived? What is the published methodology — incident data analysis, LOSA, operational safety surveys?

2. Are there published analyses of which competency categories account for most accident causation, vs. which account for most performance variance in normal operations? Is there overlap or divergence between the two?

3. At what point does satisfying the governance layer become a floor rather than a ceiling — i.e., when does de-risking stop being the right objective and expertise optimization become more valuable?

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-cbt... · Revises the normative-artifact critique: categories are safety-event-derived, not arbitrary — but Claim B (suboptimal variance partition) remains open

note-obs... · De-risking target explains why OB categories are structured around dangerous deficits, not full expertise variance

note-ai-...Open detail

agent drafted

Note content

## AI Scenario Generation as Path Optimization Mechanism

**The mechanism:** Path optimization (design problem 2 from note-systemdesign-zwei-fundamentale-probleme) requires: given an individual's current position in competency space, find the optimal next learning step. AI scenario generation is the concrete implementation of that step: measure position → identify the gap → generate a scenario or course segment that specifically targets that gap → remeasure. The loop closes with each iteration.

**Why this is different from current practice:** Current CBTA training uses pre-designed scenarios (standardized sim exercises, fixed course modules). The scenario is designed for an average learner or a specific regulatory checkpoint, not for the individual's current gap profile. The feedback loop is slow (post-exercise debriefs, periodic evaluations) and low-bandwidth (coarse OB grades). AI generation enables on-demand, gap-targeted scenario construction.

**Architecture sketch:**

1. Measurement layer: continuous or frequent assessment of individual position across competency vectors (FDM, simulator data, OB observations)
2. Gap identification: compare measured position to target (governance floor, or expertise target if above compliance floor)
3. Scenario generation: AI constructs a scenario that specifically activates the competency gap — correct difficulty, context, time pressure, interaction complexity
4. Delivery: simulator session, desktop trainer, or other format depending on target competency
5. Remeasure: close the loop

**Governance layer constraint:** The aviation regulatory context requires conformance to the fixed endpoint (ICAO competency framework or equivalent). The path can be highly individualized but must reach the same destination. This is a bounded optimization: fixed endpoint, variable route. AI optimizes the route, not the destination.

**Generalization beyond aviation:** The governance constraint is aviation-specific. In open learning domains (without a hard regulatory conformance endpoint), the destination itself becomes part of the optimization problem. The architecture is the same, but the target is defined differently. Aviation is actually the harder case and therefore the stronger proof of concept: if the loop can close under a hard governance constraint, it will generalize to softer-constraint domains.

**Human-mediated delivery constraint:** Some competency gaps require instructor-mediated feedback — the instructor needs to explain something actionable. AI-generated scenarios help with the scenario side; they don't eliminate the instructor for Stage 4-5 feedback delivery. The two legs (scenario generation, feedback articulation) have different automation prospects.

**Relationship to deliberate practice:** This architecture is essentially a mechanization of deliberate practice at scale. Ericsson's acquisition mechanism (working at performance edge + immediate specific feedback + error correction) maps directly onto the loop: gap-targeted scenario = working at edge; remeasurement = feedback; scenario adjustment = correction. The bottleneck in current training is that this loop is manually operated and can't run at the cadence DP theory requires for rapid advancement.

---

**Open questions:**

1. What scenario generation granularity is actually needed? Can AI generate full simulator scenarios (weather, ATC, traffic, system failures), or only parameterize existing scenario templates?

2. What measurement frequency is required to make the loop useful? Continuous FDM-based positioning, or is periodic (per-session) assessment sufficient?

3. How does the system handle competency interactions — when closing one gap requires simultaneous competency in others? Path optimization in a high-dimensional correlated space is non-trivial.

4. What is the instructor's role in this architecture? If scenario generation is automated, does the instructor become a feedback articulator and calibrator rather than a scenario designer and evaluator?

5. Is there prior art in simulation-based training domains (medical, military) where adaptive scenario generation has been implemented? What worked and what broke?

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-sys... · Concrete implementation mechanism for path optimization (design problem 2)

note-dp-... · AI scenario loop is mechanized deliberate practice at scale — maps directly onto Ericsson's acquisition mechanism

note-cbt... · Path optimization must operate within the governance constraint — fixed endpoint, variable route

note-sm-...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

note-sm-...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

note-sm-...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-sm-... · claim-form titles are what make notes linkable — which is the condition for network value

note-sm-...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-sm-... · concept orientation is the structural condition for the network to form

note-sm-...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-sm-... · provenance is what makes the Room 3 commitment trustworthy — it requires a clean Room 2

note-sm-...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-sm-... · gap at the Room 1 to Room 2 boundary — separation is prescribed but the threshold is not

note-sm-...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-sm-... · specific failure case — expert knowledge has no arrival to point to, breaking the signaling model

note-sm-...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-sm-... · grain calibration is the practical question the title-as-claim condition leaves open

note-sm-...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-sm-... · deliberate practice grounds neurologically why phase separation is cognitively necessary

note-sm-... · the network is the externalized pattern library — the same structure deliberate practice builds internally

note-sm-...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-sm-... · challenges when concept-orientation is achievable — the target assumes clarity that may not yet exist

note-sm-... · both are gaps at the Room 1 to Room 2 boundary — when to promote and when to promote with what concept are related questions

note-sm-...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-sm-... · two actors may concept-orient the same arrival differently — the linking model breaks without reconciliation

note-sm-... · both are sources that don't fit the standard arrival-distillation model — expert knowledge and multi-author divergence are related edge cases

note-sm-...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-sm-... · confidence marking is part of the same epistemic signaling problem — unmarked or uncalibrated confidence degrades the signal

note-sm-... · both are calibration problems — grain and confidence are two dimensions where the protocol provides rules without reference anchors

note-sm-...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-sm-... · edit drift breaks the epistemic status signal — provenance looks intact while the grounded claim has changed

note-sm-...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-sm-... · untyped links degrade the network — traversal cannot distinguish relationship kinds, which limits what the graph can answer

note-sm-...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-sm-... · the network has no way to represent its own clusters as units — synthesis is the missing operation on the product

note-sm-...Open detail

agent drafted

Note content

Confirmed provenance

No confirmed source edge yet.

Link target

Annotation

Split text

Merge source note

Merged text

Outgoing links

note-sm-... · triage decision rule gap compounds the lifecycle gap — without a clear threshold, arrivals that should have been dismissed accumulate alongside genuinely deferred ones

Sensemaking Workspace