rp-why · Collaboration Depth Model
Measuring Cognitive Complexity
in Human-AI Collaboration
A longitudinal dataset tracking three dimensions of human-AI collaboration across 141 days of instrumented practice. January 28 → June 18, 2026.
Current Data as of June 18, 2026
DOK · Depth of Knowledge
2.50
Full dataset mean (compression-adjusted)
Measures the cognitive complexity of human prompts. Scored 1.0–4.0 using Webb's Depth of Knowledge framework. 'Adjusted' accounts for compression of intent — when short prompts carry complex meaning due to established context.
+41% since baseline
TM · Tool Maturity
Tier 3
Ensemble
Measures intentional orchestration — how deliberately you coordinate AI tools, agents, and workflows. Tracked by monitoring API calls, tool calls, and agents spawned. Only counts actions the user deliberately initiated.
ADT · Agentic Delegation Trust
Frontier
TM and DOK matched, growing
Measures how much the user trusts the agent to execute — the difference between "build a website" and a prompt with specific constraints, design, accessibility, and outcomes. Low ADT = constant "fix this" corrections. High ADT = frontloaded collaboration that reduces rework.
▶Definitions
Key terms and formulas used throughout this analysis
Dimensions
- DOK (Depth of Knowledge)
- Cognitive complexity of human prompts, scored 1.0-4.0 using Webb's framework. Multi-signal classification: keyword patterns (85+), system message filtering, session-context default, and compression detection.
- DOK Adjusted
- DOK score after compression lift. When short prompts (≤8 words, position >2) carry complex intent due to established context, they receive +1 DOK. The adjustment captures trust-based delegation that raw keyword matching would miss.
- TM (Tool Maturity)
- Six-tier Orchestra Model measuring how deliberately the practitioner orchestrates AI tools. Estimated from session characteristics: prompt count, unique tools used, sub-agent presence. Capped at Tier 5 (Symphony) by the instrument; Tier 6 (Virtuoso) requires all dimensions at optimal alignment.
- ADT (Agentic Delegation Trust)
- Diagnostic zone derived from the TM × DOK matrix. Not a score but a classification: where does the practitioner's cognitive depth sit relative to their tool sophistication? Six zones from Overpowered (mismatch) to Frontier (matched and growing).
- Compression
- The percentage of prompts that are short directives (≤8 words) carrying complex intent. A signal of trust — when the practitioner says "proceed" and the agent executes a multi-step strategic workflow, that's compression. It emerges only after shared context is established.
Token Economics
- Value
DOK × (1 + DOK3+4%) × (1 + Compression%)
Multiplicative composite capturing depth, strategic proportion, and trust signal. Higher = more cognitively aligned work per session.- Token ROI
(Value_now - Value_baseline) / Cumulative_tokens
Return on token investment. Positive = tokens built durable capacity. Zero = maintenance. Negative = expenditure without growth.- Strategic Token Ratio (STR)
- Fraction of prompts at DOK 3+. "What percentage of token spend went to strategic work?" Simpler than Value but doesn't account for compression or cost normalization.
- Impact Score (IS)
(DOK × DOK3+4_factor × Comp_factor) / log(tok/prompt)
Value normalized by log of cost. Rewards depth, breadth, and trust without over-penalizing expensive sessions. The primary per-session efficiency metric.- Tokens as Capital
- The framing that distinguishes this analysis from cost-minimization. Tokens spent on DOK 1 recall are purchases (consumed, done). Tokens spent on DOK 3+ strategic work are investments (build durable capacity, compound over time, raise the practitioner's floor permanently).
- Three Returns
- Depth — DOK floor rises. Efficiency — compression emerges (trust replaces verbosity). Leverage — DOK 3+4% grows (the mix shifts from execution to design). All three compound.
How to Read This Data
The three dimensions interact to produce diagnostic zones. Understanding the framework is essential before interpreting the charts below.
DOK Levels
Scored 1.0–4.0 using Webb's framework. Analyzes language complexity across the full conversation context, not individual words.
- ● Level 1: Recall & Reproduction
- ● Level 2: Application of Skills & Concepts
- ● Level 3: Strategic Thinking
- ● Level 4: Extended Thinking
Orchestra Tiers (TM)
Six tiers of human-AI collaboration, from Solo (human works alone, AI reviews) to Virtuoso (human and AI synthesized in flow).
- ● Tier 1: Solo
- ● Tier 2: Duet
- ● Tier 3: Ensemble
- ● Tier 4: Chamber
- ● Tier 5: Symphony
- ● Tier 6: Virtuoso
ADT Diagnostic Zones
Measures the gap between Tool Complexity and Human Cognitive Depth. The TM × DOK matrix produces zones that diagnose the health of the collaboration practice.
- ● Expected
- ● Growing
- ● Frontier
- ● Thinking Ahead
- ● Underutilizing
- ● Overpowered
Orchestra Model — 6 Tiers of Collaboration
Human works alone. AI reviews after.
Back-and-forth conversation. Human prompts, AI responds, human edits.
Human provides meaningful body of work. Evaluates holistically.
Human delegates work streams. Sub-agents introduced. Orchestration required.
Multiple AI interactions coordinated toward unified goal. Minimal intervention.
Flow state. Human and AI synthesized. Optimal DOK, ADT, and TM.
Trajectory
DOK Adjusted and Tool Maturity over time. The primary growth signal.
Peak DOK: 3.70 — 2026-06-15, Jun 15: Pure strategic thinking. Extended thinking dominant — framework creation, cross-disciplinary synthesis, the day DOK 4 became reproducible.
Token Economics
Tokens as capital, not cost. Measuring return on token investment — what did the spend build?
Diagnostic Zone Map
Where each data point falls on the TM × DOK grid. The six zones reveal the relationship between tool sophistication and cognitive depth.
TM and DOK are matched and growing together. The collaboration is operating at its productive edge — tools are being used at the depth they were designed for.
Approaching a match between tool sophistication and cognitive depth. The practitioner is building toward more effective use of their current toolset.
Tool usage and cognitive depth are appropriate for the current level. A healthy starting position, but sustained time here may indicate a plateau.
Cognitive depth exceeds tool sophistication. The practitioner is thinking at a higher level than their tools support — an opportunity to adopt more powerful orchestration patterns.
Tool sophistication exceeds cognitive depth. Powerful tools are being used for simple tasks — like using an LLM to ask 'what time is it.' An opportunity to deepen the questions being asked.
Significant mismatch between tool complexity and task depth. Resources are being spent without proportional cognitive return.
DOK Distribution
How cognitive complexity distributes across sessions. The shift from DOK 1-2 dominance to DOK 3-4 dominance is the growth story.
Copying, recalling, or reproducing information. Simple factual prompts like 'what is X' or 'show me the syntax for Y.'
Applying learned skills to solve a specific problem. Using tools with purpose — 'build this component' or 'fix this error using pattern X.'
Reasoning across multiple concepts to plan, analyze, or design. Prompts that require weighing tradeoffs, sequencing work, or making architectural decisions.
Creating something entirely new — frameworks, cross-disciplinary synthesis, novel artifacts. Sustained sessions (2-4 hours) that produce original thinking documented within the LLM.
Where Growth Happens
A single mean DOK obscures growth. Phase deltas reveal where depth, compression, and strategic thinking actually shift.
Four Phases
Learning the tools. Every interaction is explicit.
Framework creation burst. Discovery peaks.
Execution-heavy. High-volume delegation.
Compression emerges. Strategic work returns.
Phase 3 → 4: The Shift
The transition from Plateau to Deepening is where the growth signal lives. Not a return to Phase 2 discovery peaks - a fundamentally different kind of depth.
Full Arc: Phase 1 → 4
The thesis: Growth is not linear DOK increase. Phase 3 drops because execution absorbs attention - the cognitive complexity moves from the prompt surface into the orchestration structure. Phase 4 recovers with a different character: compression triples, strategic thinking doubles, and the floor holds. The tool didn't get smarter. The relationship deepened.
Open Questions
Active research threads. Where the framework needs to go next.
How do we measure outcomes and impact from token economics data?
The current Value formula (DOK × strategy × trust) measures process quality — how cognitively aligned the work is. It does not measure outcomes— whether that work shipped, was adopted, or produced durable artifacts. What external signal (PRs merged, tickets closed, self-reported impact, artifact durability) would ground "value per token" in something beyond the practitioner's own dimensions?
Is there a saturation point where Token ROI approaches zero?
Marginal returns are declining (each additional billion tokens produces less incremental growth). Does ROI asymptote — indicating the practitioner has "graduated" and further tokens are maintenance — or does it plateau and then re-accelerate when a new capability tier unlocks?
Does the DOK 3.0+ efficiency inversion hold at scale?
Sessions at DOK 3.0+ are both the cheapest per prompt (180K tok/prompt) and the highest value (IS = 0.58). But N=16. Is this a real threshold effect — once you ask the right question, you don't need many tokens to get the answer — or a small-sample artifact?
What happens when TM variance increases?
88% of sessions are Tier 2 (Duet). The TM dimension is effectively flat. When sub-agents, multi-tool orchestration, and Symphony-level workflows become routine, does the DOK × TM × token relationship change fundamentally? Does higher TM reduce or amplify token spend for equivalent cognitive output?
Can this framework transfer across practitioners?
This is N=1 longitudinal data. The Three Dimensions model claims to be generalizable, but the Token ROI curve, phase transitions, and compression emergence patterns may be idiosyncratic. What would a multi-practitioner study reveal about universal vs. individual growth trajectories?
Try rp-why
Measure your own AI collaboration practice. The rp-why Goose skill runs locally against your session history and produces the same Three Dimensions analysis shown above.
Install
npx skills add https://github.com/block/agent-skills --skill rp-whyRequires the built-in skills extension enabled in Goose.
Commands
/rp-why initGenerate your personal baseline from conversation history. Takes ~30 seconds.
/rp-why currentAssess your current session. See DOK distribution, TM tier, and ADT zone in real time.
/rp-why compareCompare today against your baseline. Track growth over time.
/rp-why overallFull Three Dimensions dashboard with timeline and trajectory.
Workflow
Run /rp-why init to create your baseline.
End sessions with /rp-why current. 30 seconds to see patterns.
Run /rp-why compare to see trajectory and celebrate growth.
Before sending a prompt, ask: "Can I make this more strategic?"
Prompt Upgrades
The single highest-leverage habit: reframe prompts one DOK level higher before sending.
"What is a microservice?"
"How would I decide between microservices and a monolith for my project?"
"How do I set up CI/CD?"
"Design a CI/CD strategy that balances speed, reliability, and team workflow for a 5-person team."
"Design a caching strategy."
"Over the next few sessions, help me research, prototype, and document a caching architecture. Start by analyzing our current bottlenecks."
Mapping to the Engineering IC Ladder
DOK levels correlate with competencies on a typical engineering IC ladder. Higher DOK doesn't mean "harder" - it means access to higher-durability output.
Applying known patterns, writing tests, single-file changes. Necessary but doesn't compound. Output durability: days to weeks.
Maps to "Solve What Matters" and "Architect Systems" on the ladder. Designing systems, analyzing trade-offs, reasoning across multiple files and concerns. Output durability: months.
Maps to "Steward Systems" and cross-team influence competencies. Creating transferable knowledge, frameworks, and protocols that compound. Output durability: years.
The signal:If you're aiming to demonstrate L3/L4 competencies through AI collaboration, the indicator is spending meaningful time at DOK 3 - using the tool for design decisions and trade-off analysis, not just code generation.
The Key Insight
The three dimensions interact multiplicatively. Any one being low caps the others. High DOK + High TM + High Trust = maximum impact per token. High DOK + Low TM = frustration (thinking strategically but the tool can't deliver). Low DOK + High TM = flamethrower lighting a candle (powerful tools wasted on simple tasks). The goal is to get all three dimensions growing together, which puts you in the Frontier zone.