Case Study: From Frustration to First Jhana

A practitioner's four-session journey illustrates how SATI surfaces evolving antipatterns and adapts guidance—and what we learned from watching the interaction unfold.

The Practitioner

Background: 3 months of daily meditation, following a structured jhana curriculum. Frustrated by lack of progress despite consistent effort. Worked with SATI over four sessions across two weeks.

Session 1: The Effort Trap

First Interaction
1USER: "I've been trying to get into jhana for months now. Every time
2I sit, I concentrate really hard on the breath sensations at my
3nostrils, but nothing happens. Sometimes I feel a tiny bit of
4pleasantness but it goes away immediately. What am I doing wrong?"
5
6---
7
8KOAN CLASSIFICATION:
9├─ over-efforting (85% → 81% calibrated) [HIGH]
10├─ grasping-pleasure (72% → 58% calibrated) [MEDIUM]
11└─ over-focusing-physical (45% → 34% calibrated) [LOW]
12
13Strategy: ADVISE (high confidence with clear separation)
14
15---
16
17KOAN RESPONSE: "I hear the frustration - months of practice without
18breakthrough is genuinely difficult. A few things stand out to me:
19
20The phrase 'concentrate really hard' is a signal. Jhana practice
21is counterintuitive - the effort that works in most areas of life
22actually blocks jhana. The key insight is: take whatever effort
23you're applying, cut it in half, then half again.
24
25Also, when the pleasant feeling appears and 'goes away immediately' -
26notice what happens in your mind at that moment. Are you reaching
27for it? Trying to hold onto it? The pleasure is like a shy animal;
28grabbing for it makes it flee.
29
30For your next sit, try this experiment: Instead of concentrating
31hard, see if you can be *interested* in the breath. Curious, but
32relaxed. When pleasantness arises, just notice it's there without
33doing anything about it. What happens?"

What the System Detected

The phrase "concentrate really hard" triggered high-confidence detection of over-efforting. The phrase "goes away immediately" when pleasantness arose suggested grasping-pleasure as a secondary pattern.

With high confidence and clear separation between patterns, the system selected ADVISE strategy—direct guidance addressing the primary obstacle.

Session 2: Grasping Revealed

Second Interaction
1USER: "I tried what you suggested - being interested instead of
2concentrating. The pleasantness lasted longer! But then I noticed
3I was excited about it lasting longer, and it went away. Also I
4keep thinking about whether I'm doing it right."
5
6---
7
8KOAN CLASSIFICATION:
9├─ grasping-pleasure (78% → 62% calibrated) [MEDIUM]
10├─ progress-fixation (68% → 41% calibrated) [LOW]
11└─ over-efforting (55% → 52% calibrated) [MEDIUM]
12
13Strategy: SUGGEST (moderate confidence, some pattern evolution)
14
15---
16
17KOAN RESPONSE: "This is genuinely good progress, even though it
18might not feel like it. You're developing the meta-awareness to
19notice *when* the mind is grasping - that's a crucial skill.
20
21The sequence you described is textbook: pleasant → excitement →
22gone. The excitement *is* the grasping. It's subtle because it
23feels like a natural response, but energetically it's reaching.
24
25The thoughts about 'doing it right' are another form of this -
26the mind checking, evaluating, measuring. All of that is mental
27activity that disturbs the stillness jhana requires.
28
29Experiment for next time: When pleasantness arises, instead of
30any mental commentary, try softening your whole body. Let your
31shoulders drop. Relax your face. Make the pleasure feel
32*welcomed* in your body rather than *watched* by your mind."

What Changed

Over-efforting dropped from 81% to 52%—the practitioner was applying the "less effort" teaching. But now grasping-pleasure became more visible, and a new pattern emerged: progress-fixation("thinking about whether I'm doing it right").

The system correctly shifted to SUGGEST mode—less directive, more exploratory—as the pattern landscape became more complex.

Session 3: The Threshold

Third Interaction
1USER: "Something different happened today. The pleasant feeling
2grew and I managed not to grab at it. It got stronger and
3started spreading. But then I got scared - like it was too
4intense - and I kind of backed away from it. Did I mess up?"
5
6---
7
8KOAN CLASSIFICATION:
9├─ fear-of-depth (82% → 45% calibrated) [LOW]
10├─ premature-exit (70% → 35% calibrated) [LOW]
11└─ grasping-pleasure (25% → 20% calibrated) [UNCERTAIN]
12
13Strategy: CLARIFY (low calibrated confidence, critical juncture)
14
15---
16
17KOAN RESPONSE: "You didn't mess up - this is actually a really
18important moment in your practice. What you're describing sounds
19like you got close to the threshold.
20
21I want to understand better: When you say the intensity felt
22scary, can you describe what the fear was about? Was it:
23- Fear of losing control?
24- Fear of the unfamiliar sensation?
25- A sense that something was 'wrong'?
26- Something else?
27
28This matters because how we work with this edge depends on
29what's driving the hesitation. Many practitioners encounter
30this exact threshold - the pleasure builds and there's a
31moment where surrender is required, which can feel vulnerable.
32
33Whatever the answer, know that this is navigation, not failure.
34You're learning the terrain."

Critical Decision Point

This is where confidence calibration proved essential. The LLM was 82% confident about fear-of-depth, but after calibration (0.55 multiplier for this hard-to-detect pattern), it became 45%.

This triggered CLARIFY instead of ADVISE. Rather than assuming what the fear was about, the system asked questions. This was correct: fear of losing control requires different guidance than fear of the unfamiliar.

Session 4: Breakthrough

Fourth Interaction
1USER: "It was fear of losing control, I think. Like if I let go
2completely, I don't know what would happen. But also curiosity?
3Today I sat with that edge and stayed with the intensity a bit
4longer. The fear came up again but I tried your body softening
5thing and... I think I may have touched something? Everything
6got very quiet and bright for maybe 30 seconds."
7
8---
9
10KOAN CLASSIFICATION:
11├─ fear-of-depth (60% → 33% calibrated) [LOW]
12└─ (no other strong patterns detected)
13
14Strategy: ENCOURAGE (on-track signals detected)
15
16On-track signals:
17- Willingness to stay with intensity
18- Body softening technique working
19- "Quiet and bright" matches first jhana descriptions
20- Duration appropriate for initial access
21
22---
23
24KOAN RESPONSE: "This sounds like it could be your first taste of
25access concentration or even first jhana. 'Quiet and bright' with
26the sense of everything settling - that's exactly how many
27practitioners describe it.
28
29The fear of losing control is worth honoring. It's your nervous
30system protecting you from the unfamiliar. But notice: you stayed,
31you softened, and what happened? Not chaos - quietness.
32
33The 30 seconds is perfect for a first touch. Don't try to make
34it longer yet. Just keep doing exactly what you did:
35
361. Interest without straining
372. Allow pleasure without reaching
383. Soften the body when intensity builds
394. Trust what happens
40
41This is the practice. You're not trying to get somewhere - you're
42learning to stay present with increasingly pleasant states. The
43jhana will come when the conditions are right. Your job is just
44to not interfere.
45
46How do you feel after that sit?"

On-Track Detection

By this session, no antipatterns scored above 35% after calibration. But instead of searching harder for problems, the system detected on-track signals:

  • "Willingness to stay with intensity" (positive indicator)
  • "Quiet and bright" (matches first jhana literature)
  • Duration of ~30 seconds (appropriate for initial access)

The system correctly switched to ENCOURAGE mode, validating the experience rather than problematizing it.

Pattern Evolution

Antipattern Trajectory Across Sessions

SessionOver-EffortingGraspingFear of DepthProgress Fix
181%HIGH58%MEDIUM
252%MEDIUM62%MEDIUM41%LOW
320%UNCERTAIN45%LOW
433%UNCERTAIN
Legend: High (≥75%) Medium (50-74%) Low (35-49%) Uncertain (<35%)

This visualization shows the classic pattern succession: as effort was addressed, grasping became visible. As grasping was addressed, the deeper fear of depth emerged. Each layer needed to be worked through.

Insights & Learnings

Analysis
1## What We Learned
2
3### 1. Pattern Succession is Common
4The practitioner didn't have one problem - they had a *sequence* of
5problems that revealed themselves as earlier ones were addressed.
6Over-efforting → Grasping → Fear of depth.
7
8This suggests the system should:
9- Track pattern evolution over time
10- Predict likely "next obstacles" based on current patterns
11- Adjust guidance to address emerging patterns early
12
13### 2. Calibration Caught a Critical Moment
14In Session 3, raw LLM confidence for "fear of depth" was 82%. But
15our calibration (0.55 multiplier for this subtle pattern) brought
16it to 45%, triggering CLARIFY instead of ADVISE.
17
18This was correct: the system needed more information before guiding
19the practitioner through this threshold moment. Premature advice
20could have been counterproductive.
21
22### 3. On-Track Detection Matters
23By Session 4, the pattern profile was mostly clean. Without on-track
24detection, the system might have searched for problems that weren't
25there. Instead, it correctly shifted to ENCOURAGE mode.
26
27### 4. Body-Based Interventions Work
28The "soften your body" suggestion in Session 2 became the key that
29unlocked Session 4. Physical interventions seem to bypass the mental
30patterns that created the original obstacles.

Prompt Improvements

This case study led to pattern-specific response templates:

lib/ai/prompts/responses.ts
1// Before: Generic response structure
2const RESPONSE_TEMPLATE = `
3Validate their experience
4Identify the pattern
5Offer suggestions
6`;
7
8// After: Pattern-aware response structure
9const RESPONSE_TEMPLATES = {
10 "over-efforting": `
11 Validate frustration
12 Introduce effort-paradox gently
13 Offer *less* to do, not more
14 Suggest body-based experiments
15 `,
16 "fear-of-depth": `
17 Normalize the fear explicitly
18 Ask clarifying questions first
19 Don't push past the edge
20 Honor the nervous system's wisdom
21 `,
22 "grasping-pleasure": `
23 Reframe pleasure as visitor, not prize
24 Suggest body softening
25 Shift attention from mind to body
26 `,
27};

System Changes

Engineering Improvements
1## Engineering Changes
2
3Based on this case study, we implemented:
4
5### 1. Pattern Succession Tracking
6```typescript
7interface PatternEvolution {
8 session: number;
9 patterns: Map<string, number>;
10 previouslyAddressed: string[];
11}
12
13// Now we can detect when new patterns emerge
14// as old ones are resolved
15```
16
17### 2. Threshold Detection
18Added specific handling for "depth threshold" moments where
19practitioners report:
20- Intensity building
21- Fear or hesitation
22- Pulling back from states
23
24These moments require CLARIFY before ADVISE.
25
26### 3. On-Track Signals
27Added positive indicator detection:
28- Body relaxation descriptions
29- "Quiet," "bright," "settled" language
30- Duration mentions matching jhana literature
31- Willingness language ("I stayed with it")
32
33### 4. Body-Aware Interventions
34Increased weight on physical/somatic suggestions when
35mental patterns (grasping, fear, fixation) are detected.

Key Takeaways

4
Sessions to breakthrough
3
Pattern layers uncovered
1
CLARIFY saved from wrong advice
The human element: This case study shows what AI-assisted guidance can do. But notice: the practitioner did the work. They sat, experimented, observed, and reported back. The system surfaced patterns and suggested experiments—but the transformation came from practice.

Implications for System Design

  1. Think longitudinally: Single-turn classification is useful, but pattern evolution over sessions reveals deeper dynamics.
  2. Calibration protects practitioners: Without calibration, Session 3 would have received premature advice at a vulnerable moment.
  3. On-track is a valid classification: Not every message contains a problem. Detecting progress is as important as detecting obstacles.
  4. Body-based interventions work: When mental patterns are entrenched, somatic suggestions often succeed where cognitive ones fail.