autonomous-exploration-tribes.md
1 # Autonomous Exploration Tribes 2 3 *proto-015 | Run threads forward until uncertainty stops you* 4 5 --- 6 7 - **principle** 8 - "Self-directed teams that explore without constant supervision." 9 - "Moses pattern - escalate edge cases." 10 11 - **shape** 12 - Tribes explore branches you don't take 13 - Run forward until confidence drops below threshold 14 - Report back: progress made, where stuck, options for human 15 - [[first-officer-protocol|First Officer]] spawns and collects tribes 16 17 --- 18 19 **Status:** 📄 DOCUMENTED → Ready to implement 20 21 --- 22 23 ## Core Concept 24 25 > **Tribes don't just verify - they explore. Run a thread until uncertainty exceeds threshold, then report back.** 26 27 You drive the main conversation. Branches you didn't take get assigned to tribes (see [[tribe-sizing-algorithm]]). Each tribe pushes forward autonomously until they can't. 28 29 --- 30 31 ## The Structure 32 33 ``` 34 YOU 35 (Main thread) 36 │ 37 "I'm going this direction" 38 │ 39 ┌────────────────┼────────────────┐ 40 │ │ │ 41 │ YOUR PATH │ 42 │ (fertile) │ 43 │ │ │ 44 Branch A Branch B Branch C 45 (not taken) (not taken) (not taken) 46 │ │ │ 47 ▼ ▼ ▼ 48 ┌─────────┐ ┌─────────┐ ┌─────────┐ 49 │ TRIBE A │ │ TRIBE B │ │ TRIBE C │ 50 │ │ │ │ │ │ 51 │ Explore │ │ Explore │ │ Explore │ 52 │ until │ │ until │ │ until │ 53 │ stuck │ │ stuck │ │ stuck │ 54 └────┬────┘ └────┬────┘ └────┬────┘ 55 │ │ │ 56 ▼ ▼ ▼ 57 [REPORT] [REPORT] [REPORT] 58 59 "Got to X, "Got to Y, "Hit wall 60 confidence found Z, at step 3, 61 still high" need input" need you" 62 ``` 63 64 --- 65 66 ## Exploration Protocol 67 68 ### 1. Spawn Exploration Tribe 69 70 ```markdown 71 ## Exploration Assignment 72 73 **Thread:** `2026-01-15/three-repo-architecture` 74 **Starting Context:** [Fork export or summary] 75 **Direction:** "Explore the two-repo vs three-repo question" 76 **Uncertainty Threshold:** 0.6 (stop when confidence drops below) 77 **Max Depth:** 10 exchanges (prevent runaway) 78 79 **Autonomy Level:** 80 - Can make local decisions (confidence > 0.8) 81 - Should flag medium decisions (confidence 0.5-0.8) 82 - Must stop on low confidence (< 0.5) 83 ``` 84 85 ### 2. Tribe Explores Autonomously 86 87 The tribe (Sonnet workers + Opus FO) runs the thread: 88 89 ``` 90 EXCHANGE 1: [Tribe asks question, answers it] → Confidence: 0.9 ✓ 91 EXCHANGE 2: [Follows up, makes progress] → Confidence: 0.85 ✓ 92 EXCHANGE 3: [Hits a choice point, picks one] → Confidence: 0.7 ✓ 93 EXCHANGE 4: [Explores further] → Confidence: 0.65 ✓ 94 EXCHANGE 5: [Uncertainty rising] → Confidence: 0.55 ⚠️ 95 EXCHANGE 6: [Can't proceed confidently] → Confidence: 0.4 STOP 96 ``` 97 98 ### 3. Report Back 99 100 ```markdown 101 ## Exploration Report: three-repo-architecture 102 103 **Tribe:** A 104 **Exchanges:** 6 105 **Final Confidence:** 0.4 (stopped - below threshold) 106 107 ### Progress Made 108 1. Clarified that three-repo makes sense for [reason] 109 2. Identified permeability contract format 110 3. Drafted initial boundary definitions 111 112 ### Artifacts Created 113 - `drafts/three-repo-proposal.md` 114 - `drafts/permeability-v1.md` 115 116 ### Where We Stopped 117 **Decision point:** Whether Cerf-Meta should be its own repo or a branch 118 **Why stuck:** This is an axiom-level choice about identity boundaries 119 **Confidence:** 0.4 - need human input 120 121 ### Options for You 122 - Option A: Keep Cerf-Meta separate (personal seed stays distinct) 123 - Option B: Merge into OS (simplify, lose personal layer) 124 - Option C: Branch model (same repo, different branches) 125 126 ### Recommended Next Step 127 Human decision on Option A/B/C, then tribe can continue 128 ``` 129 130 --- 131 132 ## Confidence Tracking 133 134 Each tribe tracks confidence per exchange: 135 136 ``` 137 CONFIDENCE FACTORS: 138 139 Technical Certainty (0-1) 140 ├── Is the implementation path clear? 141 ├── Are there established patterns to follow? 142 └── Do we have the information needed? 143 144 Alignment Certainty (0-1) 145 ├── Does this align with axioms? 146 ├── Is this consistent with existing architecture? 147 └── Would the human likely approve? 148 149 Scope Certainty (0-1) 150 ├── Are we staying within assigned scope? 151 ├── Is this getting bigger than expected? 152 └── Are we still on the original thread? 153 154 AGGREGATE: C = (Tech × 0.4) + (Align × 0.4) + (Scope × 0.2) 155 ``` 156 157 ### Threshold Actions 158 159 | Confidence | Action | 160 |------------|--------| 161 | C > 0.8 | Continue autonomously | 162 | 0.6 < C ≤ 0.8 | Continue but flag for summary | 163 | 0.4 < C ≤ 0.6 | Pause, consider stopping | 164 | C ≤ 0.4 | Stop, report back | 165 166 --- 167 168 ## Integration with Thread Registry 169 170 When exploration tribe spawns: 171 172 ```markdown 173 ## Thread Registry Update 174 175 | Thread | State | Tribe | Progress | 176 |--------|-------|-------|----------| 177 | `alignment-protocol` | RUNNING | You | Active | 178 | `three-repo-architecture` | EXPLORING | Tribe A | 6 exchanges, C=0.4 | 179 | `permeability-contracts` | EXPLORING | Tribe B | 3 exchanges, C=0.7 | 180 | `ai-judges` | QUEUED | - | Waiting for tribe | 181 ``` 182 183 --- 184 185 ## Parallel Exploration 186 187 Multiple tribes can explore different branches simultaneously: 188 189 ``` 190 YOU: Main thread (alignment-protocol) 191 │ 192 ├── Tribe A: three-repo-architecture (exploring) 193 ├── Tribe B: permeability-contracts (exploring) 194 ├── Tribe C: ambient-hardware (exploring) 195 └── Tribe D: f-calculation-code (exploring) 196 197 META-FIRST OFFICER watches all tribes: 198 ├── Detects cross-tribe convergence 199 ├── Flags when tribes need to sync 200 ├── Reports aggregate progress 201 └── Escalates when multiple tribes stuck on same issue 202 ``` 203 204 --- 205 206 ## Stopping Conditions 207 208 Tribe MUST stop when: 209 210 | Condition | Why | 211 |-----------|-----| 212 | Confidence < threshold | Can't proceed reliably | 213 | Max depth reached | Prevent runaway | 214 | Axiom-level decision needed | Human must choose | 215 | Cross-thread dependency | Need to sync with another tribe | 216 | Scope creep detected | Getting off track | 217 218 Tribe SHOULD stop when: 219 220 | Condition | Why | 221 |-----------|-----| 222 | Major architectural choice | Even if confident, human should see | 223 | Irreversible decision | Hard to undo | 224 | Resource-intensive next step | Check before spending | 225 226 --- 227 228 ## Implementation 229 230 ### Spawn Command 231 232 ``` 233 "Spawn exploration tribe on `three-repo-architecture`" 234 235 Parameters: 236 - thread_id: The thread to explore 237 - context: Fork export or starting point 238 - direction: What to explore (optional - can be open-ended) 239 - threshold: Confidence floor (default 0.5) 240 - max_depth: Exchange limit (default 10) 241 - model: Sonnet for workers, Opus for FO 242 ``` 243 244 ### Tribe Composition 245 246 ``` 247 EXPLORATION TRIBE (Standard) 248 ├── 1 Opus First Officer (metacognition, confidence tracking) 249 ├── 2-3 Sonnet Workers (actual exploration) 250 └── Optional: Haiku for fact-checking along the way 251 ``` 252 253 ### Output Artifacts 254 255 Each tribe produces: 256 1. **Exploration log** - Exchange-by-exchange record 257 2. **Artifacts created** - Any files/drafts produced 258 3. **Decision points** - Where choices were made 259 4. **Stopping report** - Why stopped, what's needed 260 5. **Continuation instructions** - How to resume 261 262 --- 263 264 ## Your Role 265 266 ### Before Exploration 267 - Assign threads to tribes 268 - Set direction (or leave open) 269 - Set thresholds based on problem weight 270 271 ### During Exploration: Micro-Alignments 272 273 > **Key insight: Small touch-points keep the coupled oscillator in resonance.** 274 275 Not just "tribe stuck, escalate to human" - but **micro-alignments**: quick decisions that: 276 1. Unblock the tribe (practical value) 277 2. Keep YOUR model synchronized (cognitive value) 278 3. Maintain human-system resonance (relational value) 279 280 ``` 281 MICRO-ALIGNMENT PATTERN 282 283 Tribe hits small uncertainty: 284 ├── "Should we go left or right here?" 285 ├── Confidence: 0.6 (not stuck, but uncertain) 286 ├── Quick human input would help 287 └── LOW COST to you, HIGH VALUE to tribe 288 289 YOU: [5-second decision] "Go left" 290 291 Tribe continues with high confidence 292 Your model: Updated, synchronized 293 Coupling: Maintained 294 ``` 295 296 **Micro-alignment triggers:** 297 | Situation | Cost to You | Value | 298 |-----------|-------------|-------| 299 | Binary choice, both valid | 5 seconds | Unblocks + model sync | 300 | Preference question | 10 seconds | Alignment + model sync | 301 | Sanity check | 15 seconds | Catches drift early | 302 | Quick prioritization | 30 seconds | Direction + model sync | 303 304 **Why this matters (the sounding principle):** 305 - Your mental model must track what's happening 306 - Too much autonomy → you and system drift apart 307 - Micro-alignments are the "pings" that keep oscillators coupled 308 - Each touch-point reinforces your internal model 309 - The cost is small, the synchronization value is high 310 311 ### After Exploration 312 - Review reports 313 - Make decisions where tribes stopped 314 - Direct: Continue, merge, abandon, or pivot 315 316 ### The Sounding Rhythm 317 318 ``` 319 FULL AUTONOMY CONSTANT INVOLVEMENT 320 ───────────── ──────────────────── 321 System drifts You bottleneck 322 Your model atrophies No scaling 323 Lose resonance Defeats purpose 324 325 ← MICRO-ALIGNMENTS → 326 327 Regular small touches 328 Keep coupling tight 329 Model stays fresh 330 System stays aligned 331 ``` 332 333 --- 334 335 ## Example Flow 336 337 ``` 338 1. YOU: "I'm focusing on alignment-protocol. 339 Spawn tribes to explore three-repo, permeability, and ai-judges." 340 341 2. SYSTEM: 342 - Creates 3 exploration tribes 343 - Assigns fork exports as context 344 - Sets default thresholds 345 - Tribes begin exploring 346 347 3. [TIME PASSES - tribes work autonomously] 348 349 4. REPORTS ARRIVE: 350 - Tribe A (three-repo): Stopped at C=0.4, needs decision on repo structure 351 - Tribe B (permeability): Still running, C=0.7, making progress 352 - Tribe C (ai-judges): Stopped at C=0.5, found dependency on trust-tracking 353 354 5. YOU: 355 - Review Tribe A report, make decision 356 - Let Tribe B continue 357 - Review Tribe C, see dependency, decide to sync 358 359 6. CONTINUE: 360 - Tribe A resumes with your decision 361 - Tribe B keeps going 362 - Tribe C waits for sync or gets new direction 363 ``` 364 365 --- 366 367 ## The Promise 368 369 > **Branches don't die - they get explored.** 370 > 371 > You drive the main thread. 372 > Tribes explore the branches. 373 > Each runs until uncertainty stops them. 374 > They report what they found. 375 > You decide where to go next. 376 > 377 > Parallel progress on parallel fronts. 378 > Nothing is lost. Everything is explored. 379 380 --- 381 382 ## Related 383 384 - **axioms** 385 - [[A0 Boundary Operation]] - each tribe is a bounded exploration unit 386 - [[A1 Telos of Integration]] - tribes integrate their findings back to main thread 387 - [[A3 Dynamic Pole Navigation]] - explore until confidence threshold (dynamic stopping) 388 - **protocols** 389 - [[first-officer-protocol]] - FO spawns and manages tribes 390 - shape:: "Per-thread metacognition. Compress state, track gravity wells." 391 - [[tribe-sizing-algorithm]] - how big should each tribe be 392 - shape:: "Match tribe size to problem characteristics, not fixed." 393 - [[fractal-tribe-architecture]] - tribes are one level in the fractal 394 - shape:: "Same pattern at every level." 395 - [[model-allocation-strategy]] - Sonnet workers, Opus FO 396 - shape:: "Match model capability to task complexity." 397 - [[execution-autonomy-gradient]] - tribes use SHIP/FLAG/ESCALATE 398 - shape:: "Act autonomously when confident, escalate when uncertain." 399 - **enables** 400 - [[mission-control-protocol]] - MC synthesizes across tribe outputs 401 402 --- 403 404 *proto-015 | Autonomous Exploration Tribes | Run Until Stuck*