CIE-Unified

2026-03-31_task02_branch_a_dynamics_feedback.md

  1# Task 02: Branch A Dynamics + Feedback
  2
  3## Title
  4
  5Task 02: minimal dynamics + feedback loop + decay + degraded output
  6
  7## Direct Prompt
  8
  9Continue Branch A from commit `3e972559b9ea28fdad8c8eccd8d206afd669836f`, keep the implementation independent from other branches, strengthen the graph-native runtime beyond scaffold level, make output feedback materially affect later state, make decay observable, make degraded output real and testable, and record execution results in-repo.
 10
 11## Suggested Branch Name
 12
 13`branch-a/task02-dynamics-feedback`
 14
 15## Goal
 16
 17Implement the smallest inspectable Branch A runtime that has real local graph dynamics across `(phi, mu, J)`, real output-to-input feedback, observable forgetting, and runtime-conditioned `full` / `degraded` / `minimal` output behavior.
 18
 19## Background
 20
 21This round must continue to follow the locked conceptual and engineering constraints in:
 22
 23- `/Users/george/code/CIE-Unified/README.md`
 24- `/Users/george/code/CIE-Unified/LOCKED_IMPLEMENTATION_SPEC.md`
 25
 26Those documents remain locked and define the Branch A requirements for graph-native runtime state, output-to-input feedback, homing, decay, degraded output, and comparable observability.
 27
 28## Involved Repo
 29
 30- `/Users/george/code/CIE-Unified`
 31
 32## Scope
 33
 34- update the Branch A plan/task docs for round 2
 35- strengthen `step()` so `phi`, `mu`, and `J` evolve through explicit local rules
 36- make `emit()` queue structured internal feedback that changes later runtime state
 37- make inactivity decay and forgetting observable through retained event traces
 38- make `output_mode` depend on real runtime conditions
 39- improve snapshot observability while keeping the runtime small and stdlib-only
 40- extend tests for dynamics, feedback, decay, and degraded output
 41
 42## Allowed Modifications
 43
 44- `/Users/george/code/CIE-Unified/plans/2026-03-31_branch_a_plan.md`
 45- `/Users/george/code/CIE-Unified/tasks/2026-03-31_task01_branch_a_round1_scaffold.md`
 46- `/Users/george/code/CIE-Unified/tasks/2026-03-31_task02_branch_a_dynamics_feedback.md`
 47- `/Users/george/code/CIE-Unified/cie/__init__.py`
 48- `/Users/george/code/CIE-Unified/cie/graph.py`
 49- `/Users/george/code/CIE-Unified/cie/state.py`
 50- `/Users/george/code/CIE-Unified/cie/runtime.py`
 51- `/Users/george/code/CIE-Unified/tests/__init__.py`
 52- `/Users/george/code/CIE-Unified/tests/test_smoke.py`
 53- `/Users/george/code/CIE-Unified/tests/test_dynamics.py`
 54
 55## Avoid Modifying
 56
 57- `/Users/george/code/CIE-Unified/README.md`
 58- `/Users/george/code/CIE-Unified/LOCKED_IMPLEMENTATION_SPEC.md`
 59
 60## Must Complete
 61
 62- mark Task 01 complete and Task 02 complete in the Branch A plan
 63- create this Task 02 prompt document in-repo
 64- strengthen runtime dynamics across `phi`, `mu`, and `J`
 65- make emit-driven feedback materially affect later steps
 66- make decay/forgetting real and observable
 67- make output degradation real and testable
 68- validate with the recommended unittest command
 69- record execution details and deferred limitations here
 70
 71## Acceptance Criteria
 72
 731. `step()` performs real graph-native state evolution across `phi`, `mu`, and `J`.
 742. Output-to-input feedback materially affects subsequent runtime state.
 753. Decay/forgetting is real and observable.
 764. `output_mode` transitions among `full`, `degraded`, and `minimal` using runtime conditions.
 775. `bound_ability_core`, `drift_score`, and `anchor_pull` are more meaningful than scaffold placeholders.
 786. Tests pass.
 79
 80## Evaluation Requirements
 81
 82- use only the Python standard library in runtime code
 83- keep the implementation minimal, explicit, and inspectable
 84- stay graph-native with `(phi, mu, J)` as the canonical state
 85- avoid `exact_text_map`
 86- avoid MoE-style substitution
 87- avoid latent-vector ontology as the real runtime state
 88- keep observability stronger than hidden heuristics
 89
 90## Recommended Validation Command
 91
 92`python3 -m unittest discover -s tests -v`
 93
 94## Delivery Requirements
 95
 96- commit on `branch-a/task02-dynamics-feedback`
 97- push the branch to `origin`
 98- keep the implementation independent from `branch-b` and other later branches
 99- include execution record details for branch/base/backup/files/validation/results/limitations
100
101## Execution Record
102
103- actual branch name: `branch-a/task02-dynamics-feedback`
104- base commit: `3e972559b9ea28fdad8c8eccd8d206afd669836f`
105- backup path used for dirty-worktree handling: `none`
106- files changed:
107  - `/Users/george/code/CIE-Unified/plans/2026-03-31_branch_a_plan.md`
108  - `/Users/george/code/CIE-Unified/tasks/2026-03-31_task01_branch_a_round1_scaffold.md`
109  - `/Users/george/code/CIE-Unified/tasks/2026-03-31_task02_branch_a_dynamics_feedback.md`
110  - `/Users/george/code/CIE-Unified/cie/graph.py`
111  - `/Users/george/code/CIE-Unified/cie/state.py`
112  - `/Users/george/code/CIE-Unified/cie/runtime.py`
113  - `/Users/george/code/CIE-Unified/tests/test_smoke.py`
114  - `/Users/george/code/CIE-Unified/tests/test_dynamics.py`
115- validation command run: `python3 -m unittest discover -s tests -v`
116- concise test result summary: `Ran 10 tests; all passed.`
117- remaining limitations deferred to Task 03:
118  - sedimentation still uses lightweight stage heuristics rather than a richer path-level memory-to-skill trace
119  - merge events remain node-level promotions, not broader structural merges or loop detection
120  - feedback is now real and stateful but still local; it does not yet model richer long-horizon teacher correction
121
122## Task 03 Continuation Note
123
124Task 03 for Branch A was branched independently from commit `9823c2395c565908965b137e343bcbc1883ae3c5` on `branch-a/task03-sedimentation` to replace those sedimentation placeholders with explicit runtime state and events.