Fix

backend/Odin.py

@@ -1,13 +1,8 @@
-"""Odin — the web server and process entry point for Valhalla.
-
-Serves the React dashboard, exposes REST + WebSocket endpoints (sim
-control, roster, auth, pathfinding), and hosts the WorldEngine as a
-background asyncio task with per-tick snapshot broadcasts.
-
-Architecture: the only entry point that runs the full system; depends on
-src.core.world_engine, src.auth, pathfinder, and the frontend build.
-Design: all sim-control endpoints are auth-gated; roster edits are only
-allowed while the simulation is stopped.
+"""
+FastAPI web server for the Valhalla agent map.
+Serves the frontend (React SPA), exposes REST + WebSocket
+endpoints for pathfinding (/api/path, /api/path/stream, /ws), and
+streams simulation state via /ws/sim for live agent visualization.
 """
 
 import os
@@ -235,30 +230,6 @@ async def require_admin(authorization: str = Header(None)):
     return user
 
 
-# ---------------------------------------------------------------------------
-# Admin log relay — live view of backend logs without the Space console.
-# The relay mirror lives in src/core/log_relay.py and is installed by
-# src/core/log.py setup_logging(); only these two endpoints expose it.
-# ---------------------------------------------------------------------------
-
-@app.get("/api/logs")
-async def get_log_lines(
-    since: int = Query(default=0, ge=0),
-    _user: dict = Depends(require_admin),
-):
-    """Return relayed log lines newer than `since`; `next` is the poll cursor."""
-    from src.core.log_relay import relay_lines
-    return relay_lines(since)
-
-
-@app.post("/api/logs/clear")
-async def clear_log_lines(_user: dict = Depends(require_admin)):
-    """Clear the in-memory relay buffer. The live log file is left intact."""
-    from src.core.log_relay import clear_relay
-    clear_relay()
-    return {"ok": True}
-
-
 def _print_agent_plans(engine):
     """Print each agent's full action plan to the CLI."""
     from src.core.agent_registry import AgentRuntimeState
@@ -717,8 +688,6 @@ async def add_agent(request: AddAgentInput, _user: dict = Depends(require_admin)
             "current_time": f"{current_date} {current_hhmm}", "places": None,
             "persona_name": generated.name, "mode": "remaining" if engine.world.tick else "full_day",
             "current_location_id": generated.hostel, "upcoming_events": [],
-            "energy_level": engine._energy_baseline(persona),
-            "emotion_state": engine._emotion_baseline(persona),
         }))
         day_plan = plan_result.get("day_plan", [])
         if not day_plan:
@@ -732,7 +701,6 @@ async def add_agent(request: AddAgentInput, _user: dict = Depends(require_admin)
     engine.registry.register(AgentRuntimeState(
         agent_id=agent_id, persona=persona, persona_name=generated.name,
         manager=manager, position=position, day_plan=day_plan,
-        energy_level=engine._energy_baseline(persona),
         emotion_state=engine._emotion_baseline(persona), emotion_baseline=engine._emotion_baseline(persona),
     ))
     engine.world.register_agent(agent_id, position)

backend/Pathfinder_test.py

@@ -1,11 +1,7 @@
-"""Pathfinder_test — interactive CLI tool for verifying pixel pathfinding.
-
-Loads the campus map, computes a BFS path between two pixel coordinates,
-and renders it in a matplotlib window.
-
-Architecture: a developer tool, not part of the simulation runtime; it
-exercises backend/pathfinder.py against the real walkability map.
-Design: keeps the visual debugging loop out of the server code.
+"""
+CLI tool for pixel-level pathfinding on the Valhalla map.
+Loads map.png, computes the shortest path between two pixel coordinates
+via BFS, and displays the result with start/end markers in a matplotlib window.
 """
 
 import sys

backend/data/environment/relationship_matrix.json

@@ -1,6 +1,94 @@
 {
   "schema_version": 2,
   "relationships": {
+    "amitabh->ansh_batra": {
+      "score": 0.55,
+      "tags": [
+        "campus-acquaintance",
+        "low-pressure"
+      ],
+      "context": "Amitabh and Ansh Batra know each other through campus routines. Amitabh appreciates ansh_batra's enthusiastic plans, but the friendship is still finding its rhythm."
+    },
+    "amitabh->anubhav_prasad": {
+      "score": 0.49,
+      "tags": [
+        "campus-acquaintance",
+        "low-pressure"
+      ],
+      "context": "Amitabh and Anubhav Prasad know each other through campus routines. Amitabh appreciates anubhav_prasad's calm listening, but the friendship is still finding its rhythm."
+    },
+    "amitabh->ghanisht_kaushal": {
+      "score": 0.46,
+      "tags": [
+        "campus-acquaintance",
+        "low-pressure"
+      ],
+      "context": "Amitabh and Ghanisht Kaushal know each other through campus routines. Amitabh appreciates ghanisht_kaushal's reliable follow-through, but the friendship is still finding its rhythm."
+    },
+    "amitabh->gurnoor_singh": {
+      "score": 0.56,
+      "tags": [
+        "campus-acquaintance",
+        "low-pressure"
+      ],
+      "context": "Amitabh and Gurnoor Singh know each other through campus routines. Amitabh appreciates gurnoor_singh's big social energy, but the friendship is still finding its rhythm."
+    },
+    "amitabh->jarvis": {
+      "score": 0.63,
+      "tags": [
+        "campus-acquaintance",
+        "low-pressure"
+      ],
+      "context": "Amitabh and Jarvis know each other through campus routines. Amitabh appreciates jarvis's steady conversation, but the friendship is still finding its rhythm."
+    },
+    "amitabh->lavanya_sharma": {
+      "score": 0.5700000000000001,
+      "tags": [
+        "campus-acquaintance",
+        "low-pressure"
+      ],
+      "context": "Amitabh and Lavanya Sharma know each other through campus routines. Amitabh appreciates lavanya_sharma's direct feedback, but the friendship is still finding its rhythm."
+    },
+    "amitabh->parv_singla": {
+      "score": 0.54,
+      "tags": [
+        "campus-acquaintance",
+        "low-pressure"
+      ],
+      "context": "Amitabh and Parv Singla know each other through campus routines. Amitabh appreciates parv_singla's impulsive invitations, but the friendship is still finding its rhythm."
+    },
+    "amitabh->riya_murarka": {
+      "score": 0.3,
+      "tags": [
+        "campus-acquaintance",
+        "low-pressure"
+      ],
+      "context": "Amitabh and Riya Murarka know each other through campus routines. Amitabh appreciates riya_murarka's clear boundaries, but the friendship is still finding its rhythm."
+    },
+    "amitabh->saksham": {
+      "score": 0.53,
+      "tags": [
+        "campus-acquaintance",
+        "low-pressure"
+      ],
+      "context": "Amitabh and Saksham know each other through campus routines. Amitabh appreciates saksham's dry humour, but the friendship is still finding its rhythm."
+    },
+    "amitabh->tanishq": {
+      "score": 0.31,
+      "tags": [
+        "campus-acquaintance",
+        "low-pressure"
+      ],
+      "context": "Amitabh and Tanishq know each other through campus routines. Amitabh appreciates tanishq's quiet, improving confidence, but the friendship is still finding its rhythm."
+    },
+    "ansh_batra->amitabh": {
+      "score": 0.55,
+      "tags": [
+        "campus-acquaintance",
+        "low-pressure"
+      ],
+      "context": "Ansh Batra and Amitabh know each other through football and visual storytelling. Ansh Batra appreciates amitabh's steady conversation, but the friendship is still finding its rhythm."
+    },
     "ansh_batra->anubhav_prasad": {
       "score": 0.41,
       "tags": [
@@ -23,7 +111,14 @@
         "campus-acquaintance",
         "party-bros"
       ],
-      "context": "Ansh and Gurnoor's parties always end in legendary stories — including that one time they both woke up in the same bed after a dare and just laughed it off... mostly."
+      "context": "Ansh and Gurnoor's parties always end in legendary stories \u2014 including that one time they both woke up in the same bed after a dare and just laughed it off... mostly."
+    },
+    "ansh_batra->jarvis": {
+      "score": 0.32,
+      "tags": [
+        "new-acquaintance"
+      ],
+      "context": "Ansh Batra has only recently met Jarvis; the connection is open but untested."
     },
     "ansh_batra->lavanya_sharma": {
       "score": 0.44,
@@ -65,6 +160,14 @@
       ],
       "context": "Ansh loves hyping up Tanishq's growing confidence, especially when Tanishq blushes at compliments. It's dangerously cute."
     },
+    "anubhav_prasad->amitabh": {
+      "score": 0.49,
+      "tags": [
+        "campus-acquaintance",
+        "low-pressure"
+      ],
+      "context": "Anubhav Prasad and Amitabh know each other through co-op games and late-night chai. Anubhav Prasad appreciates amitabh's steady conversation, but the friendship is still finding its rhythm."
+    },
     "anubhav_prasad->ansh_batra": {
       "score": 0.41,
       "tags": [
@@ -89,6 +192,13 @@
       ],
       "context": "Gurnoor drags Anubhav to parties and Anubhav somehow ends up being the responsible one... until that one time he wasn't."
     },
+    "anubhav_prasad->jarvis": {
+      "score": 0.32,
+      "tags": [
+        "new-acquaintance"
+      ],
+      "context": "Anubhav Prasad has only recently met Jarvis; the connection is open but untested."
+    },
     "anubhav_prasad->lavanya_sharma": {
       "score": 0.73,
       "tags": [
@@ -129,6 +239,14 @@
       ],
       "context": "Anubhav is quietly supportive of Tanishq's confidence journey. Their interactions are soft and full of unspoken understanding."
     },
+    "ghanisht_kaushal->amitabh": {
+      "score": 0.46,
+      "tags": [
+        "campus-acquaintance",
+        "low-pressure"
+      ],
+      "context": "Ghanisht Kaushal and Amitabh know each other through badminton and blunt movie opinions. Ghanisht Kaushal appreciates amitabh's steady conversation, but the friendship is still finding its rhythm."
+    },
     "ghanisht_kaushal->ansh_batra": {
       "score": 0.66,
       "tags": [
@@ -153,6 +271,13 @@
       ],
       "context": "Gurnoor's nonstop social battery clashes with Ghanisht's chill, but the rare nights they sync are chaotic gold."
     },
+    "ghanisht_kaushal->jarvis": {
+      "score": 0.32,
+      "tags": [
+        "new-acquaintance"
+      ],
+      "context": "Ghanisht Kaushal has only recently met Jarvis; the connection is open but untested."
+    },
     "ghanisht_kaushal->lavanya_sharma": {
       "score": 0.36,
       "tags": [
@@ -193,6 +318,14 @@
       ],
       "context": "Ghanisht quietly roots for Tanishq's confidence glow-up and enjoys watching him get bolder."
     },
+    "gurnoor_singh->amitabh": {
+      "score": 0.56,
+      "tags": [
+        "campus-acquaintance",
+        "low-pressure"
+      ],
+      "context": "Gurnoor Singh and Amitabh know each other through photography and road-trip playlists. Gurnoor Singh appreciates amitabh's steady conversation, but the friendship is still finding its rhythm."
+    },
     "gurnoor_singh->ansh_batra": {
       "score": 0.53,
       "tags": [
@@ -217,6 +350,13 @@
       ],
       "context": "Gurnoor respects Ghanisht's reliability but wishes he'd loosen up more... preferably with him."
     },
+    "gurnoor_singh->jarvis": {
+      "score": 0.32,
+      "tags": [
+        "new-acquaintance"
+      ],
+      "context": "Gurnoor Singh has only recently met Jarvis; the connection is open but untested."
+    },
     "gurnoor_singh->lavanya_sharma": {
       "score": 0.5,
       "tags": [
@@ -257,6 +397,85 @@
       ],
       "context": "Gurnoor loves seeing Tanishq come out of his shell and occasionally flirts just to see him blush."
     },
+    "jarvis->amitabh": {
+      "score": 0.63,
+      "tags": [
+        "campus-acquaintance",
+        "low-pressure"
+      ],
+      "context": "Jarvis and Amitabh know each other through campus routines. Jarvis appreciates amitabh's steady conversation, but the friendship is still finding its rhythm."
+    },
+    "jarvis->ansh_batra": {
+      "score": 0.32,
+      "tags": [
+        "new-acquaintance"
+      ],
+      "context": "Jarvis is new to this circle and is still learning Ansh Batra's rhythm."
+    },
+    "jarvis->anubhav_prasad": {
+      "score": 0.32,
+      "tags": [
+        "new-acquaintance"
+      ],
+      "context": "Jarvis is new to this circle and is still learning Anubhav Prasad's rhythm."
+    },
+    "jarvis->ghanisht_kaushal": {
+      "score": 0.32,
+      "tags": [
+        "new-acquaintance"
+      ],
+      "context": "Jarvis is new to this circle and is still learning Ghanisht Kaushal's rhythm."
+    },
+    "jarvis->gurnoor_singh": {
+      "score": 0.32,
+      "tags": [
+        "new-acquaintance"
+      ],
+      "context": "Jarvis is new to this circle and is still learning Gurnoor Singh's rhythm."
+    },
+    "jarvis->lavanya_sharma": {
+      "score": 0.37,
+      "tags": [
+        "new-acquaintance"
+      ],
+      "context": "Jarvis is new to this circle and is still learning Lavanya Sharma's rhythm."
+    },
+    "jarvis->parv_singla": {
+      "score": 0.32,
+      "tags": [
+        "new-acquaintance"
+      ],
+      "context": "Jarvis is new to this circle and is still learning Parv Singla's rhythm."
+    },
+    "jarvis->riya_murarka": {
+      "score": 0.37,
+      "tags": [
+        "new-acquaintance"
+      ],
+      "context": "Jarvis is new to this circle and is still learning Riya Murarka's rhythm."
+    },
+    "jarvis->saksham": {
+      "score": 0.42,
+      "tags": [
+        "new-acquaintance"
+      ],
+      "context": "Jarvis is new to this circle and is still learning Saksham's rhythm."
+    },
+    "jarvis->tanishq": {
+      "score": 0.32,
+      "tags": [
+        "new-acquaintance"
+      ],
+      "context": "Jarvis is new to this circle and is still learning Tanishq's rhythm."
+    },
+    "lavanya_sharma->amitabh": {
+      "score": 0.5700000000000001,
+      "tags": [
+        "campus-acquaintance",
+        "low-pressure"
+      ],
+      "context": "Lavanya Sharma and Amitabh know each other through basketball and debate. Lavanya Sharma appreciates amitabh's steady conversation, but the friendship is still finding its rhythm."
+    },
     "lavanya_sharma->ansh_batra": {
       "score": 0.44,
       "tags": [
@@ -289,6 +508,13 @@
       ],
       "context": "Lavanya matches Gurnoor's energy perfectly. Their flirting is shameless and hilarious."
     },
+    "lavanya_sharma->jarvis": {
+      "score": 0.37,
+      "tags": [
+        "new-acquaintance"
+      ],
+      "context": "Lavanya Sharma has only recently met Jarvis; the connection is open but untested."
+    },
     "lavanya_sharma->parv_singla": {
       "score": 0.43,
       "tags": [
@@ -321,6 +547,14 @@
       ],
       "context": "Lavanya is proudly watching Tanishq's glow-up and isn't shy about hyping him up."
     },
+    "parv_singla->amitabh": {
+      "score": 0.54,
+      "tags": [
+        "campus-acquaintance",
+        "low-pressure"
+      ],
+      "context": "Parv Singla and Amitabh know each other through running and music. Parv Singla appreciates amitabh's steady conversation, but the friendship is still finding its rhythm."
+    },
     "parv_singla->ansh_batra": {
       "score": 0.56,
       "tags": [
@@ -353,6 +587,13 @@
       ],
       "context": "Parv and Gurnoor are basically soulmates in crime. Their friendship includes shared hangovers, secrets, and blurry memories."
     },
+    "parv_singla->jarvis": {
+      "score": 0.32,
+      "tags": [
+        "new-acquaintance"
+      ],
+      "context": "Parv Singla has only recently met Jarvis; the connection is open but untested."
+    },
     "parv_singla->lavanya_sharma": {
       "score": 0.43,
       "tags": [
@@ -385,6 +626,14 @@
       ],
       "context": "Parv loves hyping Tanishq up and watching him gain confidence."
     },
+    "riya_murarka->amitabh": {
+      "score": 0.3,
+      "tags": [
+        "campus-acquaintance",
+        "low-pressure"
+      ],
+      "context": "Riya Murarka and Amitabh know each other through reading circles and long runs. Riya Murarka appreciates amitabh's steady conversation, but the friendship is still finding its rhythm."
+    },
     "riya_murarka->ansh_batra": {
       "score": 0.5,
       "tags": [
@@ -417,6 +666,13 @@
       ],
       "context": "Riya finds Gurnoor's energy entertaining in small doses."
     },
+    "riya_murarka->jarvis": {
+      "score": 0.37,
+      "tags": [
+        "new-acquaintance"
+      ],
+      "context": "Riya Murarka has only recently met Jarvis; the connection is open but untested."
+    },
     "riya_murarka->lavanya_sharma": {
       "score": 0.57,
       "tags": [
@@ -449,6 +705,14 @@
       ],
       "context": "Riya notices Tanishq's respectful efforts and finds it sweet, but keeps things slow and platonic for now."
     },
+    "saksham->amitabh": {
+      "score": 0.53,
+      "tags": [
+        "campus-acquaintance",
+        "low-pressure"
+      ],
+      "context": "Saksham and Amitabh know each other through badminton and strategy games. Saksham appreciates amitabh's steady conversation, but the friendship is still finding its rhythm."
+    },
     "saksham->ansh_batra": {
       "score": 0.34,
       "tags": [
@@ -481,6 +745,13 @@
       ],
       "context": "Saksham finds Gurnoor's energy exhausting but entertaining."
     },
+    "saksham->jarvis": {
+      "score": 0.42,
+      "tags": [
+        "new-acquaintance"
+      ],
+      "context": "Saksham has only recently met Jarvis; the connection is open but untested."
+    },
     "saksham->lavanya_sharma": {
       "score": 0.47,
       "tags": [
@@ -513,6 +784,14 @@
       ],
       "context": "Saksham quietly supports Tanishq's confidence growth with dry but kind humor."
     },
+    "tanishq->amitabh": {
+      "score": 0.31,
+      "tags": [
+        "campus-acquaintance",
+        "low-pressure"
+      ],
+      "context": "Tanishq and Amitabh know each other through strategy games and playlists. Tanishq appreciates amitabh's steady conversation, but the friendship is still finding its rhythm."
+    },
     "tanishq->ansh_batra": {
       "score": 0.56,
       "tags": [
@@ -545,6 +824,13 @@
       ],
       "context": "Tanishq is slowly getting pulled into Gurnoor's fun orbit and enjoying it."
     },
+    "tanishq->jarvis": {
+      "score": 0.32,
+      "tags": [
+        "new-acquaintance"
+      ],
+      "context": "Tanishq has only recently met Jarvis; the connection is open but untested."
+    },
     "tanishq->lavanya_sharma": {
       "score": 0.63,
       "tags": [
@@ -578,4 +864,4 @@
       "context": "Tanishq enjoys Saksham's sarcasm and finds it comforting in its own way."
     }
   }
-}
+}

backend/data/personalities/amitabh/amitabh.json

@@ -0,0 +1,22 @@
+{
+  "Name": "Amitabh",
+  "Age": "20",
+  "Gender": "Male",
+  "Branch": "Computer Science",
+  "Home City": "Delhi",
+  "Hostel": "Beas",
+  "daily_plan_req": "Gym in the morning, classes, evening coding + chai sessions, night gaming or deep talks",
+  "innate": "Calm, observant, secretly sarcastic, gets easily flustered by bold flirting but plays it cool",
+  "learned": "How to give good advice while hiding his own chaos, how to handle friends' impulsiveness",
+  "lifestyle": "Lowkey chill but down for spontaneous shit at 2am. Lowkey addicted to emotional tension and slow-burn crushes",
+  "hobbies": "Hardware tinkering, playlists, late-night chai, overthinking texts, secret meme saving",
+  "goals": "Graduate with good grades, figure out what he wants in relationships, maybe finally make a move on someone",
+  "interests": [
+    "Deep conversations",
+    "Tech",
+    "Flirty banter",
+    "Gym",
+    "Music",
+    "Quiet tension with girls/guys"
+  ]
+}

backend/data/personalities/jarvis/jarvis.json

@@ -0,0 +1,22 @@
+{
+  "Name": "Jarvis",
+  "Age": "18",
+  "Gender": "Non-binary",
+  "Branch": "Mechanical Engineering",
+  "Home City": "Indore",
+  "Hostel": "Chenab",
+  "daily_plan_req": "Wake up at 7 AM for a morning run or quick gym session. Attend core engineering lectures from 9 AM to 4 PM, with a lunch break at the mess. Dedicate 4 PM to 6 PM to library studies and assignment completion. The evenings are prioritized for socializing, communal dinner with friends, and engaging in light-hearted hostel activities. Late nights are reserved for deep dives into single-player gaming sessions and relaxing before lights out at midnight.",
+  "innate": "I possess a natural curiosity for how physical systems and machinery work, which pairs well with my optimistic and social nature. I am inherently empathetic and quick to make friends, always looking to find common ground with those around me to foster a welcoming social environment.",
+  "learned": "Through my first year, I have learned how to manage heavy academic workloads effectively without sacrificing my mental health. I have developed strong skills in CAD software, collaborative problem-solving, and the art of navigating complex social dynamics in a communal living setting.",
+  "lifestyle": "I lead a balanced life that centers around the high-energy environment of my hostel. I value my friendships deeply and make it a point to be an active presence in the student community. My routine allows for professional growth through academics while keeping enough space for creative decompression via gaming and hobbies.",
+  "hobbies": "My hobbies include immersive PC gaming, particularly narrative-driven RPGs, tinkering with basic electronics and hardware, playing casual chess in the common room, and curating indie music playlists to share with friends.",
+  "goals": "To successfully secure a prestigious internship in the robotics field by my third year, maintain a consistent academic record above 8.5 CGPA, and cultivate a supportive social circle that enriches my college experience.",
+  "interests": [
+    "PC Gaming",
+    "Robotics",
+    "Mechanical Design",
+    "Chess",
+    "Indie Music",
+    "Photography"
+  ]
+}

backend/pathfinder.py

@@ -1,12 +1,7 @@
-"""pathfinder — pixel-space BFS pathfinding over the campus walkability map.
-
-Loads path.png (white pixels = walkable) and provides shortest_path(),
-stats(), and is_walkable() for the engine and the /api/path endpoints.
-
-Architecture: consumed by Odin.py and the agent action manager (Actions.py)
-to compute routes between buildings; anchors come from entrypoint.json.
-Design: 4-neighbor BFS with nearest-walkable endpoint snapping, because
-doors and interiors sit just off the walkable network.
+"""
+Core pathfinding module — imported by Odin.py and pixel_pathfinder.py.
+Loads path.png into a set of walkable (white) pixels and provides
+BFS shortest_path(), stats(), and is_walkable() helpers.
 """
 
 import os

backend/src/agents/Actions.py

@@ -1,14 +1,23 @@
-"""Actions — the agent's executor: turns a validated day plan into movement
-and activity on the map, tick by tick.
+"""
+Actions -- manages agent action execution: last, current, next.
+
+Takes the raw day plan produced by day_planner.py and drives it forward
+tick by tick. Handles three action types:
+
+  1. MOVE    -- agent walks from place A to place B (pathfinder.py)
+  2. MISC    -- static activity: studying, coding, eating, etc.
+  3. CONVERSATION -- triggered when two agents are in proximity.
 
-Owns the AgentActionManager (last/current/next action state machine),
-location resolution (place id -> pixel position), route computation via
-pathfinder, and conversation freeze/resume.
+The module converts location_id strings (from day plans) into pixel
+coordinates (from entrypoint.json) and uses the BFS pathfinder to
+compute walkable paths between locations.
 
-Architecture: the only module that advances agent position; called by the
-brain/body each tick and checkpointed whole by checkpoint_manager.
-Design: travel is schedule-aware (arrives by the plan's end time); the
-agent keeps its origin location_id while in transit.
+Usage:
+    from src.agents.Actions import AgentActionManager, LocationResolver
+
+    resolver = LocationResolver()
+    manager = AgentActionManager("parv_singla", day_plan, initial_position)
+    state = manager.tick(world_tick, snapshot)
 """
 
 from __future__ import annotations
@@ -73,7 +82,6 @@ class ActionState(BaseModel):
     path_index: int = 0       # current position along path
     energy_change: float = 0.0      # total change over entire action
     emotion_change: float = 0.0     # total change over entire action
-    energy_target: Optional[float] = None  # declared cumulative energy at action end (0-1); None = delta-based
     is_final_plan_action: bool = False
     event_id: Optional[str] = None  # data-driven world event, when applicable
 
@@ -348,7 +356,6 @@ class AgentActionManager:
             position=position,
             energy_change=plan_action.get("energy_change", 0.0),
             emotion_change=plan_action.get("emotion_change", 0.0),
-            energy_target=plan_action.get("energy_target"),
             is_final_plan_action=bool(self.day_plan and plan_action is self.day_plan[-1]),
             event_id=plan_action.get("world_event_id"),
         )

backend/src/agents/Long_term.py

@@ -1,15 +1,10 @@
-"""Long_term — Qdrant-backed long-term memory interface.
+"""Qdrant-backed long-term memory interface.
 
-Provides the process-wide MemoryRetriever singleton over vector_memory.py;
-long-term records live only in Qdrant, while per-day short-term files are
-summarized and indexed at day handoff.
-
-Architecture: consumed by brain.py and the engine for retrieval; paired
-with Short_term.py (operational memory) and vector_memory.py (storage).
-Design: deliberately no JSON archive reader — Qdrant is the single
-long-term source of truth.
+Long-term agent memory is stored only in Qdrant.  Short-term JSON files remain
+the operational record for the active simulation day; they are summarized and
+indexed at handoff, then removed.  This module deliberately has no JSON
+archive reader or keyword-search fallback.
 """
-
 from __future__ import annotations
 
 from typing import List, Optional, Protocol, runtime_checkable

backend/src/agents/Short_term.py

@@ -1,13 +1,13 @@
-"""Short_term — per-agent, per-day operational memory.
+"""
+Short-term Memory -- per-agent, per-day detailed memory store.
+
+Stores the full day's data (plan, events, conversations, world snapshots)
+as a single JSON file per persona per simulation date.
 
-Stores the full day (plan, events, conversations, world snapshots, LLM
-daily summary) as one JSON file per persona per date, with atomic writes,
-and archives the day to Qdrant long-term memory at handoff.
+File layout:
+  data/Short_term_db/<persona_name>/<YYYY-MM-DD>.json
 
-Architecture: the memory layer between the engine and long-term storage;
-written by world_engine.py, read by brain.py and day_planner.py.
-Design: short-term files are the operational record of the active day
-and are removed after successful archival.
+Implements MemoryStreamProtocol (from tick_graph.py) for tick-graph integration.
 """
 
 from __future__ import annotations

backend/src/agents/Single_agent.py

@@ -1,12 +1,24 @@
-"""Single_agent — single-agent planning graph used for one-persona runs.
-
-A minimal LangGraph (retrieve memories -> generate day plan) that
-exercises the planner for a single persona in isolation.
-
-Architecture: a debugging/study tool parallel to the full engine; the
-docstring reserves future nodes (execute_tick, reflect, conversation).
-Design: kept intentionally small so single-agent experiments do not drag
-in the whole WorldEngine.
+"""
+Main brain / command centre of a single agent.
+
+Makes decisions, calls and delegates tasks to sub-modules (day_planner,
+memory, reflection, etc.), and runs the agent's action loop.
+
+Exports:
+  create_agent_graph() -> CompiledGraph[AgentState]
+    A single-agent LangGraph. Currently one node: generate_day_plan.
+    Future: execute_tick, reflect, update_memory, conversation.
+
+Usage as a library (for the multi-agent orchestrator):
+  graph = create_agent_graph()
+  result = graph.invoke({
+      "persona_name": "parv_singla",
+      "persona": {...},
+      "current_time": "2026-07-03 06:00",
+  })
+
+Usage from CLI:
+  python Single_agent.py parv_singla
 """
 
 from __future__ import annotations

backend/src/agents/autonomy.py

@@ -1,12 +1,9 @@
-"""autonomy — schema for a future per-minute LLM "deviate from plan" decision.
-
-Defines AutonomyDecision (deviate / deviation_type / reason / duration)
-as the structured contract for a behavior switch the brain may request.
+"""
+Autonomy — schema for the per-minute LLM decision to deviate from the plan.
 
-Architecture: prepared for brain.py and the engine; currently no call
-site exists (the decision is gated off until enabled).
-Design: kept as a standalone schema so enabling autonomy later requires
-no changes to existing callers.
+The brain calls this once per agent per minute (when enabled). The LLM
+sees the agent's persona, current plan, and nearby surroundings, then
+decides whether to continue the plan or deviate temporarily.
 """
 
 from __future__ import annotations

backend/src/agents/body.py

@@ -1,13 +1,23 @@
-"""body — the agent's motor layer; the only way the brain moves the body.
-
-Wraps the AgentActionManager so the brain can advance, enter or resume
-conversations, and read position/current action without touching executor
-details.
-
-Architecture: sits between brain.py and Actions.py; used by the engine's
-act phase via brain.act().
-Design: the brain never manipulates the manager directly — the body is
-the single command interface for movement.
+"""
+Body -- the agent's "limbs". The motor layer the brain commands.
+
+In the human-like architecture the *brain* (brain.py) does the thinking:
+it perceives, recalls, and decides. It never moves the agent directly.
+Instead it issues motor commands to this Body, which is the only thing that
+actually changes the agent's position and current activity.
+
+The Body is a thin, behaviour-preserving adapter around the existing action
+state machine (`AgentActionManager` in Actions.py) -- the proven executor
+that walks paths and steps through the day plan. Wrapping it (rather than
+replacing it) means the body/brain split is a clean architectural layer with
+zero change to how movement and actions actually run.
+
+Motor command surface (all 0-LLM):
+  - advance(tick)              : take the next step of the current plan
+  - enter_conversation(name)   : freeze into a conversation with someone
+  - resume(day_plan)           : leave conversation / reload the plan
+Read-only senses of the body's own state:
+  - position, current_action, is_last_action
 """
 
 from __future__ import annotations

backend/src/agents/brain.py

@@ -1,13 +1,13 @@
-"""brain — per-agent cognition: the LLM-backed decide step.
+"""
+Brain -- the agent's cognition / command centre.
 
-Each tick, when the engine detects novel observations (and gates pass),
-the brain asks the LLM whether to continue the current plan or replan,
-and issues motor commands through the body.
+Each tick the brain may be called to decide (via LLM) whether the agent
+should continue their current plan or replan, based on novel observations.
+The LLM call is gated: it only fires when the perceive phase detects a change
+in the set of (agent_id, action_description) within 50px.
 
-Architecture: called by WorldEngine._phase_llm_decide; consumes memories
-from Short_term/Long_term and produces TickDecision for the replan phase.
-Design: conservative by prompt (replan only for significant events) and
-by default (any LLM failure falls back to "continue").
+When no novel observations exist, the brain returns "continue" without an LLM
+call — the agent follows its existing plan.
 """
 
 from __future__ import annotations

backend/src/agents/conversation.py

@@ -1,13 +1,33 @@
-"""conversation — generates agent-to-agent dialogue and its effects.
+"""
+Conversation -- generates dialogue between two agents via a single LLM call.
+
+The WorldEngine calls `generate_conversation()` when two agents share a
+location_id, are both in compatible actions (not sleeping), and neither is
+already mid-conversation.
+
+The single LLM call produces the full conversation (messages, summary,
+duration, sentiment, relationship delta). Both agents get their current
+action overwritten to "Chatting with X" for the duration, then naturally
+replan via the tick graph when it expires.
 
-One LLM call produces a full 4-12 message conversation with duration,
-sentiment, per-agent relationship, energy, and emotion effects; also owns
-the persistent RelationshipMatrix (directed scores, tags, context).
+Usage:
+    from src.agents.conversation import generate_conversation, RelationshipMatrix
 
-Architecture: triggered by WorldEngine proximity detection; results are
-applied to Short_term memory, the relationship matrix, and wellbeing.
-Design: whole-conversation generation trades turn-by-turn fidelity for
-roughly an order of magnitude fewer LLM calls and structured effects.
+    matrix = RelationshipMatrix()
+    result = generate_conversation(
+        agent_a_id="parv_singla",
+        agent_b_id="tanishq",
+        persona_a=gray_wilder_persona,
+        persona_b=jules_persona,
+        plan_a=gray_wilder_plan,
+        plan_b=jules_plan,
+        action_a=gray_wilder_current_action,
+        action_b=jules_current_action,
+        rel_a_to_b=matrix.get("parv_singla", "tanishq"),
+        rel_b_to_a=matrix.get("tanishq", "parv_singla"),
+        location_id="mess",
+        current_hhmm="08:05",
+    )
 """
 
 from __future__ import annotations
@@ -84,13 +104,6 @@ class ConversationResult(BaseModel):
     duration_minutes: int = Field(ge=6, le=20)
     sentiment: Literal["positive", "neutral", "negative"]
     relationship_delta: float = Field(ge=-0.15, le=0.15)
-    # LLM-decided net wellbeing effect of the chat for each participant.
-    # Energy and mood each run from 0.0 to 1.0; the resulting value after the
-    # chat must stay inside that range (never below 0% or above 100%).
-    energy_delta_a: float = 0.0
-    emotion_delta_a: float = 0.0
-    energy_delta_b: float = 0.0
-    emotion_delta_b: float = 0.0
     # Folded-in replan decision: avoids a separate 4-call day-plan regeneration
     # per agent after every conversation. True only when the conversation
     # genuinely changes an agent's immediate intentions.
@@ -545,9 +558,6 @@ Return a JSON object with:
 - "duration_minutes": integer from 6 to 20 that matches the amount of dialogue
 - "sentiment": "positive" | "neutral" | "negative"
 - "relationship_delta": float between -0.15 and 0.15 (how this conversation changes their relationship)
-- "energy_delta_a" / "energy_delta_b": each agent's net ENERGY change from this chat (positive = recharged, negative = drained)
-- "emotion_delta_a" / "emotion_delta_b": each agent's net MOOD change from this chat (positive = lifted, negative = dampened)
-- ENERGY AND MOOD each run from 0.0 to 1.0 (0% to 100%). Add each delta to the agent's current value shown above; the resulting value must stay between 0.0 and 1.0 — never above 100% or below 0%
 - "should_replan": boolean — true ONLY if this conversation genuinely changes what one of them intends to do next (e.g. they agree to meet, go somewhere together, or drop a task). Default false; most casual chats do NOT require replanning.
 - "plan_change": short string describing the change if should_replan is true, else null"""
 

backend/src/agents/daily_flavor.py

@@ -1,13 +1,9 @@
-"""daily_flavor — random daily theme and emotion for plan variety.
-
-Picks one of ten themes (e.g. Sports, Academics) and one of ten emotions
-(e.g. Excited, Melancholic) per agent per day, injected into planner
-prompts so days do not feel scripted.
+"""
+Daily flavor — random theme and emotion pickers for day_planner.
 
-Architecture: consumed by day_planner.py at plan time; a single, tiny
-dependency-free module.
-Design: deliberate randomness is the point — variety is the seed of
-unpredictable-but-plausible schedules.
+Each day an agent gets a random theme (what they focus on) and emotion
+(their mood), injected into the planner prompts so the schedule doesn't
+feel identical every day.
 """
 
 from __future__ import annotations

backend/src/agents/day_planner.py

@@ -1,16 +1,33 @@
-"""day_planner — the cognitive core: generates validated day plans.
-
-A LangGraph pipeline that decomposes a day coarse (5-8 blocks) -> hourly
--> fine (5-15 min actions), validates every level (time coverage, tiling,
-locations, academic venue policy, content safety), retries up to
-MAX_PLAN_RETRIES, and force-accepts a deterministic fallback schedule if
-the model keeps failing.
-
-Architecture: the only producer of day_plan; called by the engine at
-startup, day handoff, and replan; consumes persona, places, memories,
-events, and current wellbeing.
-Design: validation is local and deterministic before any LLM semantic
-QA; energy targets are embedded in every action (see engine glide).
+"""
+A script that plans the day of an agentic personality when handed over required data
+Per plan takes 4 LLM calls (atlest) Coarse, Hourly, Fine, Validation, for Planning a
+day in one agent's life.
+
+Tier-1 LangGraph subgraph: agent day-planning.
+
+Pipeline (mirrors Generative Agents' Planning module, coarse -> hourly -> fine,
+with a validation/retry loop):
+
+    generate_coarse_plan -> decompose_hourly -> decompose_fine -> validate_plan
+                                                                        |
+                                                        conflict? --yes-+ (loop back to generate_coarse_plan)
+                                                            |
+                                                            no -> END
+
+LLM backend: Google Gemini via the `google-genai` SDK.
+
+For now `relevant_memories` and `yesterday_summary` are expected to arrive
+empty ([] / None) -- the prompts already handle that gracefully so you can
+wire in real retrieval/memory later without touching this file's structure.
+
+FILE NOTES:
+Prompt structure can be improved
+Places are being feed in Name : , Desc : format, this can be improved
+disabled location check in validate plan : can add more places
+
+Prompt templates have to improve
+
+Have to figure out how to run this in the backend server, currently it is running standalone
 """
 
 from __future__ import annotations
@@ -90,7 +107,7 @@ def _academic_venue_policy(persona: Dict[str, Any]) -> str:
         return "No branch-specific policy is known; choose the listed location that explicitly fits."
     return (
         f"This student is in {branch}. Branch-specific classes, tutorials, and labs may use "
-        f"`{destination}` or LHC/ SAB. LHC is only for common/core/elective/guest/large shared sessions and classes. "
+        f"`{destination}` or Library/ SAB. LHC is only for common/core/elective/guest/large shared sessions and classes. "
         
     )
 
@@ -120,11 +137,10 @@ def _local_academic_venue_check(actions: List[Dict[str, Any]], persona: Dict[str
             continue
         if any(word in description for word in _SHARED_SESSION_WORDS):
             continue
-        allowed_venues = {required, "SAB", "LHC"}
-        if location not in allowed_venues:
+        if location != required:
             return (
                 f"branch-specific academic action '{action.get('action')}' for {persona.get('Branch')} "
-                f"must use {required}, SAB, or LHC, not {location}"
+                f"must use {required}, not {location}"
             )
     return None
 
@@ -196,14 +212,6 @@ class CoarseBlock(BaseModel):
     )
     energy_change: float = 0.0
     emotion_change: float = 0.0
-    energy_target: Optional[float] = Field(
-        default=None,
-        description=(
-            "Optional: your declared cumulative energy level (0.0-1.0) the agent "
-            "should have when this block ends. The runtime glides energy toward "
-            "this target. Follow the TIME-OF-DAY rules in the guidance."
-        ),
-    )
 
 
 class CoarsePlanOutput(BaseModel):
@@ -236,10 +244,6 @@ class HourlyBlock(BaseModel):
     parent_activity: str = Field(description="The coarse block this refines")
     energy_change: float = 0.0
     emotion_change: float = 0.0
-    energy_target: Optional[float] = Field(
-        default=None,
-        description="Optional cumulative energy level (0.0-1.0) at the end of this block.",
-    )
 
 
 class HourlyPlanOutput(BaseModel):
@@ -255,14 +259,6 @@ class FineAction(BaseModel):
     sub_area: Optional[str] = Field(default=None, description="One of that place's sub_areas, if applicable")
     energy_change: float = Field(description="Energy change [-1.0, 1.0] over this action; positive=restorative, negative=tiring")
     emotion_change: float = Field(description="Emotion change [-1.0, 1.0] over this action; positive=uplifting, negative=draining")
-    energy_target: Optional[float] = Field(
-        default=None,
-        description=(
-            "Optional cumulative energy level (0.0-1.0) the agent should have when "
-            "this action ends. The runtime glides energy toward this declared target; "
-            "follow the TIME-OF-DAY rules in the guidance."
-        ),
-    )
 
 
 class FinePlanOutput(BaseModel):
@@ -275,7 +271,6 @@ class AtomicLocationAssignment(BaseModel):
     sub_area: Optional[str] = None
     energy_change: float = 0.0
     emotion_change: float = 0.0
-    energy_target: Optional[float] = None
 
 class AtomicLocationOutput(BaseModel):
     assignments: List[AtomicLocationAssignment]
@@ -386,21 +381,6 @@ def _flavor_block(state: DayPlannerState) -> str:
     return f"Today's vibe: {emotion}, leaning into {theme}."
 
 
-def _agent_name(state: DayPlannerState) -> str:
-    persona = state.get("persona") or {}
-    return str(persona.get("Name") or persona.get("name") or "unknown")
-
-
-def _conflict_feedback(state: DayPlannerState) -> str:
-    reason = state.get("conflict_reason")
-    if not reason:
-        return ""
-    return (
-        f"\n\nNOTE: a previous attempt was rejected for this reason, avoid repeating it:\n"
-        f"{reason}"
-    )
-
-
 def _planning_window(state: DayPlannerState) -> tuple[str, str]:
     """Return the exact time window owned by this planner invocation."""
     current_time = state.get("current_time", "")
@@ -463,38 +443,6 @@ def _within_source_windows(record: Dict[str, Any], sources: List[Dict[str, Any]]
 # Nodes
 # ---------------------------------------------------------------------------
 
-_WELLBEING_GUIDANCE = (
-    "For EACH block/action, assign energy_change and emotion_change values. "
-    "These are the NET changes to the agent's running energy and mood caused "
-    "by that activity.\n"
-    "- energy_change: positive = restores energy, negative = drains energy\n"
-    "- emotion_change: positive = lifts mood, negative = dampens mood\n"
-    "Energy and mood each run from 0.0 to 1.0 (0% to 100%). Track the day "
-    "cumulatively: after EVERY activity the running total of energy and of "
-    "mood must stay between 0.0 and 1.0 — never above 100% or below 0%.\n"
-    "Be realistic for the persona: a full night's sleep restores a lot "
-    "(about +0.2 to +0.5), meals and rest restore a little, hard exercise and "
-    "all-nighters drain substantially, ordinary classes and chores sit in "
-    "between.\n"
-    "TIME-OF-DAY (circadian) RULES — energy must follow the clock, and YOU "
-    "choose the exact numbers:\n"
-    "- Energy is highest after waking (aim for roughly 0.65-0.85 at day "
-    "start), dips mid-afternoon around 14:00, declines through the evening, "
-    "and is LOWEST before bed (roughly 0.15-0.35 by 22:00-23:00).\n"
-    "- A day that ends near where it started is unrealistic: plan the day to "
-    "end at least 0.2-0.4 BELOW its morning energy.\n"
-    "- After ~22:00 nothing restores energy except sleep — late study, "
-    "screens, and socialising drain or stay neutral.\n"
-    "- Only sleep, meals, and genuine rest restore; classes, labs, study, "
-    "and exercise drain at least a little, sized to how long and demanding "
-    "the activity is.\n"
-    "Optionally declare energy_target for EACH block/action: your cumulative "
-    "energy level (0.0-1.0) at the moment it ends. The runtime glides the "
-    "agent's energy toward each declared target, so use it to encode the "
-    "circadian curve above. When omitted, the runtime uses energy_change "
-    "directly."
-)
-
 def generate_coarse_plan(state: DayPlannerState) -> DayPlannerState:
     persona = state["persona"]
     mode = state.get("mode", "full_day")
@@ -528,18 +476,10 @@ def generate_coarse_plan(state: DayPlannerState) -> DayPlannerState:
         "worth planning separately. Examples: sleeping, attending a class/lecture, "
         "sitting an exam, watching a movie, a long uninterrupted study/deep-work session.\n"
         "- flexible: an activity that naturally contains distinct on-site sub-activities.\n\n"
-        + _WELLBEING_GUIDANCE
+        "For EACH block, assign realistic energy_change and emotion_change values. Routine classes, labs, study, meals, and chores should be near neutral (usually -0.03 to +0.03); reserve larger positive changes for rare, meaningful events."
     )
     if current_loc:
         loc_hint = f"\nThe agent is currently at: {current_loc}. Start the plan from this location."
-    wellbeing_line = (
-        f"CURRENT WELLBEING: energy {state['current_energy']:.2f}/1.0, "
-        f"emotion {state['current_emotion']:.2f}/1.0 — plan the remaining day "
-        "from these values, keeping the cumulative energy and mood totals "
-        "between 0.0 and 1.0.\n\n"
-        if state.get("current_energy") is not None and state.get("current_emotion") is not None
-        else ""
-    )
     user_prompt = (
         f"PERSONA:\n{_persona_block(persona)}\n\n"
         f"RELEVANT MEMORIES:\n{_memories_block(state.get('relevant_memories', []))}\n\n"
@@ -550,13 +490,15 @@ def generate_coarse_plan(state: DayPlannerState) -> DayPlannerState:
         f"Plan mode: {mode}\n"
         f"REQUIRED OUTPUT WINDOW: {_window_constraint(state)}\n"
         f"Agent location: {current_loc or 'unknown'}{loc_hint}\n\n"
-        f"{wellbeing_line}"
         f"DAY-HANDOFF CONTINUITY:\n{state.get('handoff_context') or '(none)'}\n\n"
         "Generate the coarse plan now."
     )
 
     if state.get("conflict_reason"):
-        user_prompt += _conflict_feedback(state)
+        user_prompt += (
+            f"\n\nNOTE: a previous attempt was rejected for this reason, avoid repeating it:\n"
+            f"{state['conflict_reason']}"
+        )
 
     required_start = _planning_window(state)[0]
     result = call_gemini(
@@ -565,7 +507,7 @@ def generate_coarse_plan(state: DayPlannerState) -> DayPlannerState:
         _coarse_output_schema(required_start),
         "default",
     )
-    logger.info("[day_planner][%s] coarse plan generated: %d blocks", _agent_name(state), len(result.blocks))
+    logger.info("[day_planner] coarse plan generated: %d blocks", len(result.blocks))
 
     return {
         **state,
@@ -582,16 +524,14 @@ def validate_coarse_window(state: DayPlannerState) -> DayPlannerState:
         action_key="activity",
     )
     if issue:
-        logger.info("[day_planner][%s] coarse-window validation failed: %s", _agent_name(state), issue)
+        logger.info("[day_planner] coarse-window validation failed: %s", issue)
         return {
             **state,
             "conflict_detected": True,
             "conflict_reason": issue,
             "retry_count": state.get("retry_count", 0) + 1,
         }
-    # Keep conflict_reason across the retry cycle: intermediate stages must not
-    # clear the last rejection before the failing stage regenerates.
-    return {**state, "conflict_detected": False}
+    return {**state, "conflict_detected": False, "conflict_reason": None}
 
 
 def decompose_hourly(state: DayPlannerState) -> DayPlannerState:
@@ -619,7 +559,6 @@ def decompose_hourly(state: DayPlannerState) -> DayPlannerState:
             "granularity": "atomic",
             "energy_change": b.get("energy_change", 0.0),
             "emotion_change": b.get("emotion_change", 0.0),
-            "energy_target": b.get("energy_target"),
         }
         for b in atomic_blocks
     ]
@@ -636,8 +575,12 @@ def decompose_hourly(state: DayPlannerState) -> DayPlannerState:
             "(e.g. a meal block of 2 hours can contain 'walk to mess', 'eat', 'socialize'). "
             "Only the blocks provided here need refining. Do not create walk, commute, or "
             "transit sub-blocks: refine only activities performed at the destination.\n\n"
-            + _WELLBEING_GUIDANCE
-            + f"{_window_constraint(state)}"
+            "For EACH block, assign realistic energy_change and emotion_change values:\n"
+            "- energy_change: positive = restorative, negative = tiring\n"
+            "- emotion_change: positive = uplifting, negative = draining\n"
+            "- Routine work, classes, and meals should usually stay within -0.03 to +0.03; do not make ordinary productivity euphoric\n"
+            "- Be realistic for the persona\n\n"
+            f"{_window_constraint(state)}"
         )
         user_prompt = (
             f"PERSONA:\n{_persona_block(persona)}\n\n"
@@ -647,15 +590,13 @@ def decompose_hourly(state: DayPlannerState) -> DayPlannerState:
             f"REQUIRED OUTPUT WINDOW: {_window_constraint(state)}\n\n"
             "Produce the hourly-resolution plan for the blocks listed under "
             "'BLOCKS TO REFINE' only."
-            + _conflict_feedback(state)
         )
         result = call_gemini(system_prompt, user_prompt, HourlyPlanOutput, "default")
         raw_refined = [b.model_dump() for b in result.blocks]
         refined = [block for block in raw_refined if _within_source_windows(block, flexible_blocks)]
         if len(refined) != len(raw_refined):
             logger.warning(
-                "[day_planner][%s] discarded %d hourly refinement block(s) outside flexible source windows",
-                _agent_name(state),
+                "[day_planner] discarded %d hourly refinement block(s) outside flexible source windows",
                 len(raw_refined) - len(refined),
             )
         for b in refined:
@@ -664,8 +605,7 @@ def decompose_hourly(state: DayPlannerState) -> DayPlannerState:
 
     hourly_blocks.sort(key=lambda b: b["start"])
     logger.info(
-        "[day_planner][%s] hourly plan: %d atomic passthrough + %d refined",
-        _agent_name(state),
+        "[day_planner] hourly plan: %d atomic passthrough + %d refined",
         len(passthrough_hourly), len(hourly_blocks) - len(passthrough_hourly),
     )
 
@@ -707,100 +647,14 @@ def validate_hourly_refinement(state: DayPlannerState) -> DayPlannerState:
                 issue = f"hourly refinement '{block.get('activity', 'unknown')}' exceeds its flexible source window"
                 break
     if issue:
-        logger.info("[day_planner][%s] hourly refinement validation failed: %s", _agent_name(state), issue)
+        logger.info("[day_planner] hourly refinement validation failed: %s", issue)
         return {
             **state,
             "conflict_detected": True,
             "conflict_reason": issue,
             "retry_count": state.get("retry_count", 0) + 1,
         }
-    # Keep conflict_reason across the retry cycle (see validate_coarse_window).
-    return {**state, "conflict_detected": False}
-
-def _match_atomic_assignment(
-    activity: str, loc_by_activity: Dict[str, deque]
-) -> Optional[AtomicLocationAssignment]:
-    """Pick a location assignment for an atomic block: exact activity match
-    first, then a case-insensitive label match. The location model sometimes
-    rewrites activity labels between stages, so an exact-only lookup leaves
-    blocks with no assignment and the whole day falls back to force-accept."""
-    candidates = loc_by_activity.get(activity)
-    if candidates:
-        return candidates.popleft()
-    for key in list(loc_by_activity):
-        if not key:
-            continue
-        if (
-            key.lower() == activity.lower()
-            or key.lower() in activity.lower()
-            or activity.lower() in key.lower()
-        ):
-            return loc_by_activity[key].popleft()
-    return None
-
-
-def _fallback_location_id(
-    activity: str,
-    places: List[Place],
-    current_loc: str,
-    persona: Dict[str, Any],
-) -> Optional[str]:
-    """Deterministic venue fallback for atomic blocks the location model did
-    not assign. Returns a valid location_id whenever any place exists, so a
-    missing assignment can never poison validation with location_id None."""
-    valid_ids = {p.id for p in places}
-
-    def pick(candidates: List[Any]) -> Optional[str]:
-        for candidate in candidates:
-            if candidate and candidate in valid_ids:
-                return candidate
-        return None
-
-    text = activity.lower()
-    has = lambda *words: any(word in text for word in words)
-
-    # 1) Explicit venue words inside the activity label win first.
-    if has("library"):
-        return pick(
-            [p.id for p in places if "library" in p.id.lower() or "library" in p.name.lower()]
-        )
-    if has("hostel"):
-        return pick([p.id for p in places if p.type == "residential"] + [persona.get("Hostel")])
-    if has("mess", "canteen", "dining"):
-        return pick(
-            [p.id for p in places if "mess" in p.id.lower() or "mess" in p.name.lower()
-             or "canteen" in p.id.lower() or "canteen" in p.name.lower()]
-        )
-    if has("gym", "workout"):
-        return pick([p.id for p in places if "gym" in p.id.lower() or "gym" in p.name.lower()])
-    if has("sab"):
-        return pick(["SAB", *[p.id for p in places if p.id == "SAB"]])
-    if has("lhc", "lecture hall"):
-        return pick(["LHC", *[p.id for p in places if p.id == "LHC"]])
-
-    # 2) Category-based defaults, matching the venue policy used in prompts.
-    if has("class", "lecture", "lab", "exam", "study", "project", "tutorial", "seminar"):
-        return pick(
-            [p.id for p in places if "department" in p.id.lower() or "department" in p.name.lower()]
-            + ["LHC", "SAB", "library"]
-            + [p.id for p in places if "library" in p.id.lower() or "library" in p.name.lower()]
-        )
-    if has("breakfast", "lunch", "dinner", "meal", "eat", "food"):
-        return pick(
-            [p.id for p in places if "mess" in p.id.lower() or "mess" in p.name.lower()
-             or "canteen" in p.id.lower() or "canteen" in p.name.lower()]
-        )
-    if has("sport", "cricket", "football", "badminton", "exercise", "run", "fitness"):
-        return pick(
-            [p.id for p in places if "sport" in p.id.lower() or "sport" in p.name.lower()
-             or "gym" in p.id.lower() or "gym" in p.name.lower()]
-        )
-    if has("sleep", "rest", "nap", "recover", "personal", "wind down", "chat", "chill", "socialize", "room", "bunk"):
-        return pick([p.id for p in places if p.type == "residential"] + [persona.get("Hostel")])
-
-    # 3) Generic fallback: current position, then hostel, then any place.
-    return pick([current_loc, persona.get("Hostel")]) or next(iter(valid_ids), None)
-
+    return {**state, "conflict_detected": False, "conflict_reason": None}
 
 def decompose_fine(state: DayPlannerState) -> DayPlannerState:
     persona = state["persona"]
@@ -824,9 +678,13 @@ def decompose_fine(state: DayPlannerState) -> DayPlannerState:
             "Default to locations that make sense for this specific persona. "
             "Do not suggest splitting the activity.\n\n"
             "ACADEMIC VENUE POLICY: obey the branch-specific policy provided with the persona. "
-            "Prefer the branch department for classes and labs; SAB and LHC are acceptable.\n\n"
-            + _WELLBEING_GUIDANCE
-            + f"{_window_constraint(state)}"
+            "Do not use SAB as a generic lecture/lab default.\n\n"
+            "For EACH block, assign realistic energy_change and emotion_change values:\n"
+            "- energy_change: positive = restorative, negative = tiring\n"
+            "- emotion_change: positive = uplifting, negative = draining\n"
+            "- Routine work, classes, and meals should usually stay within -0.03 to +0.03; do not make ordinary productivity euphoric\n"
+            "- Be realistic for the persona\n\n"
+            f"{_window_constraint(state)}"
         )
         user_prompt = (
             f"PERSONA:\n{_persona_block(persona)}\n\n"
@@ -841,7 +699,6 @@ def decompose_fine(state: DayPlannerState) -> DayPlannerState:
             "rest, and personal activities use the hostel unless the activity explicitly "
             "requires another place.\n\n"
             "Assign a location to each activity now."
-            + _conflict_feedback(state)
         )
         result = call_gemini(system_prompt, user_prompt, AtomicLocationOutput, "default")
         # Activity labels are not unique (for example, two separate study
@@ -852,40 +709,21 @@ def decompose_fine(state: DayPlannerState) -> DayPlannerState:
             loc_by_activity[assignment.activity].append(assignment)
 
         for b in atomic_blocks:
-            assignment = _match_atomic_assignment(b["activity"], loc_by_activity)
-            if assignment:
-                location_id = assignment.location_id
-                sub_area = assignment.sub_area
-                energy_change = assignment.energy_change
-                emotion_change = assignment.emotion_change
-                energy_target = (
-                    assignment.energy_target
-                    if assignment.energy_target is not None
-                    else b.get("energy_target")
-                )
-            else:
-                location_id = _fallback_location_id(b["activity"], places, current_loc, persona)
-                sub_area = None
-                energy_change = b.get("energy_change", 0.0)
-                emotion_change = b.get("emotion_change", 0.0)
-                energy_target = b.get("energy_target")
-                if location_id is not None:
-                    logger.info(
-                        "[day_planner][%s] no location assignment for '%s' -- deterministic fallback to '%s'",
-                        _agent_name(state),
-                        b["activity"],
-                        location_id,
-                    )
+            candidates = loc_by_activity.get(b["activity"])
+            assignment = candidates.popleft() if candidates else None
             fine_actions.append({
                 "action": b["activity"],
                 "start": b["start"],
                 "end": b["end"],
                 "parent_activity": b["parent_activity"],
-                "location_id": location_id,
-                "sub_area": sub_area,
-                "energy_change": energy_change,
-                "emotion_change": emotion_change,
-                "energy_target": energy_target,
+                "location_id": assignment.location_id if assignment else None,
+                "sub_area": assignment.sub_area if assignment else None,
+                "energy_change": (
+                    assignment.energy_change if assignment else b.get("energy_change", 0.0)
+                ),
+                "emotion_change": (
+                    assignment.emotion_change if assignment else b.get("emotion_change", 0.0)
+                ),
             })
 
     # Flexible blocks: full fine-grained breakdown, as before.
@@ -894,25 +732,23 @@ def decompose_fine(state: DayPlannerState) -> DayPlannerState:
             "You refine hourly blocks into fine-grained, directly executable actions "
             "at roughly 5-15 minute granularity. Each hourly block should be broken "
             "into one or more fine actions spanning exactly its start/end range, no "
-            "gaps or overlaps. The full day must tile exactly, minute by minute: the "
-            "first action of the day starts exactly at 00:00, each action starts the "
-            "instant the previous one ends, and the LAST action of the day ends "
-            "exactly at 24:00 (write the final boundary as 24:00, never 23:59 or "
-            "0:00). Each group of fine actions must start exactly at its assigned "
-            "block's start and end exactly at its block's end -- never spill outside "
-            "your assigned windows. Every action MUST be assigned a location_id, chosen "
+            "gaps or overlaps. Every action MUST be assigned a location_id, chosen "
             "EXACTLY from the provided list -- never invent one.\n\n"
             "Do NOT output walking, commuting, travel, transit, leaving, or arriving "
             "as an action. The runtime owns visible routes between places; every action "
             "you output must be an on-site activity at its assigned location.\n\n"
             "ACADEMIC VENUE POLICY: obey the branch-specific policy provided with the persona. "
-            "Prefer the branch department for classes and labs; SAB and LHC are acceptable.\n\n"
+            "Branch-specific classes/labs must not silently fall back to SAB.\n\n"
             "Make action boundaries feel natural — group related sub-actions together. "
             "Consider typical on-site durations: eating ~20-40min and studying "
             "~30-120min. Keep adjacent location changes realistic by leaving enough "
             "time for the executor to animate transit before the next activity.\n\n"
-            + _WELLBEING_GUIDANCE
-            + f"{_window_constraint(state)}"
+            "For EACH action, assign realistic energy_change and emotion_change values:\n"
+            "- energy_change: positive = restorative, negative = tiring\n"
+            "- emotion_change: positive = uplifting, negative = draining\n"
+            "- Routine work, classes, and meals should usually stay within -0.03 to +0.03; do not make ordinary productivity euphoric\n"
+            "- Be realistic for the persona\n\n"
+            f"{_window_constraint(state)}"
         )
         user_prompt = (
             f"PERSONA:\n{_persona_block(persona)}\n\n"
@@ -927,21 +763,19 @@ def decompose_fine(state: DayPlannerState) -> DayPlannerState:
             "rest, and personal activities use the hostel unless the activity explicitly "
             "requires another place.\n\n"
             "Produce the fine-grained action plan for these blocks now."
-            + _conflict_feedback(state)
         )
         result = call_gemini(system_prompt, user_prompt, FinePlanOutput, "default")
         raw_actions = [action.model_dump() for action in result.actions]
         scoped_actions = [action for action in raw_actions if _within_source_windows(action, flexible_blocks)]
         if len(scoped_actions) != len(raw_actions):
             logger.warning(
-                "[day_planner][%s] discarded %d fine action(s) outside flexible source windows",
-                _agent_name(state),
+                "[day_planner] discarded %d fine action(s) outside flexible source windows",
                 len(raw_actions) - len(scoped_actions),
             )
         fine_actions.extend(scoped_actions)
 
     fine_actions.sort(key=lambda a: a["start"])
-    logger.info("[day_planner][%s] fine plan: %d total actions", _agent_name(state), len(fine_actions))
+    logger.info("[day_planner] fine plan: %d total actions", len(fine_actions))
 
     return {**state, "fine_plan": fine_actions}
 
@@ -1045,7 +879,7 @@ def validate_plan(state: DayPlannerState) -> DayPlannerState:
         state["fine_plan"], state.get("places", [])
     ) or _local_academic_venue_check(state["fine_plan"], state["persona"]) or _local_content_safety_check(state["fine_plan"])
     if local_issue:
-        logger.info("[day_planner][%s] local validation failed: %s", _agent_name(state), local_issue)
+        logger.info("[day_planner] local validation failed: %s", local_issue)
         return {
             **state,
             "conflict_detected": True,
@@ -1073,7 +907,7 @@ def validate_plan(state: DayPlannerState) -> DayPlannerState:
     result = call_gemini(system_prompt, user_prompt, ValidationResult, "default")
 
     if not result.valid:
-        logger.info("[day_planner][%s] semantic validation failed: %s", _agent_name(state), result.reason)
+        logger.info("[day_planner] semantic validation failed: %s", result.reason)
         return {
             **state,
             "conflict_detected": True,
@@ -1081,7 +915,7 @@ def validate_plan(state: DayPlannerState) -> DayPlannerState:
             "retry_count": state.get("retry_count", 0) + 1,
         }
 
-    logger.info("[day_planner][%s] plan validated successfully", _agent_name(state))
+    logger.info("[day_planner] plan validated successfully")
     return {
         **state,
         "conflict_detected": False,
@@ -1099,8 +933,7 @@ def route_after_validation(state: DayPlannerState) -> str:
         return "accept"
     if state.get("retry_count", 0) >= MAX_PLAN_RETRIES:
         logger.warning(
-            "[day_planner][%s] max retries (%d) reached, force-accepting last plan with error flag",
-            _agent_name(state),
+            "[day_planner] max retries (%d) reached, force-accepting last plan with error flag",
             MAX_PLAN_RETRIES,
         )
         return "give_up"
@@ -1261,8 +1094,6 @@ def run(agent: Any, world_state: dict) -> dict:
         "places": places,
         "mode": mode,
         "current_location_id": world_state.get("current_location_id"),
-        "current_energy": world_state.get("energy_level"),
-        "current_emotion": world_state.get("emotion_state"),
         "handoff_context": world_state.get("handoff_context"),
         "upcoming_events": world_state.get("upcoming_events", []),
         "daily_theme": theme,

backend/src/agents/memory_index.py

@@ -1,14 +1,12 @@
-"""memory_index — CLI admin tool for the Qdrant memory backend.
+"""Operate Qdrant-only long-term memory.
 
-Supports status, migrate-json (one-time import of legacy JSON archives
-into Qdrant), and clear operations for per-agent collections.
+``migrate-json`` imports legacy Long_term_db archives once.  Pass
+``--delete-source`` only after checking the reported indexed count; it removes
+the obsolete JSON archives after their idempotent Qdrant upsert succeeds.
 
-Architecture: a developer utility that calls vector_memory.py directly;
-not part of the simulation runtime.
-Design: migration is idempotent (deterministic point ids) so reruns are
-safe; --delete-source is explicit and documented.
+``clear`` removes only Valhalla's durable Qdrant collections. It never touches
+short-term runtime files, checkpoints, or the live simulation process.
 """
-
 from __future__ import annotations
 
 import argparse

backend/src/agents/react.py

@@ -1,14 +1,16 @@
-"""react — deterministic 0-LLM "keep going or replan?" filter.
-
-Given the agent's current action and fresh observations, returns a
-decision: no action -> replan, action finished -> replan, mid-action with
-no novelty -> continue. The LLM variant of this decision lives in
-brain.decide_tick and fires only on novelty.
-
-Architecture: retained as the cheap reflex layer; the production engine
-currently routes decisions through brain.py and WorldEngine phases.
-Design: every path has a fixed answer — this step must never stall the
-simulation on a model call.
+"""
+React -- decides, given an agent's current action (if any) and what it just
+perceived, whether to keep executing that action or interrupt and replan.
+
+Design notes
+------------
+All calls into this module are cheap heuristic checks with no LLM round-trip:
+  - no current action yet                       -> always replan
+  - current action's end_tick has passed         -> always replan
+  - mid-action, nothing new perceived this tick  -> always continue
+
+The LLM-based decision layer has moved to brain.decide_tick(), which runs
+only when the perceive phase detects novel observations.
 """
 
 from __future__ import annotations

backend/src/agents/vector_memory.py

@@ -1,15 +1,10 @@
-"""vector_memory — persistent semantic long-term memory (Qdrant + Gemini).
+"""Persistent, per-persona Cloud Qdrant long-term memory and RAG retrieval.
 
-Archives each agent's completed days as memory records, embeds queries
-with Gemini embeddings, and retrieves with a 65/20/15 semantic/importance/
-recency score plus diversity and storage-quota constraints.
-
-Architecture: the storage engine behind Long_term.py; called at day
-handoff (archive), by planning (retrieval), and by brain decisions.
-Design: importance is a static per-kind table, recency decays by real
-days, and retention pruning keeps the store within a storage budget.
+Qdrant is the sole long-term store.  The active day's short-term JSON is
+converted into durable memory records during handoff; once indexing succeeds,
+that operational file can be removed.  Retrieval returns query-relevant,
+ranked context for model prompts.
 """
-
 from __future__ import annotations
 
 import hashlib

backend/src/auth/__init__.py

@@ -1,9 +0,0 @@
-"""auth — admin authentication package for the Valhalla web dashboard.
-
-Exposes the session manager (manager.py) and HTTP routes (routes.py) that
-protect simulation-control and roster endpoints in Odin.py.
-
-Design: viewers can watch the simulation unauthenticated; only control
-endpoints require a session.
-"""
-

backend/src/auth/manager.py

@@ -1,12 +1,9 @@
-"""manager — in-memory admin session authentication.
-
-Loads email:password pairs from ADMIN_CREDENTIALS, hashes with salted
-scrypt, issues 24-hour bearer tokens, and validates/revokes them.
+"""
+In-memory session-based authentication for Valhalla web admin.
 
-Architecture: used by auth/routes.py to guard all sim-control and roster
-endpoints in Odin.py.
-Design: in-memory sessions are intentionally simple (lost on restart);
-when no credentials are configured, login is disabled with a warning.
+Admin credentials come from the ADMIN_CREDENTIALS env var as
+email:password pairs separated by semicolons. Sessions are stored
+in-memory (lost on server restart).
 """
 
 import os

backend/src/auth/routes.py

@@ -1,11 +1,5 @@
-"""routes — HTTP endpoints for admin login, logout, and session check.
-
-Mounts POST /api/auth/login, POST /api/auth/logout, GET /api/auth/me on
-the FastAPI app, backed by auth/manager.py.
-
-Architecture: a thin transport layer between the React dashboard and the
-session store; consumed by frontend/src/hooks/useAuth.jsx.
-Design: tokens travel as bearer headers; no cookie handling.
+"""
+Auth API routes — login, logout, session check.
 """
 
 from fastapi import APIRouter, HTTPException, Header

backend/src/config.py

@@ -1,13 +1,7 @@
-"""config — single source of truth for every tunable in the simulation.
-
-Defines paths, clock parameters (tick length, speed), perception and
-conversation radii, replan and budget caps, memory and LLM settings, and
-API key handling, with precedence CLI > environment > built-in default.
-
-Architecture: imported by virtually every module; never imports other
-project modules, so it can be loaded without side effects.
-Design: all settings are overridable via SIM_* environment variables so
-experiments can vary parameters without code changes.
+"""
+Project-wide path configuration.
+Resolves the project root, backend, frontend, data, and output directories
+so all modules can reference consistent paths.
 """
 
 from pathlib import Path
@@ -184,15 +178,6 @@ MEMORY_STORAGE_PRUNE_TARGET = min(MEMORY_STORAGE_PRUNE_THRESHOLD, max(0.05, _env
 # Cap on full day-plan regenerations triggered mid-day per agent (budget guard).
 MAX_REPLANS_PER_AGENT_PER_DAY = _env_int("SIM_MAX_REPLANS_PER_AGENT_PER_DAY", 3)
 
-# Deterministic backstop for agents stranded on the "Unscheduled downtime"
-# fallback schedule.  Every tick, the engine checks remaining plans; an
-# agent with upcoming downtime gets a remaining-day replan.  The cooldown
-# (in ticks) stops a repeatedly-rejected replan from hammering the planner,
-# and the horizon (in minutes) skips replans when too little of the day is
-# left to be worth one.
-DOWNTIME_REPLAN_COOLDOWN_TICKS = _env_int("SIM_DOWNTIME_REPLAN_COOLDOWN_TICKS", 60)
-DOWNTIME_REPLAN_MIN_HORIZON = _env_int("SIM_DOWNTIME_REPLAN_MIN_HORIZON", 60)
-
 # Budget governor: soft ceiling on LLM calls per real hour across the whole sim.
 # 0 = no ceiling. When exceeded, cognition degrades gracefully (skip reflex,
 # defer replans) — the sim keeps running on the 0-LLM executor path.
@@ -219,20 +204,12 @@ SIM_CREATIVITY = min(1.0, max(0.0, _env_float("SIM_CREATIVITY", 1.0)))
 # independent from creativity: an observer can ask for more varied plans
 # without making students' energy and mood unrealistically volatile.
 SIM_WELLBEING_VARIABILITY = min(1.0, max(0.0, _env_float("SIM_WELLBEING_VARIABILITY", 0.75)))
-# How fast the runtime glides an agent's energy toward the day planner's
-# declared per-action energy_target.  This is a control rate, not an energy
-# value: 0.0 disables the glide (pure delta-based energy, as before), larger
-# values converge faster (0.03 => ~84% of the gap closed per 60-min action).
-SIM_ENERGY_FOLLOW_RATE = _env_float("SIM_ENERGY_FOLLOW_RATE", 0.03)
 TEMPERATURE = 0.7 + (0.4 * SIM_CREATIVITY)  # planning and decisions: 1.1 at lively
 CONVERSATION_TEMPERATURE = 0.6 + (0.4 * SIM_CREATIVITY)  # 1.0 at lively
 SUMMARY_TEMPERATURE = 0.5
-# The simulation uses a primary model, with a same-key fallback model.
-# ``gemini_client`` tries the primary model on each key, then the fallback
-# model on the same key, before advancing to the next key; that key
-# rotation is the only provider-recovery behaviour.
-GEMINI_MODEL = _env_str("SIM_GEMINI_MODEL", "gemini-3.5-flash-lite")
-GEMINI_MODEL_FALLBACK = _env_str("SIM_GEMINI_MODEL_FALLBACK", "gemini-3.1-flash-lite")
+# The simulation intentionally uses one model.  Key traversal, implemented in
+# ``gemini_client``, is the only provider recovery behaviour.
+GEMINI_MODEL = _env_str("SIM_GEMINI_MODEL", "gemini-3.1-flash-lite")
 
 # Support multiple API keys (comma-separated in env var). When numbered
 # variables are used, they are read in ascending numeric order.
@@ -305,7 +282,6 @@ _OVERRIDE_MAP = {
     "decide_cooldown_ticks": "DECIDE_COOLDOWN_TICKS",
     "conversation_min_energy": "CONVERSATION_MIN_ENERGY",
     "conversation_min_emotion": "CONVERSATION_MIN_EMOTION",
-    "energy_follow_rate": "SIM_ENERGY_FOLLOW_RATE",
     "day_handoff_conversation_timeout_seconds": "DAY_HANDOFF_CONVERSATION_TIMEOUT_SECONDS",
 }
 
@@ -374,11 +350,10 @@ def describe_settings() -> str:
     return (
         "Valhalla simulation settings\n"
         f"  API keys loaded              : {API_KEY_COUNT} (head resets to index 1/call)\n"
-        f"  Gemini model                 : {GEMINI_MODEL} (fallback {GEMINI_MODEL_FALLBACK} on same key)\n"
+        f"  Gemini model                 : {GEMINI_MODEL}\n"
         f"  Simulation creativity        : {SIM_CREATIVITY:.2f} "
         f"(plan/decision {TEMPERATURE:.2f}, conversation {CONVERSATION_TEMPERATURE:.2f}, summary {SUMMARY_TEMPERATURE:.2f})\n"
         f"  Wellbeing variation           : {SIM_WELLBEING_VARIABILITY:.2f}\n"
-        f"  Energy follow rate            : {SIM_ENERGY_FOLLOW_RATE:.3f}/sim-min\n"
         f"  Memory backend               : {MEMORY_BACKEND}\n"
         f"  Semantic memory              : {'ON' if SEMANTIC_MEMORY_ENABLED else 'OFF'}\n"
         f"  Perception                   : {'ON' if PERCEPTION_ENABLED else 'OFF'} (radius {PERCEPTION_RADIUS_PX}px)\n"

backend/src/core/agent_registry.py

@@ -1,13 +1,12 @@
-"""agent_registry — single source of truth for every agent's runtime state.
+"""
+Agent Registry — single source of truth for every agent's runtime state.
 
-Holds one AgentRuntimeState per agent (persona, position, day plan,
-wellbeing, conversation state, replan budget) and the AgentRegistry that
-the engine, Actions, conversations, and memory all read and write.
+The WorldEngine owns one `AgentRegistry` instance. All modules (Actions,
+conversation, day_planner, Short_term) read from and write to it through
+the engine — never directly.
 
-Architecture: central state container consumed by every subsystem; the
-engine syncs it to WorldState each tick.
-Design: consolidates position into one registry to eliminate the
-dual-source position drift of earlier versions.
+This replaces the dual-source problem where Actions.py had its own
+position/action and WorldState had a separate copy that drifted.
 """
 
 from __future__ import annotations

backend/src/core/budget.py

@@ -1,14 +1,20 @@
-"""budget — LLM call budget governor.
-
-Tracks all LLM calls in a rolling one-real-hour window and answers
-can_afford(kind, cost) so cognitive steps degrade gracefully when a soft
-hourly ceiling is exceeded; the simulation keeps running on its 0-LLM
-executor path.
-
-Architecture: a process-wide singleton (GOVERNOR) consulted by the
-engine's decide/replan phases and by gemini_client.py.
-Design: cost weights reflect relative expense (decide=1, replan=4); the
-ceiling is opt-in (SIM_LLM_HOURLY_CEILING, default 0 = unlimited).
+"""
+LLM budget governor -- one place that knows how much LLM spend has happened
+recently and whether the simulation can afford more.
+
+Why this exists
+---------------
+Free-tier Gemini keys are scarce (a handful of keys, a few requests/minute
+each). Turning on perception + proximity conversations + a decision-making
+brain could, if left ungated, burn the whole quota in minutes. Every
+cognitive call site (day planner, conversation, reflex escalation) asks the
+governor `can_afford()` before spending, and calls `record()` after. When the
+soft ceiling is exceeded the governor says "no", and the caller degrades
+gracefully -- the simulation keeps running on its 0-LLM executor path.
+
+The governor is intentionally simple: a rolling one-real-hour window of call
+timestamps, plus lifetime counters for observability (used by the budget
+stress test and the on-screen/logged stats).
 """
 
 from __future__ import annotations

backend/src/core/checkpoint_manager.py

@@ -1,13 +1,14 @@
-"""checkpoint_manager — per-tick world state persistence.
+"""
+Checkpoint Manager — per-tick state save/load for crash recovery.
 
-Saves WorldState, AgentRegistry (including the serialized action manager),
-engine-owned state, and Python RNG state to gzip JSON after every tick,
-and loads them for resume/rewind; prunes to a one-day window.
+Saves WorldState + AgentRegistry after every tick to compressed
+``backend/data/checkpoints/tick_{00001}.json.gz`` files.
 
-Architecture: called by WorldEngine each tick and at day handoff; enables
-the server's resume/rewind endpoints and the --resume-checkpoint flag.
-Design: atomic writes (.tmp + rename); saved RNG state makes collision
-resolution replay identically.
+Supports:
+  - Save: full simulation state as JSON
+  - Load: reconstruct from any saved tick
+  - List: available checkpoint ticks
+  - Prune: auto-delete old checkpoints, keep last N
 """
 
 from __future__ import annotations

backend/src/core/log.py

@@ -1,13 +1,27 @@
-"""log — centralized logging setup for the whole project.
+"""
+Centralized logging setup for the whole project. Every module imports
+`get_logger(__name__)` from here instead of calling `logging.basicConfig`
+or `logging.getLogger` directly, so the format + destinations stay
+identical everywhere
+
+Usage
+-----
+Call `setup_logging()` ONCE, as early as possible in the process (top of
+whatever your real entrypoint is -- backend/main.py, or the `run()` /
+`__main__` block of a standalone script):
+
+    from src.core.log import setup_logging, get_logger
+    setup_logging(run_id="run_name")   # run_id is optional
+    logger = get_logger(__name__)
 
-Every module imports get_logger(__name__) here so format, destinations,
-and level stay consistent across the engine, planner, and server.
+Every other module then just does:
 
-Architecture: a dependency of every backend module; no business logic.
-Design: logs land in backend/output/logs with timestamps for post-hoc
-analysis of long runs. Console output is off by default — the admin log
-terminal (src/core/log_relay.py) mirrors every line to the live log file
-and the admin-only /api/logs endpoints instead.
+    from src.core.log import get_logger
+    logger = get_logger(__name__)
+
+If some module gets imported/used before setup_logging() runs (import
+order accidents happen), get_logger() will lazily call setup_logging()
+with defaults so you still get sane output instead of silence.
 """
 
 from __future__ import annotations
@@ -39,7 +53,7 @@ def _build_log_path(run_id: Optional[str] = None) -> Path:
 def setup_logging(
         level: int = DEFAULT_LEVEL,
         run_id: Optional[str] = None,
-        console: bool = False,
+        console: bool = True,
         file: bool = True,
         max_bytes: int = 5 * 1024 * 1024,
         backup_count: int = 5,
@@ -55,8 +69,7 @@ def setup_logging(
     run_id        -- optional tag folded into the log filename, e.g. a
                       persona name or simulation id, so a run's logs are
                       easy to find in output/logs/
-    console       -- also stream logs to stdout (default off; watch the
-                      admin log terminal instead)
+    console       -- also stream logs to stdout
     file          -- also write logs to output/logs/<timestamp>[_<run_id>].log
     max_bytes /
     backup_count  -- rotation settings for the file handler
@@ -116,14 +129,6 @@ def setup_logging(
     logging.config.dictConfig(config)
     _configured = True
 
-    # Mirror every emitted line into the admin log terminal's buffer and the
-    # live log file, regardless of the console/file toggles above.
-    try:
-        from src.core.log_relay import install_relay
-        install_relay()
-    except Exception:
-        pass
-
     if log_path:
         logging.getLogger(__name__).info("[log] logging initialized -> %s", log_path)
 

backend/src/core/log_relay.py

@@ -1,82 +0,0 @@
-"""log_relay — mirror of every project log line for the admin log terminal.
-
-Attaches a second handler to the ROOT logger: each emitted record is
-formatted with the project's standard format and appended to a bounded
-in-memory ring buffer (polled by the admin-only /api/logs endpoints) and
-to a live file backend/output/logs/live.log, so the same output also
-survives restarts and stays viewable without the Space console.
-
-Architecture: installed once by src/core/log.py setup_logging(); consumed
-by Odin.py's admin-gated endpoints and the frontend LogTerminal panel.
-Design: stdlib-only, a one-way mirror — never changes existing handlers,
-levels, or the file rotation policy.
-"""
-
-from __future__ import annotations
-
-import logging
-import threading
-from collections import deque
-from itertools import count
-
-from src.config import LOG_DIR
-from src.core.log import DATE_FORMAT, LOG_FORMAT
-
-MAX_LINES = 2000
-LIVE_LOG_PATH = LOG_DIR / "live.log"
-
-_seq = count(1)
-_buffer: deque = deque(maxlen=MAX_LINES)
-_latest_seq = 0
-_lock = threading.Lock()
-_installed = False
-
-
-class LogRelayHandler(logging.Handler):
-    """Appends each formatted record to the ring buffer and the live file."""
-
-    def __init__(self, level: int = logging.NOTSET) -> None:
-        super().__init__(level)
-        self._formatter = logging.Formatter(LOG_FORMAT, DATE_FORMAT)
-
-    def emit(self, record: logging.LogRecord) -> None:
-        global _latest_seq
-        try:
-            text = self._formatter.format(record)
-        except Exception:
-            text = f"{record.name} | {record.getMessage()}"
-        with _lock:
-            _latest_seq = next(_seq)
-            _buffer.append({"seq": _latest_seq, "level": record.levelname, "text": text})
-        try:
-            LOG_DIR.mkdir(parents=True, exist_ok=True)
-            with open(LIVE_LOG_PATH, "a", encoding="utf-8") as fh:
-                fh.write(text + "\n")
-        except OSError:
-            pass
-
-
-def install_relay() -> None:
-    """Attach the relay handler to the root logger exactly once."""
-    global _installed
-    if _installed:
-        return
-    logging.getLogger().addHandler(LogRelayHandler())
-    _installed = True
-
-
-def relay_lines(since: int = 0) -> dict:
-    """Return relayed entries with seq > since, plus the latest seq as cursor.
-
-    The cursor tracks the last emitted sequence, so polling stays
-    incremental even after the buffer is cleared.
-    """
-    with _lock:
-        lines = [entry for entry in _buffer if entry["seq"] > since]
-        return {"lines": lines, "next": _latest_seq}
-
-
-def clear_relay() -> None:
-    """Drop the in-memory buffer (the live file is intentionally kept)."""
-    with _lock:
-        _buffer.clear()

backend/src/core/perceive.py

@@ -1,12 +1,13 @@
-"""perceive — turns a frozen WorldSnapshot into what one agent can observe.
+"""
+Perceive -- turns a WorldSnapshot into what one agent can currently observe.
 
-Pure functions, no LLM, no side effects: builds nearest-first Observation
-lists in both tile (Chebyshev) and pixel (Euclidean) spaces.
+This is a pure function of snapshot data: no LLM calls, no side effects.
+With only one agent registered (your current single-agent phase), this
+naturally returns an empty list every tick -- no special-casing needed to
+"turn on" perception later, it already does the real spatial query.
 
-Architecture: used by the engine's perceive phase via snapshot
-agents_within_px; the novelty fingerprint that gates LLM decisions is
-computed from these observations.
-Design: strict purity keeps perception reproducible and cheap.
+Radius/distance logic itself lives on `WorldSnapshot` (core/snapshot.py) so
+there's exactly one implementation of "who's nearby" in the codebase. (in snapshot.py)
 """
 
 from __future__ import annotations

backend/src/core/runtime_health.py

@@ -1,14 +1,4 @@
-"""runtime_health — bounded per-tick anomaly detection for the live sim.
-
-Scans for stalled travel, overdue actions, position desyncs, and
-conversation timeouts; emits reports every N ticks or immediately on
-anomaly, feeding the dashboard debug panel and the sidecar monitor.
-
-Architecture: called by WorldEngine at the end of each tick; consumed by
-the frontend snapshot (health block).
-Design: state is O(agents) and history-free, so monitoring never grows
-with runtime.
-"""
+"""Bounded runtime health checks for a live Valhalla simulation."""
 
 from __future__ import annotations
 

backend/src/core/snapshot.py

@@ -1,12 +1,34 @@
-"""snapshot — immutable point-in-time view of WorldState.
+"""
+Snapshot -- point-in-time read-only view of WorldState (Immutable)
+
+Every tick takes exactly one `WorldSnapshot` and hands the
+*same* frozen object to every agent's tick graph via asyncio.gather(). That's
+what makes parallel agent decisions safe -- nobody is reading a WorldState
+that's being mutated mid-tick by someone else's action.
+
+`WorldSnapshot` is a deliberately separate class from `WorldState`, not just
+a deep copy of it. It exposes zero mutating methods, so there is no method
+an agent's perceive/react/plan code could accidentally call that would
+corrupt the resolve phase's assumptions. If you need a new read-only query
+(e.g. "what's the nearest free table"), add it here as a method on
+`WorldSnapshot` -- don't reach into `.agents`/`.occupancy` directly from
+perceive.py and reimplement the same query logic in multiple places.
+
+Usage
+-----
+    from src.core.world_state import WorldState, Position
+    from src.core.snapshot import take_snapshot
+
+    world = WorldState()
+    world.register_agent("gurnoor", Position(x=4, y=2, location_id="dorm_room_1"))
 
-The engine freezes exactly one WorldSnapshot per tick and hands the same
-object to every agent's parallel phase, so reads can never observe a
-mid-tick mutation.
+    snap = take_snapshot(world)          # take ONCE per tick, before decide phase
+    snap.get_agent("gurnoor")            # read-only query
+    snap.agents_near("gurnoor", radius=3)
+    snap.is_free("cafeteria_table_3")
 
-Architecture: consumed by perceive.py, the engine's observation builder,
-and the frontend snapshot projection.
-Design: frozen data structures; agents see the same world each tick.
+    # snap.tick = 999          <- raises, frozen model
+    # snap.agents["x"] = ...   <- raises, frozen model
 """
 
 from __future__ import annotations

backend/src/core/tick_graph.py

@@ -1,13 +1,20 @@
-"""tick_graph — per-agent LangGraph pipeline and standalone CLI debug tool.
+"""
+Agent -- per-agent LangGraph subgraph + standalone CLI debug tool.
+
+Production pipeline (used by WorldEngine via build_tick_graph):
+
+    perceive -> retrieve_memories -> react --[replan]--> day_planner -> write_back_memory
+                                              \\_[continue]__> keep_current /
+
+Only agents where scheduler.py's `agents_ready_for_decision()` returns
+True are invoked each tick -- mid-action agents are skipped entirely.
+
+Standalone CLI debug mode (python Agent.py <persona>):
 
-Defines the perceive -> retrieve_memories -> react -> day_planner ->
-write_back_memory subgraph and a day-planning CLI.
+    retrieve_memories from Short_term -> call day_planner.run() -> print plan table
 
-Architecture: the graph the engine was designed around; the production
-WorldEngine currently implements its own phase loop and calls brain.py
-directly, so this module serves as the reference single-agent pipeline.
-Design: kept as the canonical per-agent graph for experiments and
-debugging; its memory-stream integration is intentionally pluggable.
+Useful for testing a single persona's day plan without standing up the
+full tick loop.
 """
 
 from __future__ import annotations

backend/src/core/world_engine.py

@@ -1,15 +1,16 @@
-"""world_engine — the main simulation orchestrator.
-
-Owns the tick loop (snapshot, act, perceive, decide, replan, resolve),
-day handoff, proximity conversations, events, wellbeing updates, health
-observation, and per-tick checkpointing; also runs headless via CLI.
-
-Architecture: the hub of the backend — every subsystem (brain, planner,
-Actions, memory, events, checkpoints) is called from here; Odin.py hosts
-it as a background task.
-Design: LLM calls run in worker threads and asyncio tasks so latency
-never freezes the clock; deterministic phases (resolve) and stochastic
-phases (decide) are strictly separated.
+"""
+WorldEngine — the main simulation orchestrator.
+
+Controls the tick loop: advances time, runs agent actions in parallel,
+detects proximity for conversations, handles end-of-day transitions,
+and keeps WorldState in sync with the agent registry.
+
+Usage:
+    from src.core.world_engine import WorldEngine
+
+    engine = WorldEngine()
+    await engine.initialize()
+    await engine.run(max_ticks=1440)   # one full day at 1 tick/sec
 """
 
 from __future__ import annotations
@@ -17,7 +18,6 @@ from __future__ import annotations
 import asyncio
 import hashlib
 import json
-import math
 import random
 import sys
 import time as _time
@@ -126,10 +126,6 @@ class WorldEngine:
         # Decisions are advisory; a slow provider response must not freeze the
         # simulation clock or WebSocket snapshots at an action boundary.
         self._decision_tasks: Dict[str, asyncio.Task] = {}
-        # Advisory throttle: last tick an "unscheduled downtime" recovery
-        # replan was attempted per agent.  Not checkpointed — on restore a
-        # fresh attempt is harmless.
-        self._downtime_replan_tick: Dict[str, int] = {}
 
     @staticmethod
     def _conversation_key(first_id: str, second_id: str) -> str:
@@ -353,41 +349,69 @@ class WorldEngine:
                 baseline -= 0.04
         return max(0.56, min(0.86, baseline))
 
-    def _action_wellbeing_deltas(self, state: AgentRuntimeState, action: Any, duration: int) -> tuple[float, float]:
-        """Return the total wellbeing effect for one action.
-
-        When the day planner declares an energy_target for the action, the
-        runtime glides the agent's energy from its current level toward that
-        declared cumulative target (the LLM owns every value; this is only a
-        smooth, deterministic path to it).  The exponential progress factor
-        means short actions barely move energy while long ones converge, and
-        because it depends only on stored state and the plan it stays
-        checkpoint-reproducible.
+    def _action_wellbeing_deltas(self, state: AgentRuntimeState, action: Any) -> tuple[float, float]:
+        """Compute a deterministic total wellbeing effect for one action.
 
-        Without a declared target the legacy delta path applies: the planner's
-        energy_change drives the level, with the small deterministic jitter
-        keyed by agent/action so replays remain reproducible.
+        LLM-supplied deltas are useful hints, but are normally very small.  A
+        shared local activity model therefore gives classes, travel, rest, and
+        social time their ordinary human cost or benefit.  The small stable
+        variation is keyed by agent/action, rather than sampled each tick, so
+        replaying a checkpoint remains reproducible.
         """
         description = (getattr(action, "description", "") or "").lower()
-        emotion = float(getattr(action, "emotion_change", 0.0) or 0.0)
-
-        target = getattr(action, "energy_target", None)
-        if target is not None:
-            target = max(0.0, min(1.0, float(target)))
-            remaining = target - state.energy_level
-            progress = 1.0 - math.exp(-_cfg.SIM_ENERGY_FOLLOW_RATE * max(1, duration))
-            energy = remaining * progress
-            return energy, emotion
-
-        energy = float(getattr(action, "energy_change", 0.0) or 0.0)
+        action_type = str(getattr(action, "action_type", "")).lower()
+        # The planner can add personality-specific flavour, but it must not
+        # turn an otherwise restorative meal or quiet break into a day-long
+        # energy drain.  The local physical activity model is authoritative.
+        planner_energy = max(-0.08, min(0.08, float(getattr(action, "energy_change", 0.0))))
+        planner_emotion = max(-0.12, min(0.12, float(getattr(action, "emotion_change", 0.0))))
+        energy, emotion = 0.0, 0.0
+
+        if action_type.endswith("move") or any(word in description for word in ("walk", "travel", "commute", "go to")):
+            energy, emotion = -0.075, -0.008
+        elif "sleep" in description:
+            energy, emotion = 0.50, 0.025
+        elif any(word in description for word in ("nap", "rest", "recharge", "lie down")):
+            energy, emotion = 0.20, 0.020
+        elif any(word in description for word in (
+            "meme", "memes", "scroll", "social media", "youtube", "video",
+            "reading for pleasure", "quiet reading", "reading quietly", "bench", "downtime",
+            "free time", "relax", "relaxing", "wind-down", "wind down",
+        )):
+            energy, emotion = 0.090, 0.025
+        elif any(word in description for word in ("class", "lecture", "lab", "tutorial", "study", "assignment", "coding", "project", "exam")):
+            energy, emotion = -0.070, -0.025
+        elif any(word in description for word in ("gym", "sport", "run", "football", "basketball", "badminton", "workout", "cardio", "weightlift", "training")):
+            energy, emotion = -0.180, 0.075
+        elif any(word in description for word in ("breakfast", "lunch", "dinner", "meal", "food", "tea", "chai", "eat", "eating")):
+            energy, emotion = 0.130, 0.025
+        elif any(word in description for word in ("friends", "club", "music", "open mic", "game", "movie", "social", "hangout")):
+            energy, emotion = 0.015, 0.075
+        elif any(word in description for word in ("laundry", "clean", "errand", "admin", "queue", "chore")):
+            energy, emotion = -0.080, -0.025
+        elif any(word in description for word in ("stand", "standing", "wait", "waiting")):
+            energy, emotion = -0.040, -0.005
+        else:
+            # Neutral, seated or low-intensity tasks should not silently push
+            # every agent toward exhaustion merely because their wording was
+            # not anticipated above.
+            energy, emotion = -0.005, 0.0
+
+        # Introverted students generally enjoy a good conversation but spend
+        # more energy on it; this keeps personality visible without judging it.
+        traits = " ".join(str(state.persona.get(key, "")) for key in ("innate", "lifestyle", "learned")).lower()
+        if any(word in description for word in ("friends", "club", "social", "hangout")) and any(
+            marker in traits for marker in ("introverted", "quiet", "reserved")
+        ):
+            energy -= 0.03
 
         variation = _cfg.SIM_WELLBEING_VARIABILITY
         token = f"{state.agent_id}|{getattr(action, 'start_time', '')}|{getattr(action, 'end_time', '')}|{description}"
         digest = hashlib.blake2s(token.encode("utf-8"), digest_size=4).digest()
         jitter = (int.from_bytes(digest, "big") / 0xFFFFFFFF) * 2.0 - 1.0
-        energy += jitter * 0.035 * variation
-        emotion += jitter * 0.045 * variation
-        return energy, emotion
+        energy += planner_energy + jitter * 0.035 * variation
+        emotion += planner_emotion + jitter * 0.045 * variation
+        return max(-0.28, min(0.30, energy)), max(-0.18, min(0.16, emotion))
 
     def _memory_context(self, persona_name, persona, before_date=None, query_hint=""):
         """Build (relevant_memories, rolling_summary) for a day-planner call.
@@ -508,8 +532,6 @@ class WorldEngine:
                             "persona_name": name,
                             "mode": "full_day",
                             "current_location_id": hostel,
-                            "energy_level": self._energy_baseline(persona),
-                            "emotion_state": self._emotion_baseline(persona),
                             "upcoming_events": self.event_manager.snapshot(self.sim_start_date, self.sim_start_hhmm).get("upcoming", []),
                         },
                     ),
@@ -679,26 +701,6 @@ class WorldEngine:
                 return_exceptions=True,
             )
 
-        # ══════ PHASE 5b: Unscheduled downtime recovery (parallel, deterministic) ══════
-        # A force-accepted fallback plan strands the agent on "Unscheduled
-        # downtime" for the rest of the day.  The LLM decide path may never
-        # fire for such an agent, so detect it here and replan the remaining
-        # whole day explicitly.
-        downtime_agents = [
-            s for s in agent_states if self._has_unscheduled_downtime(s, current_tick)
-        ]
-        if downtime_agents:
-            for s in downtime_agents:
-                self._downtime_replan_tick[s.agent_id] = current_tick
-                logger.info(
-                    "[WorldEngine] agent '%s' stuck on unscheduled downtime — replanning remaining day",
-                    s.persona_name,
-                )
-            await asyncio.gather(
-                *[self._phase_replan(s, current_tick, hhmm) for s in downtime_agents],
-                return_exceptions=True,
-            )
-
         # ══════ PHASE 6: Resolve (sequential) ══════
         await self._check_last_action_triggers(current_tick, hhmm)
         self._apply_finished_event_effects(self.sim_start_date, hhmm)
@@ -885,8 +887,6 @@ class WorldEngine:
                         "persona_name": state.persona_name,
                         "mode": "remaining",
                         "current_location_id": state.position.location_id,
-                        "energy_level": state.energy_level,
-                        "emotion_state": state.emotion_state,
                         "upcoming_events": self.event_manager.snapshot(self.sim_start_date, hhmm).get("upcoming", []),
                     },
                 ),
@@ -913,38 +913,6 @@ class WorldEngine:
                 "[WorldEngine] replan failed for '%s': %s", state.persona_name, e,
             )
 
-    def _has_unscheduled_downtime(self, state: AgentRuntimeState, tick: int) -> bool:
-        """Detect agents stranded on the deterministic fallback schedule.
-
-        The LLM decide path is gated (novelty, energy/emotion, cooldown,
-        budget), so a force-accepted fallback day can leave an agent stuck on
-        "Unscheduled downtime" for hours with no replan ever firing.  This
-        backstop scans the remaining plan every tick and flags it."""
-        if state.paused:
-            return False
-        if state.day_archived:
-            return False
-        if state.manager is None:
-            return False
-        if state.replan_count >= _cfg.MAX_REPLANS_PER_AGENT_PER_DAY:
-            return False
-        if (
-            tick - self._downtime_replan_tick.get(state.agent_id, -10**9)
-            < _cfg.DOWNTIME_REPLAN_COOLDOWN_TICKS
-        ):
-            return False
-        now_minutes = tick % (24 * 60)
-        if now_minutes >= 24 * 60 - _cfg.DOWNTIME_REPLAN_MIN_HORIZON:
-            # Too little of the day remains to justify a replan.
-            return False
-        for action in state.day_plan:
-            end = self._hhmm_to_minutes(str(action.get("end", "")))
-            if end > now_minutes and "unscheduled downtime" in str(
-                action.get("action", "")
-            ).lower():
-                return True
-        return False
-
     async def _run_agent_act(
         self, state: AgentRuntimeState, tick: int, hhmm: str
     ) -> None:
@@ -1064,11 +1032,15 @@ class WorldEngine:
             end_min = self._hhmm_to_minutes(action.end_time)
             duration = max(1, end_min - start_min)
             tick_step = _cfg.SIM_MINUTES_PER_TICK
-            action_energy_change, action_emotion_change = self._action_wellbeing_deltas(state, action, duration)
+            action_energy_change, action_emotion_change = self._action_wellbeing_deltas(state, action)
             energy_tick = (action_energy_change / duration) * tick_step
             emotion_tick = (action_emotion_change / duration) * tick_step
-            state.energy_level = max(0.0, min(1.0, state.energy_level + energy_tick))
-            state.emotion_state = max(0.0, min(1.0, state.emotion_state + emotion_tick))
+            state.energy_level = max(0.08, min(0.97, state.energy_level + energy_tick))
+            baseline = state.emotion_baseline
+            # Mood has a weak pull towards personality baseline, but day
+            # events are allowed to remain visible for several actions.
+            recovery = (baseline - state.emotion_state) * min(0.015, 0.0005 * tick_step)
+            state.emotion_state = max(0.10, min(0.90, state.emotion_state + emotion_tick + recovery))
         except Exception:
             pass
 
@@ -1511,14 +1483,22 @@ class WorldEngine:
         self.relationship_matrix.update(b.agent_id, a.agent_id, conv_result.relationship_delta)
         self.relationship_matrix.save()
         # Conversations affect the people having them, not only their stored
-        # relationship score.  The LLM decides each participant's net energy
-        # and mood change for the chat; only a 0..1 safety clamp is applied.
-        for state, energy_delta, emotion_delta in (
-            (a, conv_result.energy_delta_a, conv_result.emotion_delta_a),
-            (b, conv_result.energy_delta_b, conv_result.emotion_delta_b),
-        ):
-            state.energy_level = max(0.0, min(1.0, state.energy_level + float(energy_delta or 0.0)))
-            state.emotion_state = max(0.0, min(1.0, state.emotion_state + float(emotion_delta or 0.0)))
+        # relationship score.  A warm chat is a modest lift; an awkward one is
+        # draining.  The effect is applied once per completed conversation.
+        relationship_delta = max(-0.20, min(0.20, conv_result.relationship_delta))
+        sentiment = (getattr(conv_result, "sentiment", "neutral") or "neutral").lower()
+        for state in (a, b):
+            social_cost = 0.045 if any(marker in " ".join(
+                str(state.persona.get(key, "")) for key in ("innate", "lifestyle", "learned")
+            ).lower() for marker in ("introverted", "quiet", "reserved")) else 0.025
+            state.energy_level = max(0.08, min(0.97, state.energy_level - social_cost))
+            if sentiment in ("positive", "warm", "friendly"):
+                mood_delta = 0.035 + max(0.0, relationship_delta) * 0.25
+            elif sentiment in ("negative", "tense", "awkward"):
+                mood_delta = -0.035 + min(0.0, relationship_delta) * 0.25
+            else:
+                mood_delta = relationship_delta * 0.08
+            state.emotion_state = max(0.10, min(0.90, state.emotion_state + mood_delta))
         logger.info(
             "[WorldEngine] conversation '%s' <-> '%s' active until tick %d",
             a.persona_name, b.persona_name, self.world.tick + conv_result.duration_minutes,
@@ -1546,8 +1526,6 @@ class WorldEngine:
                     "persona_name": state.persona_name,
                     "mode": "remaining",
                     "current_location_id": state.position.location_id,
-                    "energy_level": state.energy_level,
-                    "emotion_state": state.emotion_state,
                 },
             )
             return plan_result.get("day_plan", [])
@@ -1802,10 +1780,7 @@ class WorldEngine:
                 f"The previous day ended while the agent was {action_text} at {location}. "
                 f"Energy is {state.energy_level:.2f}/1.0 and emotion is "
                 f"{state.emotion_state:.2f}/1.0. Continue naturally from this "
-                "physical and emotional state; do not abruptly relocate them. "
-                "When assigning energy_change/emotion_change for the new day, "
-                "keep the cumulative energy and mood totals between 0.0 and "
-                "1.0 at all times."
+                "physical and emotional state; do not abruptly relocate them."
             )
 
         async def _plan_next_day(state: AgentRuntimeState) -> tuple[AgentRuntimeState, list]:
@@ -1828,8 +1803,6 @@ class WorldEngine:
                         "mode": "next_day",
                         "current_location_id": state.position.location_id,
                         "handoff_context": _handoff_context(state),
-                        "energy_level": state.energy_level,
-                        "emotion_state": state.emotion_state,
                         "upcoming_events": self.event_manager.snapshot(next_date, "00:00").get("upcoming", []),
                     },
                 )
@@ -1858,7 +1831,6 @@ class WorldEngine:
         self._recent_convs.clear()
         self._in_range.clear()
         self._last_decision_tick.clear()
-        self._downtime_replan_tick.clear()
         self._last_obs.clear()
         self._tick_observations.clear()
         self._applied_event_effects.clear()

backend/src/core/world_events.py

@@ -1,13 +1,10 @@
-"""world_events — deterministic, LLM-free campus events.
+"""Deterministic, data-driven campus events and safe plan opportunities.
 
-Loads the event calendar, decides attendance from persona interest, social
-score, and seeded noise, splices events into flexible plan windows, and
-applies wellbeing/relationship effects once events end.
-
-Architecture: called by WorldEngine at init, day handoff, and restore;
-consumes places/personas/relationship matrix and edits day plans.
-Design: events only replace entirely flexible time windows (never classes
-or sleep), keeping the calendar safe to apply automatically.
+The event calendar is deliberately independent of the LLM.  It makes an
+attendance decision from a persona, its social context, schedule conflicts,
+and a seeded tie-breaker, then edits only an entirely-flexible time window.
+This keeps festivals and interruptions lively without allowing them to erase
+classes, meals, sleep, exams, or an in-progress route.
 """
 
 from __future__ import annotations

backend/src/core/world_state.py

@@ -1,13 +1,39 @@
-"""world_state — the canonical mutable world.
+"""
+World State -- the single source of truth for the simulation.
+
+`WorldState` holds everything that is true about the world at a given tick:
+where every agent is, what they're currently doing, and who/what currently
+holds any contested resource (a chair, an NPC's attention, a location slot).
+
+Ownership rule:
+    Only `WorldEngine`'s resolve phase should ever call the mutating methods
+    on this class directly (`set_agent_action`, `move_agent`, `occupy`, ...).
+
+    Every agent tick graph (perceive -> retrieve -> react -> day_planner ->
+    act) must only ever see a frozen copy produced by `core/snapshot.py`.
+
+    That separation is what keeps the decide phase safely parallelizable
+    with asyncio.gather() -- nobody is reading a WorldState that something
+    else is mutating mid-tick.
+
+Usage
+-----
+    from src.core.world_state import WorldState, Position
+
+    world = WorldState()
+    world.register_agent("gurnoor", Position(x=4, y=2, location_id="dorm_room_1"))
+    world.register_resource("cafeteria_table_3")
 
-Holds agent positions, occupancy, append-only history, the absolute tick
-clock, and conversation cooldowns; only the engine's resolve phase may
-mutate it.
+    # inside the resolve phase, after an agent's tick graph proposed an action:
+    world.occupy("cafeteria_table_3", "gurnoor")
+    world.set_agent_action("gurnoor", CurrentAction(
+        description="eating breakfast",
+        start_tick=world.tick,
+        end_tick=world.tick + 20,
+        target_object_id="cafeteria_table_3",
+    ))
 
-Architecture: the bottom layer under snapshot.py (read view), checkpoint
-manager, and the engine's sync step.
-Design: mutation is restricted by convention — the engine mirrors the
-registry into WorldState exactly twice per tick.
+    world.advance_tick(minutes=10)
 """
 
 from __future__ import annotations

backend/src/llm/gemini_client.py

@@ -1,15 +1,8 @@
-"""gemini_client — the project's only LLM access point.
-
-Wraps Google Gemini with deterministic head-first key rotation (each call
-tries every configured key exactly once), JSON-schema structured output,
-embedding support, and a provider_failure circuit breaker that halts the
-simulation cleanly when every key fails.
-
-Architecture: called by day_planner.py, brain.py, conversation.py,
-vector_memory.py, and the roster generator; records spend with the budget
-governor.
-Design: deliberately no retries or timeouts in this module — resilience
-lives in the engine's checkpoint/resume path and the budget governor.
+"""Gemini access through a deterministic, head-first key ring.
+
+Every API call starts with the first configured key.  On *any* exception it
+tries the next node exactly once.  There are deliberately no delays, retries,
+timeouts, cooldowns, key reservations, or model changes in this module.
 """
 
 from __future__ import annotations
@@ -23,13 +16,7 @@ from google import genai
 from google.genai import types
 from pydantic import BaseModel
 
-from src.config import (
-    API_KEYS,
-    GEMINI_MODEL,
-    GEMINI_MODEL_FALLBACK,
-    MEMORY_EMBEDDING_MODEL,
-    TEMPERATURE,
-)
+from src.config import API_KEYS, GEMINI_MODEL, MEMORY_EMBEDDING_MODEL, TEMPERATURE
 from src.core.log import get_logger
 
 logger = get_logger(__name__)
@@ -136,29 +123,28 @@ def call_gemini(
     complexity: str = "default",
     temperature: float = TEMPERATURE,
 ) -> BaseModel:
-    """Try both configured models on each key before advancing the ring."""
+    """Call exactly one model, moving through the ring on any failure."""
     errors: list[Exception] = []
     for node in _new_ring().traverse_from_head():
-        for model in (GEMINI_MODEL, GEMINI_MODEL_FALLBACK):
-            try:
-                response = _get_client(node.key).models.generate_content(
-                    model=model,
-                    contents=user_prompt,
-                    config=types.GenerateContentConfig(
-                        system_instruction=system_prompt,
-                        response_mime_type="application/json",
-                        response_schema=schema,
-                        temperature=temperature,
-                        thinking_config=types.ThinkingConfig(thinking_level="medium"),
-                    ),
-                )
-                result = response.parsed if getattr(response, "parsed", None) is not None else schema.model_validate(json.loads(response.text))
-                logger.info("[gemini] model=%s key_index=%d ok", model, node.index)
-                _record_success(complexity)
-                return result
-            except Exception as exc:
-                errors.append(exc)
-                logger.warning("[gemini] model=%s key_index=%d failed (%s); advancing", model, node.index, type(exc).__name__)
+        try:
+            response = _get_client(node.key).models.generate_content(
+                model=GEMINI_MODEL,
+                contents=user_prompt,
+                config=types.GenerateContentConfig(
+                    system_instruction=system_prompt,
+                    response_mime_type="application/json",
+                    response_schema=schema,
+                    temperature=temperature,
+                    thinking_config=types.ThinkingConfig(thinking_level="medium"),
+                ),
+            )
+            result = response.parsed if getattr(response, "parsed", None) is not None else schema.model_validate(json.loads(response.text))
+            logger.info("[gemini] model=%s key_index=%d ok", GEMINI_MODEL, node.index)
+            _record_success(complexity)
+            return result
+        except Exception as exc:
+            errors.append(exc)
+            logger.warning("[gemini] model=%s key_index=%d failed (%s); advancing", GEMINI_MODEL, node.index, type(exc).__name__)
     raise _quota_exhausted(GEMINI_MODEL, errors)
 
 

backend/test_gemini_client.py

@@ -1,60 +0,0 @@
-from pydantic import BaseModel
-
-from src.llm import gemini_client
-
-
-class _Result(BaseModel):
-    value: str
-
-
-class _Response:
-    parsed = _Result(value="ok")
-
-
-class _Models:
-    def __init__(self, calls, failures):
-        self._calls = calls
-        self._failures = failures
-
-    def generate_content(self, *, model, **_kwargs):
-        self._calls.append(model)
-        if model in self._failures:
-            raise RuntimeError(model)
-        return _Response()
-
-
-class _Client:
-    def __init__(self, calls, failures):
-        self.models = _Models(calls, failures)
-
-
-def test_fallback_model_runs_on_same_key_before_next_key(monkeypatch):
-    calls = []
-    monkeypatch.setattr(gemini_client, "API_KEYS", ["key-1", "key-2"])
-    monkeypatch.setattr(gemini_client, "GEMINI_MODEL", "primary")
-    monkeypatch.setattr(gemini_client, "GEMINI_MODEL_FALLBACK", "fallback")
-    monkeypatch.setattr(gemini_client, "_get_client", lambda key: _Client(calls, {"primary"}))
-    monkeypatch.setattr(gemini_client, "_record_success", lambda _complexity: None)
-
-    result = gemini_client.call_gemini("system", "user", _Result)
-
-    assert result.value == "ok"
-    assert calls == ["primary", "fallback"]
-
-
-def test_key_advances_only_after_both_models_fail(monkeypatch):
-    calls = []
-    monkeypatch.setattr(gemini_client, "API_KEYS", ["key-1", "key-2"])
-    monkeypatch.setattr(gemini_client, "GEMINI_MODEL", "primary")
-    monkeypatch.setattr(gemini_client, "GEMINI_MODEL_FALLBACK", "fallback")
-    monkeypatch.setattr(
-        gemini_client,
-        "_get_client",
-        lambda key: _Client(calls, {"primary", "fallback"} if key == "key-1" else set()),
-    )
-    monkeypatch.setattr(gemini_client, "_record_success", lambda _complexity: None)
-
-    result = gemini_client.call_gemini("system", "user", _Result)
-
-    assert result.value == "ok"
-    assert calls == ["primary", "fallback", "primary"]

backend/tools/sidecar_monitor.py

@@ -1,13 +1,11 @@
-"""sidecar_monitor — read-only, non-LLM overnight monitor.
+"""Read-only, non-LLM overnight monitor for a running Valhalla simulation.
 
-Polls the same state endpoint the frontend uses and writes machine-
-readable JSONL plus a human-readable report of the night's run,
-including health anomalies.
+It polls the same state endpoint used by the frontend and writes two files:
+* ``*.jsonl``: every sampled snapshot and every detected finding (machine-readable)
+* ``*.txt``: a concise, chronological report suitable for morning review
 
-Architecture: an external observer of Odin.py; never touches checkpoints
-or control endpoints.
-Design: read-only by construction so monitoring can never perturb the
-simulation it observes.
+The monitor never imports simulation modules, calls an LLM, writes checkpoints,
+or invokes any control endpoint. Stop it with Ctrl+C; it flushes a final summary.
 """
 
 from __future__ import annotations

frontend/src/App.jsx

@@ -1,17 +1,3 @@
-/**
- * App — root layout and state composer of the dashboard.
- *
- * Owns auth + simulation snapshot state, renders the map canvas, the
- * agent windows, and all side panels, and handles error banners and
- * day-handoff/provider-failure packets from the backend.
- *
- * Architecture: consumed by main.jsx; renders SimCanvas, InfoBar,
- * ConversationFeed, EventsPanel, DebugPanel, RosterManager, LoginButton.
- *
- * Design: one component per concern, all fed from a single WebSocket
- * snapshot hook (useSimState).
- */
-
 import { useState, useEffect, useCallback } from "react";
 import useSimState from "./hooks/useSimState";
 import { AuthProvider, useAuth } from "./hooks/useAuth";
@@ -23,7 +9,6 @@ import EventsPanel from "./components/EventsPanel";
 import DebugPanel from "./components/DebugPanel";
 import RosterManager from "./components/RosterManager";
 import LoginButton from "./components/LoginButton";
-import LogTerminal from "./components/LogTerminal";
 import { apiUrl } from "./utils/api";
 import "./App.css";
 
@@ -31,6 +16,7 @@ function compactTabPosition(index) {
   const side = index % 2;
   const row = Math.floor(index / 2);
   return {
+    // Reserve the full inspector width even while this card is compact. A
     // right-hand card can then expand without its controls leaving the view.
     x: side ? Math.max(16, window.innerWidth - 276) : 16,
     y: 66 + row * 92,
@@ -57,7 +43,6 @@ function AppContent() {
   const [controlError, setControlError] = useState(null);
   const [simulationRunning, setSimulationRunning] = useState(true);
   const [rosterOpen, setRosterOpen] = useState(false);
-  const [showTerminal, setShowTerminal] = useState(false);
 
   useEffect(() => {
     if (!snapshot) return;
@@ -155,7 +140,6 @@ function AppContent() {
       <EventsPanel events={snapshot?.events} />
       {showDebug && <DebugPanel health={snapshot?.health} />}
       <RosterManager open={rosterOpen} onClose={() => setRosterOpen(false)} simulationRunning={simulationRunning} onError={setControlError} isAuthenticated={isAuthenticated} />
-      {isAuthenticated && showTerminal && <LogTerminal onClose={() => setShowTerminal(false)} />}
 
       {agentIds.map((id, index) => {
         const expanded = expandedAgentIds.has(id);
@@ -188,8 +172,6 @@ function AppContent() {
         }}
         onToggleRoster={() => setRosterOpen((value) => !value)}
         isAuthenticated={isAuthenticated}
-        terminalOpen={showTerminal}
-        onToggleTerminal={() => setShowTerminal((value) => !value)}
       />
       <LoginButton />
     </div>

frontend/src/components/ActionDetail.jsx

@@ -1,14 +1,3 @@
-/**
- * ActionDetail — renders an agent's current action description.
- *
- * Shows the action text with its time range and route progress for
- * movement actions.
- *
- * Architecture: rendered inside AgentWindow's expanded inspector.
- *
- * Design: purely presentational; receives the action object as props.
- */
-
 export default function ActionDetail({ action }) {
   if (!action) {
     return (

frontend/src/components/AgentWindow.jsx

@@ -1,17 +1,3 @@
-/**
- * AgentWindow — per-agent draggable inspector card.
- *
- * Compact card by default; expanded view shows location, current action
- * (via ActionDetail), energy/emotion gauges, pause state, and the agent's
- * live conversation transcript (ChatPanel).
- *
- * Architecture: one instance per agent, laid out by App.jsx; fed from the
- * shared simulation snapshot.
- *
- * Design: windows are draggable (react-draggable) so multiple agents can
- * be inspected simultaneously.
- */
-
 import { useRef, useState, useEffect } from "react";
 import Draggable from "react-draggable";
 import WindowHeader from "./WindowHeader";
@@ -40,6 +26,7 @@ export default function AgentWindow({ agentId, data, speed, defaultPosition, exp
   const conversationId = conversation
     ? `${conversation.partner_id || conversation.partner_name}_${conversation.started_tick ?? "pending"}`
     : null;
+  // A transcript belongs to the live conversation state, not to the agent
   // card's lifetime.  The backend clears this state when the simulated chat
   // finishes; rendering only an active/generating conversation automatically
   // collapses the chat panel as the agent starts their next task.

frontend/src/components/ChatBubble.jsx

@@ -1,14 +1,3 @@
-/**
- * ChatBubble — a single message bubble in a conversation transcript.
- *
- * Aligns right for the focused agent (self) and left for the partner,
- * colored by the speaker's agent color.
- *
- * Architecture: rendered by ChatPanel for each revealed message.
- *
- * Design: purely presentational; no state.
- */
-
 export default function ChatBubble({ text, isSelf, color }) {
   return (
     <div style={{

frontend/src/components/ChatPanel.jsx

@@ -1,16 +1,3 @@
-/**
- * ChatPanel — read-only live transcript of one agent-to-agent conversation.
- *
- * Reveals messages one by one to mirror the backend's staged generation,
- * and auto-scrolls to the newest message.
- *
- * Architecture: rendered inside AgentWindow; fed the conversation object
- * and the revealed-count from the simulation snapshot.
- *
- * Design: observation-only — there is deliberately no input to talk to
- * agents from the UI.
- */
-
 import { useEffect, useRef } from "react";
 import ChatBubble from "./ChatBubble";
 

frontend/src/components/ConversationFeed.jsx

@@ -1,16 +1,3 @@
-/**
- * ConversationFeed — draggable log of recent campus conversations.
- *
- * Lists recent conversations with sentiment color dots, participants,
- * simulation time, and location, with auto-refresh on new entries.
- *
- * Architecture: rendered by App.jsx from the snapshot's
- * recent_conversations block.
- *
- * Design: the sentiment dot is derived from the conversation's structured
- * sentiment field, not guessed from text.
- */
-
 import { useEffect, useRef, useState } from "react";
 import Draggable from "react-draggable";
 
@@ -28,6 +15,7 @@ export default function ConversationFeed({ conversations, minimized = false, onT
     y: 16,
   }));
 
+  // Keep the panel usable after resizing or minimizing, matching agent cards.
   useEffect(() => {
     const clampToViewport = () => {
       const node = nodeRef.current;

frontend/src/components/DebugPanel.jsx

@@ -1,15 +1,3 @@
-/**
- * DebugPanel — engine health readout for researchers.
- *
- * Shows tick, agent/moving/paused counts, background task counts, and
- * per-agent anomalies reported by the backend health monitor.
- *
- * Architecture: toggled from InfoBar; fed by the snapshot's health block.
- *
- * Design: keeps runtime anomalies visible without cluttering the main
- * dashboard.
- */
-
 export default function DebugPanel({ health }) {
   if (!health) return null;
   return (

frontend/src/components/EventsPanel.jsx

@@ -1,15 +1,3 @@
-/**
- * EventsPanel — "campus pulse" list of live and upcoming events.
- *
- * Renders active events as LIVE and upcoming events with time range,
- * category color, and attendance count vs capacity.
- *
- * Architecture: rendered by App.jsx from the snapshot's events block.
- *
- * Design: read-only; event data originates in the deterministic event
- * calendar on the backend.
- */
-
 const CATEGORY_COLOR = {
   "technical-cultural": "#5b9bd5",
   sports: "#51cf66",

frontend/src/components/InfoBar.jsx

@@ -1,19 +1,6 @@
-/**
- * InfoBar — status bar and timeline controls.
- *
- * Shows tick/time/day/pace and agent counts, plus auth-gated controls:
- * rewind (ticks or hours), fast-forward, slow-down, stop/start, and the
- * roster button.
- *
- * Architecture: rendered by App.jsx; posts to the backend's sim-control
- * endpoints with the admin bearer token.
- *
- * Design: controls are hidden for unauthenticated viewers.
- */
-
 import { useState } from "react";
 
-export default function InfoBar({ snapshot, showDebug, onToggleDebug, onFastForward, onSlowDown, onRewind, simulationRunning, onToggleSimulation, onToggleRoster, isAuthenticated, terminalOpen, onToggleTerminal }) {
+export default function InfoBar({ snapshot, showDebug, onToggleDebug, onFastForward, onSlowDown, onRewind, simulationRunning, onToggleSimulation, onToggleRoster, isAuthenticated }) {
   const [rewindAmount, setRewindAmount] = useState("10");
   const [rewindUnit, setRewindUnit] = useState("ticks");
   if (!snapshot) return null;
@@ -140,20 +127,11 @@ export default function InfoBar({ snapshot, showDebug, onToggleDebug, onFastForw
         {showDebug ? "DEBUG ON" : "DEBUG"}
       </button>
       {isAuthenticated && (
-        <>
-          <button onClick={onToggleRoster} style={{
-            pointerEvents: "auto", border: "1px solid rgba(212,160,74,.3)", borderRadius: 3,
-            background: "rgba(212,160,74,.08)", color: "#e7bd70", padding: "2px 5px",
-            fontFamily: "'Space Mono', monospace", fontSize: 8, cursor: "pointer",
-          }} title="Add, retire, or rename agents while the simulation is stopped">ROSTER</button>
-          <button onClick={onToggleTerminal} style={{
-            pointerEvents: "auto", border: "1px solid rgba(91,155,213,.3)", borderRadius: 3,
-            background: terminalOpen ? "rgba(91,155,213,.22)" : "transparent", color: "#8fbbe8", padding: "2px 5px",
-            fontFamily: "'Space Mono', monospace", fontSize: 8, cursor: "pointer",
-          }} title="Open the live backend log terminal">
-            {terminalOpen ? "TERMINAL ON" : "TERMINAL"}
-          </button>
-        </>
+        <button onClick={onToggleRoster} style={{
+          pointerEvents: "auto", border: "1px solid rgba(212,160,74,.3)", borderRadius: 3,
+          background: "rgba(212,160,74,.08)", color: "#e7bd70", padding: "2px 5px",
+          fontFamily: "'Space Mono', monospace", fontSize: 8, cursor: "pointer",
+        }} title="Add, retire, or rename agents while the simulation is stopped">ROSTER</button>
       )}
     </div>
   );

frontend/src/components/Legend.jsx

@@ -1,16 +1,4 @@
-/**
- * Legend — color-to-name roster legend (unused, kept for reference).
- *
- * Maps each agent's dot color to their name and lets the user focus the
- * camera by clicking a name.
- *
- * Architecture: not currently rendered — SimCanvas draws labels directly;
- * retained as the planned clickable roster.
- *
- * Design: superseded by on-canvas labels; left in the tree until the
- * focus UX is finalized.
- */
-
+// Color legend: maps each agent's dot color to their name. Click to focus.
 export default function Legend({ agents, focusedId, onFocus }) {
   if (!agents) return null;
   const entries = Object.entries(agents);

frontend/src/components/LogTerminal.jsx

@@ -1,189 +0,0 @@
-/**
- * LogTerminal — admin-only live view of the backend log relay.
- *
- * Polls GET /api/logs?since=<cursor> with the admin bearer token and
- * renders the captured logger output as a read-only terminal.
- *
- * Architecture: rendered by App.jsx only for authenticated users; toggled
- * by the TERMINAL button in InfoBar.
- *
- * Design: 2s REST polling reuses the existing bearer-token flow instead of
- * adding WebSocket authentication; auto-scroll halts while the pointer is
- * over the terminal so lines can be read mid-stream.
- */
-
-import { useEffect, useRef, useState, useCallback } from "react";
-import { useAuth } from "../hooks/useAuth";
-import { apiUrl } from "../utils/api";
-
-const POLL_MS = 2000;
-const LEVEL_COLORS = {
-  DEBUG: "#6b6b78",
-  INFO: "#8fbbe8",
-  WARNING: "#e7bd70",
-  ERROR: "#ff8b8b",
-  CRITICAL: "#ff6b6b",
-};
-
-export default function LogTerminal({ onClose }) {
-  const { token } = useAuth();
-  const [lines, setLines] = useState([]);
-  const [cursor, setCursor] = useState(0);
-  const [error, setError] = useState(null);
-  const [paused, setPaused] = useState(false);
-  const bodyRef = useRef(null);
-  const pausedRef = useRef(false);
-
-  const fetchNew = useCallback(async () => {
-    try {
-      const res = await fetch(apiUrl(`/api/logs?since=${cursor}`), {
-        headers: token ? { Authorization: `Bearer ${token}` } : {},
-      });
-      if (!res.ok) throw new Error(`Log fetch failed (${res.status})`);
-      const data = await res.json();
-      if (data.lines?.length) {
-        setLines((previous) => [...previous.slice(-2000), ...data.lines]);
-      }
-      if (typeof data.next === "number") setCursor(data.next);
-      setError(null);
-    } catch (err) {
-      setError(err.message);
-    }
-  }, [cursor, token]);
-
-  useEffect(() => {
-    fetchNew();
-    const timer = setInterval(fetchNew, POLL_MS);
-    return () => clearInterval(timer);
-  }, [fetchNew]);
-
-  useEffect(() => {
-    const el = bodyRef.current;
-    if (el && !pausedRef.current) el.scrollTop = el.scrollHeight;
-  }, [lines]);
-
-  async function clearLogs() {
-    try {
-      const res = await fetch(apiUrl("/api/logs/clear"), {
-        method: "POST",
-        headers: token ? { Authorization: `Bearer ${token}` } : {},
-      });
-      if (!res.ok) throw new Error(`Clear failed (${res.status})`);
-      setLines([]);
-    } catch (err) {
-      setError(err.message);
-    }
-  }
-
-  async function copyLogs() {
-    try {
-      await navigator.clipboard.writeText(lines.map((line) => line.text).join("\n"));
-    } catch {
-      /* clipboard unavailable — ignore */
-    }
-  }
-
-  const headerButton = {
-    background: "rgba(91,155,213,.08)",
-    border: "1px solid rgba(91,155,213,.30)",
-    borderRadius: 3,
-    color: "#8fbbe8",
-    padding: "2px 6px",
-    fontFamily: "'Space Mono', monospace",
-    fontSize: 8,
-    cursor: "pointer",
-  };
-
-  return (
-    <aside
-      style={{
-        position: "fixed",
-        right: 16,
-        top: 66,
-        bottom: 96,
-        width: 460,
-        maxWidth: "calc(100vw - 32px)",
-        display: "flex",
-        flexDirection: "column",
-        background: "rgba(8,10,12,0.95)",
-        backdropFilter: "blur(8px)",
-        border: "1px solid rgba(91,155,213,0.30)",
-        borderRadius: 6,
-        boxShadow: "0 14px 44px rgba(0,0,0,0.6)",
-        fontFamily: "'Space Mono', monospace",
-        zIndex: 1300,
-      }}
-    >
-      <header
-        style={{
-          display: "flex",
-          alignItems: "center",
-          gap: 8,
-          padding: "8px 12px",
-          borderBottom: "1px solid rgba(255,255,255,0.07)",
-        }}
-      >
-        <span style={{ color: "#8fbbe8", fontWeight: 700, fontSize: 10, letterSpacing: "0.14em" }}>
-          LOG TERMINAL
-        </span>
-        <span style={{ fontSize: 9, color: "#6b6b78" }}>{lines.length} lines</span>
-        <span style={{ flex: 1 }} />
-        <button onClick={copyLogs} style={headerButton} title="Copy all relayed lines to the clipboard">
-          COPY
-        </button>
-        <button onClick={clearLogs} style={headerButton} title="Clear the in-memory relay buffer">
-          CLEAR
-        </button>
-        <button onClick={onClose} style={headerButton} title="Close the log terminal">
-          X
-        </button>
-      </header>
-
-      <div
-        ref={bodyRef}
-        onMouseEnter={() => {
-          pausedRef.current = true;
-          setPaused(true);
-        }}
-        onMouseLeave={() => {
-          pausedRef.current = false;
-          setPaused(false);
-        }}
-        style={{
-          flex: 1,
-          overflowY: "auto",
-          padding: "10px 12px",
-          fontSize: 9,
-          lineHeight: 1.65,
-        }}
-      >
-        {lines.length === 0 && (
-          <div style={{ color: "#6b6b78" }}>no lines yet — the relay captures log output once the simulation starts</div>
-        )}
-        {lines.map((entry) => (
-          <div
-            key={entry.seq}
-            style={{
-              color: LEVEL_COLORS[entry.level] || "#d0d0da",
-              whiteSpace: "pre-wrap",
-              wordBreak: "break-word",
-            }}
-          >
-            {entry.text}
-          </div>
-        ))}
-      </div>
-
-      <footer
-        style={{
-          padding: "6px 12px",
-          borderTop: "1px solid rgba(255,255,255,0.07)",
-          fontSize: 9,
-          color: error ? "#ff8b8b" : paused ? "#e7bd70" : "#6b6b78",
-        }}
-      >
-        {error ? `ERROR: ${error}` : paused ? "PAUSED — move the pointer away to resume auto-scroll" : "TAILING — 2s poll"}
-      </footer>
-    </aside>
-  );
-}

frontend/src/components/LoginButton.jsx

@@ -1,15 +1,3 @@
-/**
- * LoginButton — auth entry point for admin controls.
- *
- * Shows a LOGIN pill when unauthenticated (opens LoginModal) and a user
- * dropdown with logout when a session is active.
- *
- * Architecture: rendered by App.jsx; uses the useAuth context shared with
- * InfoBar and RosterManager.
- *
- * Design: anonymous users can watch; only admins get controls.
- */
-
 import { useState } from "react";
 import { useAuth } from "../hooks/useAuth";
 import LoginModal from "./LoginModal";