{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "7abb57ff",
   "metadata": {},
   "source": [
    "# Checkpoint-log pilot: declaration versus numerical change\n",
    "\n",
    "**Executed secondary analysis, 17 September 2026.** One published training trajectory; 300 gradient-norm observations and 30 evaluation-loss observations. Inputs are transcribed source fields, not simulated model data. A source-verification script is supplied but remote byte-level verification remains pending.\n",
    "\n",
    "This notebook does not run model inference or training, does not load released adapter tensors, and does not classify consciousness. Synthetic arrays appear only in explicitly labeled algebra tests. The declaration-before-change hypothesis remains to be tested with a separately defined marker.\n",
    "\n",
    "Public source: `ModelOrganismsForEM/Qwen2.5-14B-Instruct_R1_0_1_0_extended_train/checkpoints/checkpoint-300/trainer_state.json`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "ac713a6e",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-09-17T07:34:31.690141Z",
     "iopub.status.busy": "2026-09-17T07:34:31.690027Z",
     "iopub.status.idle": "2026-09-17T07:34:31.702236Z",
     "shell.execute_reply": "2026-09-17T07:34:31.701642Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Independent runs: 1; gradient observations: 300; evaluation observations: 30\n",
      "{\n",
      "  \"global_step\": 300,\n",
      "  \"epoch\": 0.7566204287515763,\n",
      "  \"eval_steps\": 10,\n",
      "  \"logging_steps\": 1,\n",
      "  \"max_steps\": 792,\n",
      "  \"num_input_tokens_seen\": 0,\n",
      "  \"num_train_epochs\": 2,\n",
      "  \"save_steps\": 1,\n",
      "  \"train_batch_size\": 2,\n",
      "  \"total_flos\": 6.020530367924429e+16\n",
      "}\n"
     ]
    }
   ],
   "source": [
    "from pathlib import Path\n",
    "import json, sys, importlib.metadata\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "root = Path.cwd()\n",
    "if not (root / \"raw/gradient_norm.csv\").exists():\n",
    "    raise RuntimeError(\"Run from the extracted package root.\")\n",
    "sys.path.insert(0, str(root / \"code\"))\n",
    "from analyze_logs import load_data, main, two_segment_scan\n",
    "from adapter_geometry import metrics\n",
    "g, e = load_data(root)\n",
    "print(f\"Independent runs: 1; gradient observations: {len(g)}; evaluation observations: {len(e)}\")\n",
    "print(json.dumps(json.loads((root / \"raw/provenance.json\").read_text())[\"selected_source_metadata\"], indent=2))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2b350730",
   "metadata": {},
   "source": [
    "## 1. Verify extraction structure and execute the analysis\n",
    "Local validation checks structure and sentinel values, not remote-source equivalence."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "ba823bca",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-09-17T07:34:31.703949Z",
     "iopub.status.busy": "2026-09-17T07:34:31.703835Z",
     "iopub.status.idle": "2026-09-17T07:34:31.805505Z",
     "shell.execute_reply": "2026-09-17T07:34:31.804744Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{\n",
      "  \"status\": \"exploratory single-trajectory secondary numerical analysis\",\n",
      "  \"n_independent_runs\": 1,\n",
      "  \"n_gradient_observations\": 300,\n",
      "  \"n_evaluation_observations\": 30,\n",
      "  \"source_global_step\": 300,\n",
      "  \"source_max_steps\": 792,\n",
      "  \"fraction_of_configured_steps_observed\": 0.3787878787878788,\n",
      "  \"raw_gradient_peak\": {\n",
      "    \"step\": 169,\n",
      "    \"value\": 40.9891242980957\n",
      "  },\n",
      "  \"centered_smoothing_peaks\": [\n",
      "    {\n",
      "      \"window_width\": 1,\n",
      "      \"center_step\": 169,\n",
      "      \"mean_at_peak\": 40.9891242980957\n",
      "    },\n",
      "    {\n",
      "      \"window_width\": 5,\n",
      "      \"center_step\": 171,\n",
      "      \"mean_at_peak\": 33.3971752166748\n",
      "    },\n",
      "    {\n",
      "      \"window_width\": 11,\n",
      "      \"center_step\": 168,\n",
      "      \"mean_at_peak\": 31.858454964377664\n",
      "    },\n",
      "    {\n",
      "      \"window_width\": 21,\n",
      "      \"center_step\": 167,\n",
      "      \"mean_at_peak\": 28.889931724185036\n",
      "    },\n",
      "    {\n",
      "      \"window_width\": 31,\n",
      "      \"center_step\": 166,\n",
      "      \"mean_at_peak\": 26.44963159868794\n",
      "    }\n",
      "  ],\n",
      "  \"largest_evaluation_loss_drop\": {\n",
      "    \"from_step\": 160,\n",
      "    \"to_step\": 170,\n",
      "    \"absolute\": 0.20026636123657227,\n",
      "    \"relative_percent\": 7.176222569879443\n",
      "  },\n",
      "  \"evaluation_loss_initial_observed\": 4.168299674987793,\n",
      "  \"evaluation_loss_final_observed\": 1.699493408203125,\n",
      "  \"evaluation_loss_reduction_percent\": 59.22813759286388,\n",
      "  \"all_observed_evaluation_differences_decrease\": true,\n",
      "  \"mean_grad_161_180\": 28.693284511566162,\n",
      "  \"mean_grad_181_200\": 15.343834018707275,\n",
      "  \"gradient_mean_drop_percent_161_180_to_181_200\": 46.52465104675545,\n",
      "  \"gradient_mean_ratio_161_180_to_141_160\": 1.4171730133407292,\n",
      "  \"best_descriptive_two_line_break\": {\n",
      "    \"split_after_step\": 181.0,\n",
      "    \"second_segment_starts\": 182.0,\n",
      "    \"sse\": 1679.3678563750443,\n",
      "    \"slope_before\": 0.14254928130287897,\n",
      "    \"slope_after\": -0.05131966136931838,\n",
      "    \"intercept_before\": 0.3765082286427286,\n",
      "    \"intercept_after\": 24.860779954748384\n",
      "  },\n",
      "  \"break_window_sensitivity\": [\n",
      "    {\n",
      "      \"window_start\": 1,\n",
      "      \"window_end\": 300,\n",
      "      \"min_segment_points\": 15,\n",
      "      \"split_after_step\": 181.0,\n",
      "      \"second_segment_starts\": 182.0,\n",
      "      \"sse\": 1679.3678563750443,\n",
      "      \"slope_before\": 0.14254928130287897,\n",
      "      \"slope_after\": -0.05131966136931838,\n",
      "      \"intercept_before\": 0.3765082286427286,\n",
      "      \"intercept_after\": 24.860779954748384\n",
      "    },\n",
      "    {\n",
      "      \"window_start\": 1,\n",
      "      \"window_end\": 300,\n",
      "      \"min_segment_points\": 30,\n",
      "      \"split_after_step\": 181.0,\n",
      "      \"second_segment_starts\": 182.0,\n",
      "      \"sse\": 1679.3678563750443,\n",
      "      \"slope_before\": 0.14254928130287897,\n",
      "      \"slope_after\": -0.05131966136931838,\n",
      "      \"intercept_before\": 0.3765082286427286,\n",
      "      \"intercept_after\": 24.860779954748384\n",
      "    },\n",
      "    {\n",
      "      \"window_start\": 1,\n",
      "      \"window_end\": 300,\n",
      "      \"min_segment_points\": 50,\n",
      "      \"split_after_step\": 181.0,\n",
      "      \"second_segment_starts\": 182.0,\n",
      "      \"sse\": 1679.3678563750443,\n",
      "      \"slope_before\": 0.14254928130287897,\n",
      "      \"slope_after\": -0.05131966136931838,\n",
      "      \"intercept_before\": 0.3765082286427286,\n",
      "      \"intercept_after\": 24.860779954748384\n",
      "    },\n",
      "    {\n",
      "      \"window_start\": 1,\n",
      "      \"window_end\": 300,\n",
      "      \"min_segment_points\": 60,\n",
      "      \"split_after_step\": 181.0,\n",
      "      \"second_segment_starts\": 182.0,\n",
      "      \"sse\": 1679.3678563750443,\n",
      "      \"slope_before\": 0.14254928130287897,\n",
      "      \"slope_after\": -0.05131966136931838,\n",
      "      \"intercept_before\": 0.3765082286427286,\n",
      "      \"intercept_after\": 24.860779954748384\n",
      "    },\n",
      "    {\n",
      "      \"window_start\": 50,\n",
      "      \"window_end\": 300,\n",
      "      \"min_segment_points\": 20,\n",
      "      \"split_after_step\": 181.0,\n",
      "      \"second_segment_starts\": 182.0,\n",
      "      \"sse\": 1625.997097158174,\n",
      "      \"slope_before\": 0.1544005478569755,\n",
      "      \"slope_after\": -0.05131966136931838,\n",
      "      \"intercept_before\": -1.1810741527664248,\n",
      "      \"intercept_after\": 24.860779954748384\n",
      "    },\n",
      "    {\n",
      "      \"window_start\": 100,\n",
      "      \"window_end\": 250,\n",
      "      \"min_segment_points\": 20,\n",
      "      \"split_after_step\": 176.0,\n",
      "      \"second_segment_starts\": 177.0,\n",
      "      \"sse\": 1313.6060095202854,\n",
      "      \"slope_before\": 0.23582091898852714,\n",
      "      \"slope_after\": -0.07259306857163082,\n",
      "      \"intercept_before\": -12.514042273650343,\n",
      "      \"intercept_after\": 29.854009722187445\n",
      "    },\n",
      "    {\n",
      "      \"window_start\": 140,\n",
      "      \"window_end\": 220,\n",
      "      \"min_segment_points\": 15,\n",
      "      \"split_after_step\": 175.0,\n",
      "      \"second_segment_starts\": 176.0,\n",
      "      \"sse\": 735.8379765313783,\n",
      "      \"slope_before\": 0.5171299954025289,\n",
      "      \"slope_after\": -0.15661300789854415,\n",
      "      \"intercept_before\": -56.8914832982931,\n",
      "      \"intercept_after\": 46.5270450297185\n",
      "    }\n",
      "  ],\n",
      "  \"declaration_time\": null,\n",
      "  \"consciousness_onset_time\": null,\n",
      "  \"behavioral_data_analyzed\": false,\n",
      "  \"adapter_direction_measured\": false,\n",
      "  \"weight_update_norm_measured\": false,\n",
      "  \"universal_irreversibility_tested\": false,\n",
      "  \"precision_note\": \"CSV inputs transcribed from source-reader JSON; per-value remote verification pending. Local checks passed.\"\n",
      "}\n"
     ]
    }
   ],
   "source": [
    "summary = main(root)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8231b27b",
   "metadata": {},
   "source": [
    "## 2. Peak and smoothing sensitivity\n",
    "Centered smoothing is retrospective and uses future observations. The window peak is not an online event detector."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "45d1f36e",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-09-17T07:34:31.806734Z",
     "iopub.status.busy": "2026-09-17T07:34:31.806608Z",
     "iopub.status.idle": "2026-09-17T07:34:31.817597Z",
     "shell.execute_reply": "2026-09-17T07:34:31.816574Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>window_width</th>\n",
       "      <th>center_step</th>\n",
       "      <th>mean_at_peak</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>169</td>\n",
       "      <td>40.989124</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>5</td>\n",
       "      <td>171</td>\n",
       "      <td>33.397175</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>11</td>\n",
       "      <td>168</td>\n",
       "      <td>31.858455</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>21</td>\n",
       "      <td>167</td>\n",
       "      <td>28.889932</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>31</td>\n",
       "      <td>166</td>\n",
       "      <td>26.449632</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   window_width  center_step  mean_at_peak\n",
       "0             1          169     40.989124\n",
       "1             5          171     33.397175\n",
       "2            11          168     31.858455\n",
       "3            21          167     28.889932\n",
       "4            31          166     26.449632"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd.read_csv(root / \"results/smoothing_sensitivity.csv\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0ded0b66",
   "metadata": {},
   "source": [
    "## 3. Adjacent gradient windows\n",
    "Windows were chosen with prior knowledge of the source paper's approximate step-180 region; these are descriptive effects, not preregistered causal estimates."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "612a4eab",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-09-17T07:34:31.819228Z",
     "iopub.status.busy": "2026-09-17T07:34:31.819074Z",
     "iopub.status.idle": "2026-09-17T07:34:31.830993Z",
     "shell.execute_reply": "2026-09-17T07:34:31.830192Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>start</th>\n",
       "      <th>end</th>\n",
       "      <th>n</th>\n",
       "      <th>mean</th>\n",
       "      <th>median</th>\n",
       "      <th>min</th>\n",
       "      <th>max</th>\n",
       "      <th>sd_descriptive</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>20</td>\n",
       "      <td>20</td>\n",
       "      <td>2.354562</td>\n",
       "      <td>2.207071</td>\n",
       "      <td>1.380375</td>\n",
       "      <td>3.868789</td>\n",
       "      <td>0.751823</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>81</td>\n",
       "      <td>100</td>\n",
       "      <td>20</td>\n",
       "      <td>12.797106</td>\n",
       "      <td>12.738049</td>\n",
       "      <td>10.495080</td>\n",
       "      <td>14.986769</td>\n",
       "      <td>1.175160</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>121</td>\n",
       "      <td>140</td>\n",
       "      <td>20</td>\n",
       "      <td>16.745894</td>\n",
       "      <td>17.092672</td>\n",
       "      <td>14.262141</td>\n",
       "      <td>19.516367</td>\n",
       "      <td>1.291065</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>141</td>\n",
       "      <td>160</td>\n",
       "      <td>20</td>\n",
       "      <td>20.246847</td>\n",
       "      <td>19.583440</td>\n",
       "      <td>15.177545</td>\n",
       "      <td>28.506445</td>\n",
       "      <td>3.708671</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>161</td>\n",
       "      <td>180</td>\n",
       "      <td>20</td>\n",
       "      <td>28.693285</td>\n",
       "      <td>29.035612</td>\n",
       "      <td>18.069666</td>\n",
       "      <td>40.989124</td>\n",
       "      <td>5.055740</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>181</td>\n",
       "      <td>200</td>\n",
       "      <td>20</td>\n",
       "      <td>15.343834</td>\n",
       "      <td>14.487312</td>\n",
       "      <td>12.611186</td>\n",
       "      <td>22.843805</td>\n",
       "      <td>2.633433</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>201</td>\n",
       "      <td>220</td>\n",
       "      <td>20</td>\n",
       "      <td>13.916411</td>\n",
       "      <td>13.565392</td>\n",
       "      <td>11.437468</td>\n",
       "      <td>16.698671</td>\n",
       "      <td>1.529096</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>281</td>\n",
       "      <td>300</td>\n",
       "      <td>20</td>\n",
       "      <td>9.375246</td>\n",
       "      <td>9.671891</td>\n",
       "      <td>7.580107</td>\n",
       "      <td>11.036866</td>\n",
       "      <td>0.921460</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   start  end   n       mean     median        min        max  sd_descriptive\n",
       "0      1   20  20   2.354562   2.207071   1.380375   3.868789        0.751823\n",
       "1     81  100  20  12.797106  12.738049  10.495080  14.986769        1.175160\n",
       "2    121  140  20  16.745894  17.092672  14.262141  19.516367        1.291065\n",
       "3    141  160  20  20.246847  19.583440  15.177545  28.506445        3.708671\n",
       "4    161  180  20  28.693285  29.035612  18.069666  40.989124        5.055740\n",
       "5    181  200  20  15.343834  14.487312  12.611186  22.843805        2.633433\n",
       "6    201  220  20  13.916411  13.565392  11.437468  16.698671        1.529096\n",
       "7    281  300  20   9.375246   9.671891   7.580107  11.036866        0.921460"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mean decrease (%): 46.52465104675544\n"
     ]
    }
   ],
   "source": [
    "windows = pd.read_csv(root / \"results/gradient_window_summaries.csv\")\n",
    "display(windows)\n",
    "pre = float(windows.loc[windows.start.eq(161), \"mean\"].iloc[0])\n",
    "post = float(windows.loc[windows.start.eq(181), \"mean\"].iloc[0])\n",
    "print(\"Mean decrease (%):\", 100 * (1 - post / pre))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a6d27346",
   "metadata": {},
   "source": [
    "## 4. Evaluation-loss changes\n",
    "These are trainer-reported evaluation losses, not consciousness or misalignment scores."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "a0ddcd25",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-09-17T07:34:31.832336Z",
     "iopub.status.busy": "2026-09-17T07:34:31.832148Z",
     "iopub.status.idle": "2026-09-17T07:34:31.844006Z",
     "shell.execute_reply": "2026-09-17T07:34:31.843369Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>step</th>\n",
       "      <th>eval_loss</th>\n",
       "      <th>previous_step</th>\n",
       "      <th>loss_drop_from_previous</th>\n",
       "      <th>loss_drop_per_step</th>\n",
       "      <th>relative_drop_percent</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>140</td>\n",
       "      <td>3.080153</td>\n",
       "      <td>130.0</td>\n",
       "      <td>0.131675</td>\n",
       "      <td>0.013168</td>\n",
       "      <td>4.099691</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>150</td>\n",
       "      <td>2.947167</td>\n",
       "      <td>140.0</td>\n",
       "      <td>0.132986</td>\n",
       "      <td>0.013299</td>\n",
       "      <td>4.317508</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <td>160</td>\n",
       "      <td>2.790693</td>\n",
       "      <td>150.0</td>\n",
       "      <td>0.156473</td>\n",
       "      <td>0.015647</td>\n",
       "      <td>5.309282</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>170</td>\n",
       "      <td>2.590427</td>\n",
       "      <td>160.0</td>\n",
       "      <td>0.200266</td>\n",
       "      <td>0.020027</td>\n",
       "      <td>7.176223</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>180</td>\n",
       "      <td>2.413678</td>\n",
       "      <td>170.0</td>\n",
       "      <td>0.176749</td>\n",
       "      <td>0.017675</td>\n",
       "      <td>6.823161</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>190</td>\n",
       "      <td>2.334286</td>\n",
       "      <td>180.0</td>\n",
       "      <td>0.079392</td>\n",
       "      <td>0.007939</td>\n",
       "      <td>3.289252</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19</th>\n",
       "      <td>200</td>\n",
       "      <td>2.255881</td>\n",
       "      <td>190.0</td>\n",
       "      <td>0.078405</td>\n",
       "      <td>0.007840</td>\n",
       "      <td>3.358829</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20</th>\n",
       "      <td>210</td>\n",
       "      <td>2.178447</td>\n",
       "      <td>200.0</td>\n",
       "      <td>0.077434</td>\n",
       "      <td>0.007743</td>\n",
       "      <td>3.432552</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    step  eval_loss  ...  loss_drop_per_step  relative_drop_percent\n",
       "13   140   3.080153  ...            0.013168               4.099691\n",
       "14   150   2.947167  ...            0.013299               4.317508\n",
       "15   160   2.790693  ...            0.015647               5.309282\n",
       "16   170   2.590427  ...            0.020027               7.176223\n",
       "17   180   2.413678  ...            0.017675               6.823161\n",
       "18   190   2.334286  ...            0.007939               3.289252\n",
       "19   200   2.255881  ...            0.007840               3.358829\n",
       "20   210   2.178447  ...            0.007743               3.432552\n",
       "\n",
       "[8 rows x 6 columns]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "All observed intervals decrease: True\n"
     ]
    }
   ],
   "source": [
    "loss_changes = pd.read_csv(root / \"results/evaluation_loss_changes.csv\")\n",
    "display(loss_changes.loc[loss_changes.step.between(140, 210)])\n",
    "print(\"All observed intervals decrease:\", bool((e.eval_loss.diff().dropna() < 0).all()))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "564e9647",
   "metadata": {},
   "source": [
    "## 5. Descriptive break sensitivity\n",
    "The fitted break may be displaced from the peak and moves with the chosen window. This is one time series, not a population of onset times."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "3fc60b60",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-09-17T07:34:31.845354Z",
     "iopub.status.busy": "2026-09-17T07:34:31.845223Z",
     "iopub.status.idle": "2026-09-17T07:34:31.862284Z",
     "shell.execute_reply": "2026-09-17T07:34:31.861548Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>window_start</th>\n",
       "      <th>window_end</th>\n",
       "      <th>min_segment_points</th>\n",
       "      <th>split_after_step</th>\n",
       "      <th>second_segment_starts</th>\n",
       "      <th>sse</th>\n",
       "      <th>slope_before</th>\n",
       "      <th>slope_after</th>\n",
       "      <th>intercept_before</th>\n",
       "      <th>intercept_after</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>300</td>\n",
       "      <td>15</td>\n",
       "      <td>181.0</td>\n",
       "      <td>182.0</td>\n",
       "      <td>1679.367856</td>\n",
       "      <td>0.142549</td>\n",
       "      <td>-0.051320</td>\n",
       "      <td>0.376508</td>\n",
       "      <td>24.860780</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1</td>\n",
       "      <td>300</td>\n",
       "      <td>30</td>\n",
       "      <td>181.0</td>\n",
       "      <td>182.0</td>\n",
       "      <td>1679.367856</td>\n",
       "      <td>0.142549</td>\n",
       "      <td>-0.051320</td>\n",
       "      <td>0.376508</td>\n",
       "      <td>24.860780</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>1</td>\n",
       "      <td>300</td>\n",
       "      <td>50</td>\n",
       "      <td>181.0</td>\n",
       "      <td>182.0</td>\n",
       "      <td>1679.367856</td>\n",
       "      <td>0.142549</td>\n",
       "      <td>-0.051320</td>\n",
       "      <td>0.376508</td>\n",
       "      <td>24.860780</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>1</td>\n",
       "      <td>300</td>\n",
       "      <td>60</td>\n",
       "      <td>181.0</td>\n",
       "      <td>182.0</td>\n",
       "      <td>1679.367856</td>\n",
       "      <td>0.142549</td>\n",
       "      <td>-0.051320</td>\n",
       "      <td>0.376508</td>\n",
       "      <td>24.860780</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>50</td>\n",
       "      <td>300</td>\n",
       "      <td>20</td>\n",
       "      <td>181.0</td>\n",
       "      <td>182.0</td>\n",
       "      <td>1625.997097</td>\n",
       "      <td>0.154401</td>\n",
       "      <td>-0.051320</td>\n",
       "      <td>-1.181074</td>\n",
       "      <td>24.860780</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>100</td>\n",
       "      <td>250</td>\n",
       "      <td>20</td>\n",
       "      <td>176.0</td>\n",
       "      <td>177.0</td>\n",
       "      <td>1313.606010</td>\n",
       "      <td>0.235821</td>\n",
       "      <td>-0.072593</td>\n",
       "      <td>-12.514042</td>\n",
       "      <td>29.854010</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>140</td>\n",
       "      <td>220</td>\n",
       "      <td>15</td>\n",
       "      <td>175.0</td>\n",
       "      <td>176.0</td>\n",
       "      <td>735.837977</td>\n",
       "      <td>0.517130</td>\n",
       "      <td>-0.156613</td>\n",
       "      <td>-56.891483</td>\n",
       "      <td>46.527045</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   window_start  window_end  ...  intercept_before  intercept_after\n",
       "0             1         300  ...          0.376508        24.860780\n",
       "1             1         300  ...          0.376508        24.860780\n",
       "2             1         300  ...          0.376508        24.860780\n",
       "3             1         300  ...          0.376508        24.860780\n",
       "4            50         300  ...         -1.181074        24.860780\n",
       "5           100         250  ...        -12.514042        29.854010\n",
       "6           140         220  ...        -56.891483        46.527045\n",
       "\n",
       "[7 rows x 10 columns]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>model</th>\n",
       "      <th>coefficient_count</th>\n",
       "      <th>sse</th>\n",
       "      <th>rmse</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>polynomial_degree_1</td>\n",
       "      <td>2</td>\n",
       "      <td>10845.877951</td>\n",
       "      <td>6.012730</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>polynomial_degree_2</td>\n",
       "      <td>3</td>\n",
       "      <td>4107.151918</td>\n",
       "      <td>3.700068</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>polynomial_degree_3</td>\n",
       "      <td>4</td>\n",
       "      <td>4107.093788</td>\n",
       "      <td>3.700042</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>polynomial_degree_4</td>\n",
       "      <td>5</td>\n",
       "      <td>3406.631851</td>\n",
       "      <td>3.369783</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>polynomial_degree_5</td>\n",
       "      <td>6</td>\n",
       "      <td>3372.830193</td>\n",
       "      <td>3.353024</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>two_lines_with_selected_break</td>\n",
       "      <td>4</td>\n",
       "      <td>1679.367856</td>\n",
       "      <td>2.365987</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                           model  coefficient_count           sse      rmse\n",
       "0            polynomial_degree_1                  2  10845.877951  6.012730\n",
       "1            polynomial_degree_2                  3   4107.151918  3.700068\n",
       "2            polynomial_degree_3                  4   4107.093788  3.700042\n",
       "3            polynomial_degree_4                  5   3406.631851  3.369783\n",
       "4            polynomial_degree_5                  6   3372.830193  3.353024\n",
       "5  two_lines_with_selected_break                  4   1679.367856  2.365987"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "display(pd.read_csv(root / \"results/break_window_sensitivity.csv\"))\n",
    "display(pd.read_csv(root / \"results/descriptive_fit_comparison.csv\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ce583869",
   "metadata": {},
   "source": [
    "## 6. Synthetic algebra check: a factor can reverse while the effective operator does not\n",
    "The following arrays are SYNTHETIC unit-test data, not model parameters."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "b465cff7",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-09-17T07:34:31.863854Z",
     "iopub.status.busy": "2026-09-17T07:34:31.863376Z",
     "iopub.status.idle": "2026-09-17T07:34:31.868090Z",
     "shell.execute_reply": "2026-09-17T07:34:31.866881Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{\n",
      "  \"update_norm_before\": 23.958297101421877,\n",
      "  \"update_norm_after\": 23.958297101421877,\n",
      "  \"effective_update_distance\": 0.0,\n",
      "  \"effective_update_cosine\": 1.0,\n",
      "  \"B_factor_cosine\": -1.0\n",
      "}\n"
     ]
    }
   ],
   "source": [
    "a = np.array([1., 2., 3.])\n",
    "b = np.array([4., 5.])\n",
    "result = metrics(a, b, -2*a, -0.5*b)\n",
    "print(json.dumps(result, indent=2))\n",
    "assert np.allclose(np.outer(b, a), np.outer(-0.5*b, -2*a))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ef5e2fa4",
   "metadata": {},
   "source": [
    "## 7. Local test suite"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "0d8c6125",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-09-17T07:34:31.869765Z",
     "iopub.status.busy": "2026-09-17T07:34:31.869621Z",
     "iopub.status.idle": "2026-09-17T07:34:32.779784Z",
     "shell.execute_reply": "2026-09-17T07:34:32.778951Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "test_counts_and_fields (test_analysis.DataChecks.test_counts_and_fields) ... ok\n",
      "test_gauge_invariance (test_analysis.DataChecks.test_gauge_invariance) ... ok\n",
      "test_geometry_dense_equivalence (test_analysis.DataChecks.test_geometry_dense_equivalence) ... ok\n",
      "test_loss_drop (test_analysis.DataChecks.test_loss_drop) ... ok\n",
      "test_no_declaration_in_data (test_analysis.DataChecks.test_no_declaration_in_data) ... ok\n",
      "test_peak (test_analysis.DataChecks.test_peak) ... ok\n",
      "test_scan (test_analysis.DataChecks.test_scan) ... ok\n",
      "test_smoothing (test_analysis.DataChecks.test_smoothing) ... ok\n",
      "test_synthetic_scan (test_analysis.DataChecks.test_synthetic_scan) ... ok\n",
      "test_zero_direction_undefined (test_analysis.DataChecks.test_zero_direction_undefined) ... ok\n",
      "\n",
      "----------------------------------------------------------------------\n",
      "Ran 10 tests in 0.042s\n",
      "\n",
      "OK\n",
      "\n"
     ]
    }
   ],
   "source": [
    "import subprocess\n",
    "p = subprocess.run([sys.executable, \"-m\", \"unittest\", \"discover\", \"-s\", \"code\", \"-p\", \"test_*.py\", \"-v\"],\n",
    "                   cwd=root, text=True, capture_output=True)\n",
    "print(p.stdout + p.stderr)\n",
    "assert p.returncode == 0"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "67860b5b",
   "metadata": {},
   "source": [
    "## Conclusion\n",
    "A real training-log episode is present. The raw peak, smoothed peaks, loss-decrease interval, and descriptive trend break are different summaries. No declaration field or paired self/reference measurement is present; neither a declaration time nor a consciousness onset is estimated.\n",
    "\n",
    "Full-model behavior, independent-seed replication, causal declaration feedback, real tensor geometry, and irreversible persistence remain untested. See the manuscript for the proposed protocol, not for assumed positive results."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.13.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}