Update README.md

README.md

@@ -38,8 +38,6 @@ A reproducible, decision-theoretic benchmark where each episode logs (i) calibra
 
 ## Files in this dataset
 
-This repository contains the following primary artifacts:
-
 - **`mixed_initiative_traces.csv`**  
   Main tabular dataset. Each row corresponds to one mixed-initiative decision episode.
 
@@ -64,18 +62,18 @@ This dataset enables benchmarking of decision policies, calibrated uncertainty m
 
 ## Concept-to-data mapping (Figures 1–7)
 
-The original CHI’99 paper presents UI sketches and analytic models. This dataset provides data-driven analogs:
+This dataset provides data-driven analogs to the CHI’99 LookOut concepts:
 
 - **Fig 1 — Manual invocation**  
-  Manual inspection and explicit invocation signals  
+  Manual inspection and explicit invocation  
   (`manual_hover_inspect`, `manual_click_invoke`)
 
 - **Fig 2 — Explicit social agent**  
-  Dialog confirmation and refinement flow  
+  Dialog confirmation and refinement  
   (`dialog_confirmed`, `refine_after_accept`)
 
 - **Fig 3 — Automated scoping**  
-  Scoped actions selected under uncertainty  
+  Risk-reducing partial actions  
   (`agent_action = scope`)
 
 - **Fig 4 — Action vs no-action threshold**  
@@ -85,10 +83,11 @@ The original CHI’99 paper presents UI sketches and analytic models. This datas
   Threshold changes under attention and urgency
 
 - **Fig 6 — Dialog as a second option**  
-  Three-way EU comparison (no_action / dialog / action)
+  Three-way expected utility comparison
 
 - **Fig 7 — Dwell time vs message length**  
-  Sigmoid attention model linking `message_length_bytes` and `dwell_time_sec`
+  Sigmoid attention model  
+  (`message_length_bytes`, `dwell_time_sec`)
 
 ---
 
@@ -96,36 +95,21 @@ The original CHI’99 paper presents UI sketches and analytic models. This datas
 
 Each episode is generated as follows:
 
-1. **Latent evidence → calibrated posterior**  
-   Sample `evidence_score`, compute `p_true`, and sample `true_goal`.
-
-2. **Reported belief**  
-   Compute `p_model` via logit-space distortion of `p_true`.
-
-3. **Context and attention**  
-   Sample urgency and compute readiness from dwell-time dynamics.
-
-4. **Decision policy**  
-   Select modality and action by maximizing expected utility and compute analytic thresholds.
-
-5. **Outcome realization**  
-   Sample user response and log realized utility.
+1. Sample latent evidence and compute calibrated posterior `p_true`
+2. Apply belief distortion to obtain `p_model`
+3. Sample urgency and compute attention/readiness
+4. Compute expected utilities and analytic thresholds
+5. Select action via expected-utility maximization
+6. Sample user response and realized utility
 
 ---
 
 ## Recommended tasks
 
-1. **Calibration and uncertainty learning**  
-   Predict `true_goal` and evaluate against `p_true` and `p_model`.
-
-2. **Policy learning / imitation learning**  
-   Predict `agent_action` given context and compare to baseline policies.
-
-3. **Counterfactual decision analysis**  
-   Re-evaluate decisions using stored utilities and thresholds.
-
-4. **Interpretability benchmarks**  
-   Explain decisions using analytic thresholds and EU curves.
+1. **Calibration and uncertainty learning**
+2. **Policy learning / imitation learning**
+3. **Counterfactual decision analysis**
+4. **Interpretability and explanation benchmarking**
 
 ---
 
@@ -136,17 +120,17 @@ Each row represents a **single mixed-initiative decision episode**.
 ### Identity
 
 - **`episode_id`**  
-  Unique episode identifier.
+  Unique identifier for the episode.
 
 - **`user_id`**  
-  Synthetic user identifier for repeated-interaction simulation.
+  Synthetic user identifier used to simulate repeated interactions.
 
 ---
 
 ### Ground truth and uncertainty
 
 - **`true_goal`** *(binary)*  
-  Whether the suggested action was truly appropriate.
+  Whether the suggested action was actually appropriate.
 
 - **`p_true`** *(float, 0–1)*  
   Calibrated posterior probability of correctness.
@@ -162,13 +146,13 @@ Each row represents a **single mixed-initiative decision episode**.
 ### Attention and readiness
 
 - **`message_length_bytes`** *(int)*  
-  Size of candidate interruption message.
+  Size of the candidate interruption message.
 
 - **`dwell_time_sec`** *(float)*  
   Simulated dwell time before interruption.
 
 - **`time_since_focus_sec`** *(float)*  
-  Time since last user engagement.
+  Time elapsed since last user engagement.
 
 - **`not_ready_score`** *(float, 0–1)*  
   Aggregate interruption cost estimate.
@@ -178,17 +162,17 @@ Each row represents a **single mixed-initiative decision episode**.
 ### Context
 
 - **`urgency_0_1`** *(float, 0–1)*  
-  Urgency level affecting inaction cost.
+  Task urgency affecting the cost of inaction.
 
 ---
 
 ### Mixed-initiative decision
 
 - **`lookout_modality`** *(categorical)*  
-  `manual_invocation`, `explicit_agent`, or `auto_scoping`.
+  Interaction mode: `manual_invocation`, `explicit_agent`, or `auto_scoping`.
 
 - **`agent_action`** *(categorical)*  
-  `no_action`, `action`, `dialog`, or `scope`.
+  Chosen action: `no_action`, `action`, `dialog`, or `scope`.
 
 ---
 
@@ -205,10 +189,10 @@ Each row represents a **single mixed-initiative decision episode**.
 ### Dialog and refinement
 
 - **`dialog_confirmed`** *(binary)*  
-  Dialog acceptance indicator.
+  User accepted dialog suggestion.
 
 - **`refine_after_accept`** *(binary)*  
-  Post-acceptance refinement request.
+  User requested refinement after acceptance.
 
 ---
 
@@ -218,27 +202,34 @@ Each row represents a **single mixed-initiative decision episode**.
   `accept`, `reject`, or `ignore`.
 
 - **`realized_utility`** *(float)*  
-  Utility actually realized.
+  Utility actually realized after outcome sampling.
 
 - **`u_baseline_no_action`** *(float)*  
-  Utility under no-action baseline.
+  Utility that would have occurred under no action.
 
 ---
 
 ### Expected utilities
 
-- **`eu_no_action`**  
-- **`eu_action`**  
-- **`eu_dialog`**  
-- **`eu_scope`**
+These fields store the **expected utility values computed at decision time**, before sampling the user response. They are the quantities used by the policy to choose an action.
+
+- **`eu_no_action`** *(float)*  
+  Expected utility of taking **no action**, balancing avoided interruption cost against delay or missed-opportunity cost.
+
+- **`eu_action`** *(float)*  
+  Expected utility of taking the **primary action immediately**, combining benefit, error penalty, and interruption cost.
+
+- **`eu_dialog`** *(float)*  
+  Expected utility of initiating **dialog or clarification**, trading lower risk for a chance to improve correctness.
 
-All values represent expected utility under the model.
+- **`eu_scope`** *(float)*  
+  Expected utility of taking a **scoped or partial action**, providing reduced risk with partial benefit.
 
 ---
 
 ### Analytic decision thresholds
 
-Minimum belief probability where one action dominates another:
+Each threshold represents the **minimum belief probability** at which one action dominates another in expected utility.
 
 - **`p_star_noaction_action`**
 - **`p_star_noaction_dialog`**
@@ -253,7 +244,7 @@ All thresholds are analytic and derived directly from the utility model.
 
 ## Quality report
 
-The file `mixed_initiative_traces.quality_report.txt` documents calibration, sensitivity, and policy comparisons.
+The file `mixed_initiative_traces.quality_report.txt` documents calibration results, sensitivity checks, and policy comparisons.
 
 ---