Upload 12 files

.gitattributes

@@ -57,3 +57,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+chi99horvitz.pdf filter=lfs diff=lfs merge=lfs -text
+mixed_initiative_traces.csv filter=lfs diff=lfs merge=lfs -text
+mixed_initiative_traces.jsonl filter=lfs diff=lfs merge=lfs -text

chi99horvitz.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f2596af326e186cc1eb27da5d0f0beda8d2ff15794b82958788c636693938d2
+size 221004

mixed_initiative_traces.config.json

@@ -0,0 +1,67 @@
+{
+  "seed": 13,
+  "n_users": 40,
+  "n_episodes": 50000,
+  "output_dir": "output",
+  "file_prefix": "mixed_initiative_traces",
+  "msg_len_min": 50,
+  "msg_len_max": 2500,
+  "dwell_min_sec": 2.0,
+  "dwell_max_sec": 9.0,
+  "dwell_center_bytes": 900.0,
+  "dwell_width_bytes": 220.0,
+  "dwell_noise_std": 0.6,
+  "early_check_prob": 0.35,
+  "early_offset_std": 2.4,
+  "late_offset_std": 1.8,
+  "base_goal_rate": 0.18,
+  "evidence_strength": 2.15,
+  "evidence_noise_std": 0.35,
+  "user_logit_shift_std": 0.2,
+  "p_clip": 0.0001,
+  "model_logit_temperature": 1.05,
+  "model_logit_noise_std": 0.2,
+  "model_logit_bias": 0.0,
+  "u_no_action_goal": 0.35,
+  "u_no_action_no_goal": 0.7,
+  "u_action_goal": 0.95,
+  "u_action_no_goal_base": 0.22,
+  "u_dialog_goal": 0.75,
+  "u_dialog_no_goal_base": 0.55,
+  "u_scope_goal": 0.62,
+  "u_scope_no_goal_base": 0.62,
+  "penalty_action_no_goal_max": 0.55,
+  "penalty_dialog_no_goal_max": 0.25,
+  "penalty_scope_no_goal_max": 0.1,
+  "low_urgency_penalty_action": 0.12,
+  "low_urgency_penalty_dialog": 0.05,
+  "low_urgency_penalty_scope": 0.03,
+  "high_urgency_bonus_action": 0.05,
+  "high_urgency_bonus_dialog": 0.03,
+  "high_urgency_bonus_scope": 0.02,
+  "ignore_base": 0.05,
+  "ignore_if_not_ready_boost": 0.35,
+  "accept_if_goal_action": 0.9,
+  "accept_if_no_goal_action": 0.1,
+  "accept_if_goal_dialog": 0.82,
+  "accept_if_no_goal_dialog": 0.18,
+  "accept_if_goal_scope": 0.88,
+  "accept_if_no_goal_scope": 0.35,
+  "refine_if_accept": 0.35,
+  "realized_utility_noise_std": 0.03,
+  "modality_manual_prob": 0.25,
+  "modality_explicit_agent_prob": 0.45,
+  "modality_auto_scoping_prob": 0.3,
+  "baseline_p_dialog": 0.3,
+  "baseline_p_action": 0.7,
+  "calibration_bins": 10,
+  "report_bins_target": 12,
+  "report_min_bin_size": 500,
+  "report_top_lines": 12,
+  "not_ready_zero_epsilon": 1e-12,
+  "slope_alpha": 0.05,
+  "slope_warn_alpha": 0.2,
+  "make_plots_if_possible": true,
+  "fig3_bins_p": 25,
+  "fig3_bins_nr": 20
+}

mixed_initiative_traces.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:505495f860e7537e140359ab07511a9c0d9d1d971ec114bde7e6cf96a5abc320
+size 13783084

mixed_initiative_traces.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:798d3fc0e8e74c0b1ffeaf020de627a60030f9ae995ef00771e2bc7f5bf3bd3a
+size 52173496

mixed_initiative_traces.quality_report.txt

@@ -0,0 +1,110 @@
+LookOut-inspired Mixed-Initiative Engine — Quality Report (v1.0)
+Developed by DBbun LLC — January 2026
+Generated: 2026-01-11 19:11:06
+
+Files
+- CSV:   output\mixed_initiative_traces.csv
+- JSONL: output\mixed_initiative_traces.jsonl
+- Config:output\mixed_initiative_traces.config.json
+- Report:output\mixed_initiative_traces.quality_report.txt
+- Figures:
+  • output/fig1_manual_invocation_summary.png
+  • output/fig2_explicit_agent_flow.png
+  • output/fig3_auto_scoping_policy_map.png
+  • output/fig4_threshold_action_vs_noaction.png
+  • output/fig5_context_shifted_thresholds.png
+  • output/fig6_thresholds_action_dialog_noaction.png
+  • output/fig7_dwell_time_vs_message_length.png
+
+1) Distributions
+- rows: 50,000
+Modality distribution:
+- auto_scoping: 14,950 (29.90%)
+- explicit_agent: 22,346 (44.69%)
+- manual_invocation: 12,704 (25.41%)
+Action distribution:
+- action: 9,932 (19.86%)
+- dialog: 6,404 (12.81%)
+- no_action: 32,132 (64.26%)
+- scope: 1,532 (3.06%)
+Response distribution:
+- accept: 10,387 (20.77%)
+- ignore: 1,406 (2.81%)
+- none: 32,132 (64.26%)
+- reject: 6,075 (12.15%)
+
+2) Uncertainty quality
+- AUROC(p_true):  0.8877  (calibrated-by-construction via latent evidence)
+- AUROC(p_model): 0.8856  => PASS
+- ECE(p_model):   0.0058  => PASS
+
+Reliability (p_model): avg_pred -> empirical_rate (n)
+- 0.038 -> 0.034 (n=18255)
+- 0.146 -> 0.142 (n=7206)
+- 0.248 -> 0.239 (n=4902)
+- 0.349 -> 0.349 (n=3752)
+- 0.449 -> 0.437 (n=3137)
+- 0.549 -> 0.546 (n=2824)
+- 0.649 -> 0.654 (n=2509)
+- 0.749 -> 0.768 (n=2509)
+- 0.850 -> 0.864 (n=2477)
+- 0.948 -> 0.954 (n=2429)
+
+3) Sensitivity tests
+Action sensitivity to p_model (should be +):
+- slope: +1.120359, z: +326.111, p: 0.000e+00 => PASS
+Action sensitivity to not_ready (should be -):
+- slope: -0.022279, z: -2.091, p: 3.654e-02 => PASS
+- not_ready health (p75): 0.128 => PASS
+
+4) Fig 7 alignment (dwell time vs message length)
+- corr(message_length, dwell_time): 0.9146 (should be positive)
+
+5) Fig 1/2 behavioral summaries (data-driven)
+- Fig 1 manual_invocation n=12704: click=3392 (26.70%), hover_only=0 (0.00%), no_action=9312 (73.30%)
+- Fig 2 explicit_agent(dialog) n=4369: accept=1947 (44.56%), refine|accept=34.98%
+
+6) Binned curves (interpretability)
+P(action | p_model):
+- p≈0.006: 0.00% (n=4166, [0.000,0.011])
+- p≈0.020: 0.00% (n=4166, [0.011,0.028])
+- p≈0.040: 0.00% (n=4166, [0.028,0.052])
+- p≈0.068: 0.00% (n=4166, [0.052,0.084])
+- p≈0.107: 0.00% (n=4166, [0.084,0.129])
+- p≈0.161: 0.00% (n=4166, [0.129,0.192])
+- p≈0.232: 0.00% (n=4166, [0.192,0.272])
+- p≈0.325: 0.00% (n=4166, [0.272,0.378])
+- p≈0.443: 5.50% (n=4166, [0.378,0.508])
+- p≈0.585: 33.85% (n=4166, [0.508,0.662])
+- p≈0.746: 98.87% (n=4166, [0.662,0.829])
+- p≈0.914: 100.00% (n=4174, [0.829,0.999])
+
+P(ignore | not_ready):
+- nr≈0.000: 1.81% (n=32404, [0.000,0.000])
+- nr≈0.019: 2.52% (n=1466, [0.000,0.037])
+- nr≈0.055: 1.84% (n=1466, [0.037,0.073])
+- nr≈0.092: 3.48% (n=1466, [0.073,0.111])
+- nr≈0.129: 3.48% (n=1466, [0.111,0.146])
+- nr≈0.167: 4.02% (n=1466, [0.146,0.187])
+- nr≈0.207: 3.41% (n=1466, [0.187,0.227])
+- nr≈0.249: 3.82% (n=1466, [0.227,0.270])
+- nr≈0.294: 4.91% (n=1466, [0.270,0.318])
+- nr≈0.345: 6.62% (n=1466, [0.318,0.373])
+- nr≈0.407: 5.93% (n=1466, [0.373,0.440])
+- nr≈0.493: 6.68% (n=1466, [0.440,0.547])
+
+7) Net benefit vs baselines (counterfactual expected utility)
+B0 always no_action: mean Δ=+0.0713, median Δ=+0.0037, P(Δ>0)=53.40%
+B1 dialog if p≥0.30: mean Δ=+0.0118, median Δ=+0.0056, P(Δ>0)=54.73%
+B2 action if p≥0.70 else dialog if p≥0.30: mean Δ=-0.0016, median Δ=+0.0056, P(Δ>0)=54.73%
+- Beats B0: mean Δ=+0.0713 => PASS
+
+8) Threshold provenance (Figs 4–6)
+- Per-episode thresholds p* are computed analytically from context-shifted utilities and stored in the dataset.
+
+9) Quick rubric
+- ECE(p_model): PASS
+- AUROC(p_model): PASS
+- Sensitivity p_model: PASS
+- Sensitivity not_ready: PASS
+- Beats B0: PASS