Rebrand TRACE -> SPARK

- Full name: 'Simulation-based Posterior Amortization for Reaction Kinetics'
- Class TRACEPredictor -> SPARKPredictor; env vars TRACE_* -> SPARK_*
- Updated README, app.py title/info, all docstrings and comments.
- No model architecture or checkpoint format changes.

Co-authored-by: Cursor <cursoragent@cursor.com>

README.md

@@ -1,5 +1,5 @@
 ---
-title: TRACE
+title: SPARK
 emoji: ⚡
 colorFrom: blue
 colorTo: purple
@@ -8,10 +8,10 @@ sdk_version: "5.29.0"
 app_file: app.py
 pinned: false
 license: mit
-short_description: TRACE — Bayesian inference for CV & TPD analysis
+short_description: SPARK — Bayesian inference for CV & TPD analysis
 ---
 
-# TRACE — Trace Recognition and Amortized Conditional Estimation
+# SPARK — Simulation-based Posterior Amortization for Reaction Kinetics
 
 Amortized Bayesian inference for electrochemistry and catalysis.
 Upload cyclic voltammetry (CV) or temperature-programmed desorption (TPD) data to automatically:

app.py

@@ -1,5 +1,5 @@
 """
-TRACE (Trace Recognition and Amortized Conditional Estimation)
+SPARK (Simulation-based Posterior Amortization for Reaction Kinetics)
 
 Gradio web interface for mechanism classification and parameter inference
 from cyclic voltammetry (CV) and temperature-programmed desorption (TPD) data.
@@ -16,7 +16,7 @@ import gradio as gr
 
 
 
-from inference import TRACEPredictor
+from inference import SPARKPredictor
 from preprocessing import (
     nondimensionalize_cv,
     estimate_E0,
@@ -41,27 +41,27 @@ DEMO_RENDERS = REPO_ROOT / "demo_renders"
 EC_CHECKPOINT = REPO_ROOT / "checkpoints" / "ec_best.pt"
 TPD_CHECKPOINT = REPO_ROOT / "checkpoints" / "tpd_best.pt"
 
-# Allow override via environment variables. TRACE_*_CHECKPOINT is the
+# Allow override via environment variables. SPARK_*_CHECKPOINT is the
 # canonical name; ECFLOW_*_CHECKPOINT remains accepted for backward
 # compatibility with existing HF Space secrets.
 EC_CHECKPOINT = Path(
     os.environ.get(
-        "TRACE_EC_CHECKPOINT",
+        "SPARK_EC_CHECKPOINT",
         os.environ.get("ECFLOW_EC_CHECKPOINT", str(EC_CHECKPOINT)),
     )
 )
 TPD_CHECKPOINT = Path(
     os.environ.get(
-        "TRACE_TPD_CHECKPOINT",
+        "SPARK_TPD_CHECKPOINT",
         os.environ.get("ECFLOW_TPD_CHECKPOINT", str(TPD_CHECKPOINT)),
     )
 )
 
 # Optional image-input checkpoints. When set and the file exists, the
-# "From Image" tabs use the image-input TRACE model directly instead of
+# "From Image" tabs use the image-input SPARK model directly instead of
 # digitizing the plot. If not set, the digitizer fallback is used.
-EC_IMAGE_CHECKPOINT_PATH = os.environ.get("TRACE_EC_IMAGE_CHECKPOINT", "")
-TPD_IMAGE_CHECKPOINT_PATH = os.environ.get("TRACE_TPD_IMAGE_CHECKPOINT", "")
+EC_IMAGE_CHECKPOINT_PATH = os.environ.get("SPARK_EC_IMAGE_CHECKPOINT", "")
+TPD_IMAGE_CHECKPOINT_PATH = os.environ.get("SPARK_TPD_IMAGE_CHECKPOINT", "")
 EC_IMAGE_CHECKPOINT = (
     Path(EC_IMAGE_CHECKPOINT_PATH) if EC_IMAGE_CHECKPOINT_PATH else None
 )
@@ -73,8 +73,8 @@ TPD_IMAGE_CHECKPOINT = (
 # "From Image" tabs expose a "Joint image+waveform (Phase 2)" inference
 # strategy that fuses the rendered image with the digitizer-extracted
 # waveform inside a single encoder. Falls back to ensemble if missing.
-EC_JOINT_CHECKPOINT_PATH = os.environ.get("TRACE_EC_JOINT_CHECKPOINT", "")
-TPD_JOINT_CHECKPOINT_PATH = os.environ.get("TRACE_TPD_JOINT_CHECKPOINT", "")
+EC_JOINT_CHECKPOINT_PATH = os.environ.get("SPARK_EC_JOINT_CHECKPOINT", "")
+TPD_JOINT_CHECKPOINT_PATH = os.environ.get("SPARK_TPD_JOINT_CHECKPOINT", "")
 EC_JOINT_CHECKPOINT = (
     Path(EC_JOINT_CHECKPOINT_PATH) if EC_JOINT_CHECKPOINT_PATH else None
 )
@@ -245,7 +245,7 @@ def get_predictor():
         tpd_joint = (str(TPD_JOINT_CHECKPOINT)
                      if (TPD_JOINT_CHECKPOINT and TPD_JOINT_CHECKPOINT.exists())
                      else None)
-        predictor = TRACEPredictor(
+        predictor = SPARKPredictor(
             ec_checkpoint=ec_ckpt,
             tpd_checkpoint=tpd_ckpt,
             ec_image_checkpoint=ec_img,
@@ -359,7 +359,7 @@ def analyze_cv_image(files, scan_rate_text, E0_V, threshold, current_unit,
                      method_label="Ensemble (recommended)"):
     """Analyze CV from uploaded plot images (one per scan rate).
 
-    When the image-input TRACE model is available we run a hybrid pipeline:
+    When the image-input SPARK model is available we run a hybrid pipeline:
     image-mode prediction in parallel with the digitizer + waveform-mode path,
     and combine via `method_label` (Ensemble / Image-direct / Digitize-then-infer
     / Auto-fallback). When only the waveform model is available, falls back
@@ -477,7 +477,7 @@ def analyze_cv_image(files, scan_rate_text, E0_V, threshold, current_unit,
     )
 
 
-OOD_THRESHOLD = float(os.environ.get("TRACE_OOD_THRESHOLD", "0.5"))
+OOD_THRESHOLD = float(os.environ.get("SPARK_OOD_THRESHOLD", "0.5"))
 
 
 def _ood_banner_update(ood_score, threshold=OOD_THRESHOLD):
@@ -636,7 +636,7 @@ def _method_comparison_html(hybrid_out, threshold=OOD_THRESHOLD):
                  "methods disagree \u2014 review carefully</span>")
 
     footnote = (
-        "Image-direct (Phase 1) is the image-only TRACE; Digitize-then-infer "
+        "Image-direct (Phase 1) is the image-only SPARK; Digitize-then-infer "
         "uses the headline waveform model on a curve extracted from the plot; "
         "Joint (Phase 2) fuses both signals inside a single encoder trained "
         "with real-world distortion augmentation. When available, the joint "
@@ -1254,7 +1254,7 @@ footer { display: none !important; }
 
 def build_app():
     with gr.Blocks(
-        title="TRACE — Bayesian Inference for Electrochemistry & Catalysis",
+        title="SPARK — Bayesian Inference for Electrochemistry & Catalysis",
         theme=gr.themes.Soft(
             primary_hue="blue",
             secondary_hue="slate",
@@ -1265,7 +1265,7 @@ def build_app():
         gr.HTML("<div class='trace-page'>")
         gr.HTML(
             "<div class='main-header'>"
-            "<h1><span class='accent'>TRACE</span></h1>"
+            "<h1><span class='accent'>SPARK</span></h1>"
             "<p><strong>T</strong>race <strong>R</strong>ecognition and "
             "<strong>A</strong>mortized <strong>C</strong>onditional "
             "<strong>E</strong>stimation &mdash; Bayesian inference for "
@@ -1757,7 +1757,7 @@ def build_app():
                         gr.Markdown("""
 ## How it works
 
-**TRACE** (Trace Recognition and Amortized Conditional Estimation) uses
+**SPARK** (Simulation-based Posterior Amortization for Reaction Kinetics) uses
 **conditional normalizing flows** with a **Set Transformer** encoder to
 perform amortized Bayesian inference. Given one or more experimental
 curves, it simultaneously classifies the reaction mechanism and produces
@@ -1778,7 +1778,7 @@ weighting.
 ### Citation
 
 ```
-Yan, B. (2026). TRACE: Amortized Bayesian Inference for
+Yan, B. (2026). SPARK: Amortized Bayesian Inference for
 Mechanism Identification and Parameter Estimation in
 Electrochemistry and Catalysis via Conditional
 Normalizing Flows. [Preprint]

image_encoder.py

@@ -1,5 +1,5 @@
 """
-Image encoder for the image-input TRACE variants (CV and TPD).
+Image encoder for the image-input SPARK variants (CV and TPD).
 
 Replaces the per-scan 1-D `SignalEncoder` with a small 2-D CNN that maps a
 single rasterized plot image (grayscale, 224x224 by default) to a context

image_preprocessing.py

@@ -1,5 +1,5 @@
 """
-Image preprocessing for image-input TRACE on real-world uploads.
+Image preprocessing for image-input SPARK on real-world uploads.
 
 Real user uploads (paper-figure crops, software screenshots, photos of
 lab monitors) live in a much wider image distribution than the rendered
@@ -250,7 +250,7 @@ def prepare_for_image_mode(
     do_clean: bool = True,
     target_size: int = 224,
 ) -> Tuple[Image.Image, Dict[str, object]]:
-    """Full preprocessing pipeline for image-mode TRACE.
+    """Full preprocessing pipeline for image-mode SPARK.
 
     Steps (any can be skipped):
         crop_to_plot_region -> remove_gridlines_and_background -> resize.

inference.py

@@ -1,5 +1,5 @@
 """
-TRACE inference engine.
+SPARK inference engine.
 
 Loads trained EC and TPD models and runs end-to-end inference
 from preprocessed arrays (dimensionless for CV, physical for TPD).
@@ -37,7 +37,7 @@ def _fix_actnorm_initialized(model):
             module.initialized = True
 
 
-class TRACEPredictor:
+class SPARKPredictor:
     """Unified predictor for both EC (cyclic voltammetry) and TPD domains."""
 
     def __init__(self, ec_checkpoint=None, tpd_checkpoint=None, device=None,
@@ -145,7 +145,7 @@ class TRACEPredictor:
             k for k in unexpected if not k.endswith("_initialized")
         ]
         if suspicious_missing or suspicious_unexpected:
-            print(f"[TRACEPredictor] WARNING: state_dict mismatch on EC ckpt.")
+            print(f"[SPARKPredictor] WARNING: state_dict mismatch on EC ckpt.")
             if suspicious_missing:
                 print(f"  missing   ({len(suspicious_missing)}): "
                       f"{suspicious_missing[:6]}{' ...' if len(suspicious_missing) > 6 else ''}")
@@ -231,7 +231,7 @@ class TRACEPredictor:
             k for k in unexpected if not k.endswith("_initialized")
         ]
         if suspicious_missing or suspicious_unexpected:
-            print(f"[TRACEPredictor] WARNING: state_dict mismatch on TPD ckpt.")
+            print(f"[SPARKPredictor] WARNING: state_dict mismatch on TPD ckpt.")
             if suspicious_missing:
                 print(f"  missing   ({len(suspicious_missing)}): "
                       f"{suspicious_missing[:6]}{' ...' if len(suspicious_missing) > 6 else ''}")
@@ -338,7 +338,7 @@ class TRACEPredictor:
         suspicious_missing = [k for k in missing if not k.endswith("_initialized")]
         suspicious_unexpected = [k for k in unexpected if not k.endswith("_initialized")]
         if suspicious_missing or suspicious_unexpected:
-            print("[TRACEPredictor] WARNING: state_dict mismatch on EC image ckpt.")
+            print("[SPARKPredictor] WARNING: state_dict mismatch on EC image ckpt.")
             if suspicious_missing:
                 print(f"  missing   ({len(suspicious_missing)}): "
                       f"{suspicious_missing[:6]}{' ...' if len(suspicious_missing) > 6 else ''}")
@@ -393,7 +393,7 @@ class TRACEPredictor:
         suspicious_missing = [k for k in missing if not k.endswith("_initialized")]
         suspicious_unexpected = [k for k in unexpected if not k.endswith("_initialized")]
         if suspicious_missing or suspicious_unexpected:
-            print("[TRACEPredictor] WARNING: state_dict mismatch on TPD image ckpt.")
+            print("[SPARKPredictor] WARNING: state_dict mismatch on TPD image ckpt.")
             if suspicious_missing:
                 print(f"  missing   ({len(suspicious_missing)}): "
                       f"{suspicious_missing[:6]}{' ...' if len(suspicious_missing) > 6 else ''}")
@@ -447,7 +447,7 @@ class TRACEPredictor:
         suspicious_missing = [k for k in missing if not k.endswith("_initialized")]
         suspicious_unexpected = [k for k in unexpected if not k.endswith("_initialized")]
         if suspicious_missing or suspicious_unexpected:
-            print("[TRACEPredictor] WARNING: state_dict mismatch on EC joint ckpt.")
+            print("[SPARKPredictor] WARNING: state_dict mismatch on EC joint ckpt.")
             if suspicious_missing:
                 print(f"  missing   ({len(suspicious_missing)}): "
                       f"{suspicious_missing[:6]}{' ...' if len(suspicious_missing) > 6 else ''}")
@@ -503,7 +503,7 @@ class TRACEPredictor:
         suspicious_missing = [k for k in missing if not k.endswith("_initialized")]
         suspicious_unexpected = [k for k in unexpected if not k.endswith("_initialized")]
         if suspicious_missing or suspicious_unexpected:
-            print("[TRACEPredictor] WARNING: state_dict mismatch on TPD joint ckpt.")
+            print("[SPARKPredictor] WARNING: state_dict mismatch on TPD joint ckpt.")
             if suspicious_missing:
                 print(f"  missing   ({len(suspicious_missing)}): "
                       f"{suspicious_missing[:6]}{' ...' if len(suspicious_missing) > 6 else ''}")
@@ -900,7 +900,7 @@ class TRACEPredictor:
                     n_samples=n_samples, temperature=temperature,
                 )
             except Exception as exc:
-                print(f"[TRACEPredictor] image-mode CV failed: {exc}")
+                print(f"[SPARKPredictor] image-mode CV failed: {exc}")
 
         if self.ec_model is not None and potentials is not None and fluxes is not None:
             try:
@@ -909,7 +909,7 @@ class TRACEPredictor:
                     n_samples=n_samples, temperature=temperature,
                 )
             except Exception as exc:
-                print(f"[TRACEPredictor] waveform CV failed: {exc}")
+                print(f"[SPARKPredictor] waveform CV failed: {exc}")
 
         if (self.has_ec_joint_model and preprocessed is not None
                 and potentials is not None and fluxes is not None):
@@ -919,7 +919,7 @@ class TRACEPredictor:
                     n_samples=n_samples, temperature=temperature,
                 )
             except Exception as exc:
-                print(f"[TRACEPredictor] joint CV failed: {exc}")
+                print(f"[SPARKPredictor] joint CV failed: {exc}")
 
         return image_result, waveform_result, joint_result, preproc_meta
 
@@ -1014,7 +1014,7 @@ class TRACEPredictor:
                     n_samples=n_samples, temperature=temperature,
                 )
             except Exception as exc:
-                print(f"[TRACEPredictor] image-mode TPD failed: {exc}")
+                print(f"[SPARKPredictor] image-mode TPD failed: {exc}")
 
         if self.tpd_model is not None and temperatures is not None and rates is not None:
             try:
@@ -1023,7 +1023,7 @@ class TRACEPredictor:
                     n_samples=n_samples, temperature=temperature,
                 )
             except Exception as exc:
-                print(f"[TRACEPredictor] waveform TPD failed: {exc}")
+                print(f"[SPARKPredictor] waveform TPD failed: {exc}")
 
         if (self.has_tpd_joint_model and preprocessed is not None
                 and temperatures is not None and rates is not None):
@@ -1033,7 +1033,7 @@ class TRACEPredictor:
                     n_samples=n_samples, temperature=temperature,
                 )
             except Exception as exc:
-                print(f"[TRACEPredictor] joint TPD failed: {exc}")
+                print(f"[SPARKPredictor] joint TPD failed: {exc}")
 
         return image_result, waveform_result, joint_result, preproc_meta
 

plotting.py

@@ -1,5 +1,5 @@
 """
-Visualization utilities for the TRACE web app.
+Visualization utilities for the SPARK web app.
 
 Generates matplotlib figures for mechanism classification, parameter
 posteriors, and signal reconstruction overlays.

preprocessing.py

@@ -1,5 +1,5 @@
 """
-Preprocessing utilities for TRACE web app.
+Preprocessing utilities for SPARK web app.
 
 Handles:
 - CSV/NPZ parsing for both CV and TPD data
@@ -24,7 +24,7 @@ def nondimensionalize_cv(E_volts, i_amps, v_Vs, E0_V, T_K=298.15,
                          A_cm2=0.0707, C_A_molcm3=1e-6, D_A_cm2s=1e-5, n=1,
                          v_ref_Vs=0.1):
     """
-    Convert physical CV data to dimensionless units for the TRACE model.
+    Convert physical CV data to dimensionless units for the SPARK model.
 
     Potential and current are nondimensionalized using the Compton convention
     with the scan-rate-dependent diffusion length d = sqrt(D·RT/(nFv)):
@@ -33,7 +33,7 @@ def nondimensionalize_cv(E_volts, i_amps, v_Vs, E0_V, T_K=298.15,
 
     The dimensionless scan rate σ = v / v_ref is computed separately.
     In the Compton convention σ ≡ 1 by construction (d absorbs v), but the
-    TRACE model uses σ as an explicit conditioning variable to distinguish
+    SPARK model uses σ as an explicit conditioning variable to distinguish
     experiments at different scan rates. Setting v_ref so that σ spans the
     training range (~0.1–100) gives the model the scan-rate information.