diff --git a/bin/cli/run_signal_analysis.py b/bin/cli/run_signal_analysis.py
index f039e630..dc537814 100644
--- a/bin/cli/run_signal_analysis.py
+++ b/bin/cli/run_signal_analysis.py
@@ -26,8 +26,12 @@
 from looptrace.SpotPicker import SpotPicker
 from looptrace.utilities import find_first_option, get_either, wrap_exception, wrap_error_message
 
+FieldOfViewName: TypeAlias = str
+RawTimepoint: TypeAlias = int
+
 _A = TypeVar("_A")
 _B = TypeVar("_B")
+_K = TypeVar("_K")
 
 SIGNAL_CHANNEL_COLUMN = "signalChannel"
 
@@ -153,16 +157,21 @@ def workflow(
 
                     spot_drift_file: Path = H.drift_correction_file__coarse
                     logging.info("Reading nuclei drift file: %s", spot_drift_file)
-                    all_spot_drifts: pd.DataFrame = pd.read_csv(spot_drift_file, index_col=False)
+                    all_spot_drifts: Mapping[tuple[FieldOfViewName, RawTimepoint], DriftRecord] = read_spot_drifts_file(spot_drift_file)
 
                     nuclei_drift_file: Path = H.nuclei_coarse_drift_correction_file
                     logging.info("Reading nuclei drift file: %s", nuclei_drift_file)
-                    all_nuclei_drifts: pd.DataFrame = pd.read_csv(nuclei_drift_file, index_col=False)
+                    all_nuclei_drifts: Mapping[FieldOfViewName, DriftRecord] = read_signal_drifts_file(nuclei_drift_file)
 
                     # TODO: discard regional spot timepoints from the bigger collection: https://github.com/gerlichlab/looptrace/issues/376
                     for spec in analysis_specs:
                         # Get the ROIs of this type.
                         roi_type: RoiType = spec.roi_type
+                        if roi_type == RoiType.LocusSpecific:
+                            # TODO: will need to account for pixels vs. nanometers
+                            # TODO: will need to account for different headers (e.g., z_px and z rather than zc, yc, etc.)
+                            logging.error("Cross-channel analysis for locus-specific spots isn't yet supported, skipping!")
+                            continue
                         logging.info("Analyzing signal for ROI type '%s'", roi_type.name)
                         rois_file: Path = getattr(H, roi_type.file_attribute_on_image_handler)
                         all_rois: pd.DataFrame = pd.read_csv(rois_file, index_col=False)
@@ -170,21 +179,24 @@ def workflow(
                         # Build up the records for this ROI type, for all FOVs.
                         by_raw_channel: Mapping[int, list[dict]] = defaultdict
                         # TODO: refactor with https://github.com/gerlichlab/gertils/issues/32
-                        for fov, img in S.iter_fov_img_pairs():
+                        for fov, image_stack in S.iter_fov_img_pairs():
                             logging.info(f"Analysing signal for FOV: {fov}")
-                            nuc_drift_curr_fov: pd.DataFrame = all_nuclei_drifts[all_nuclei_drifts[FIELD_OF_VIEW_COLUMN] == fov]
-                            logging.debug(f"Shape of nuclei drifts ({type(nuc_drift_curr_fov).__name__}): {nuc_drift_curr_fov.shape}")
-                            spot_drifts_curr_fov: pd.DataFrame = all_spot_drifts[all_spot_drifts[FIELD_OF_VIEW_COLUMN] == fov]
-                            logging.debug(f"Shape of spot drifts ({type(spot_drifts_curr_fov).__name__}): {spot_drifts_curr_fov.shape}")
+                            nuc_drift: DriftRecord = all_nuclei_drifts[fov]
                             rois: pd.DataFrame = all_rois[all_rois[FIELD_OF_VIEW_COLUMN] == fov]
                             logging.debug("ROI count: %d", rois.shape[0])
+                            print("ROI count: {rois.shape[0]}") # DEBUG
                             for _, r in rois.iterrows():
-                                spot_drift = spot_drifts_curr_fov[spot_drifts_curr_fov[TIMEPOINT_COLUMN] == r[TIMEPOINT_COLUMN]]
+                                timepoint: RawTimepoint = r[TIMEPOINT_COLUMN]
+                                img = image_stack[timepoint]
+                                print(f"img dim: {img.shape}") # DEBUG
+                                spot_drift: DriftRecord = all_spot_drifts[(fov, timepoint)]
+                                print(f"Spot drift ({type(spot_drift).__name__}): {spot_drift}") # DEBUG
                                 pt0: ImagePoint3D = get_centroid_from_record(r)
+                                print(f"Point: {pt0}") # DEBUG
                                 dc_pt: ImagePoint3D = ImagePoint3D(
-                                    z=pt0.z - nuc_drift_curr_fov[Z_PX_COARSE] + spot_drift[Z_PX_COARSE], 
-                                    y=pt0.y - nuc_drift_curr_fov[Y_PX_COARSE] + spot_drift[Y_PX_COARSE], 
-                                    x=pt0.x - nuc_drift_curr_fov[X_PX_COARSE] + spot_drift[X_PX_COARSE], 
+                                    z=pt0.z - nuc_drift.z + spot_drift.z, 
+                                    y=pt0.y - nuc_drift.y + spot_drift.y, 
+                                    x=pt0.x - nuc_drift.x + spot_drift.x, 
                                 )
                                 for stats in compute_pixel_statistics(
                                     img=img,
@@ -223,6 +235,36 @@ def proc1(acc: State, a: _A) -> State:
     return Seq.of_iterable(inputs).fold(proc1, Result.Ok(Seq()))
 
 
+@dataclass(kw_only=True, frozen=True)
+class DriftRecord:
+    x: float
+    y: float
+    z: float
+
+
+@curry_flip(1)
+def read_drift_file(drift_file: Path, get_key: Callable[[pd.Series], _K]) -> Mapping[_K, "DriftRecord"]:
+    drifts: pd.DataFrame = pd.read_csv(drift_file, index_col=False)
+    colnames = {Z_PX_COARSE: "z", Y_PX_COARSE: "y", X_PX_COARSE: "x"}
+    result: Mapping[_K, DriftRecord] = {}
+    for key, rec in drifts.rename(columns=colnames).apply(
+        lambda row: (get_key(row), DriftRecord(**row[colnames.values()].to_dict())), 
+        axis=1,
+    ):
+        if key in result:
+            raise ValueError(f"Repeated key in drift file ({drift_file}): {key}")
+        result[key] = rec
+    return result
+    
+
+read_signal_drifts_file: Callable[[Path], Mapping[FieldOfViewName, "DriftRecord"]] = \
+    read_drift_file(lambda row: row[FIELD_OF_VIEW_COLUMN])
+
+read_spot_drifts_file: Callable[[Path], Mapping[tuple[FieldOfViewName, RawTimepoint], "DriftRecord"]] = \
+    read_drift_file(lambda row: (row[FIELD_OF_VIEW_COLUMN], row[TIMEPOINT_COLUMN]))
+
+
+
 def _ensure_unique(items: Iterable[_A]) -> Result[set[_A], str]:
     try:
         counts = Counter(items)