fix: gross-based batch discard, direction-correct event logs, wire CSV logger

docs/architecture.md

@@ -116,7 +116,11 @@ stats panel (Loading / Unloading / Net / last-3-batch history), and a bottom sta
   do not add keys here.
 - `src/logger.py` — `CSVLogger` appends `batch_summary.csv`
   (`batch_id,start,end,duration,loading,unloading,net`) and `sack_events.csv`
-  (`timestamp,batch_id,sack_crossed,T-<id>,direction`).
+  (`timestamp,batch_id,sack_crossed,T-<id>,direction`). **Wired into
+  `predict.py`**: one `log_event` per crossing in both event loops, one
+  `log_batch` per kept batch (`finalize_batch` + manual dashboard stop);
+  files live in `output.dir` next to `jetson_counter.db`, all calls
+  best-effort (warn on failure, never crash the pipeline).
 
 ## Production pipeline (`predict.py`)
 
@@ -142,8 +146,9 @@ dashboard integration is file-based.)
 
 `predict.py` reads mode every frame: `auto` → FSM branch; any other valid value
 → operator/state-file branch (persists `count`, `box_count`, `box_unloading`,
-**`unloading`**). Finalize discard = **both nets 0**
-(`net_sack = loading − unloading`, `net_box = box_loading − box_unloading`).
+**`unloading`**). Finalize discard = **both gross counts 0** (`count == 0` and
+`box_loading == 0`). Stored nets: `net_sack = loading − unloading`,
+`net_box = box_loading − box_unloading`.
 
 DO pipeline (dashboard): smartphone upload → OCR draft → edit/stage → start
 gates → active batch → stop-preview soft-warn → stop → `batches` + DO columns.