Stop prefiltering avg_dollar_volume on a stale price

avg_dollar_volume was marked STATIC_SAFE, meaning a screen like
`avgdvol > 20m` narrowed the universe against the snapshot before any
quote came back. But the field is price times avg_volume_3m, and price
is the one live input in that product — a stock trading through the
threshold since the snapshot was taken would get dropped before it
ever had a chance to be repriced. That's the two-clocks-in-one-number
mistake pct_52w_range already avoids by staying off STATIC_SAFE; this
field was doing the same math and hadn't gotten the same treatment.
It was already in LIVE_COLUMNS, so no repricing logic changes, only
which bucket a screen condition on it lands in.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01JRdJUc8CJPpH9hHjJQ6S7m

Author: Ben05-sys

app/screen.py

@@ -122,13 +122,12 @@
     # today's tape — same drift tolerance as `earnings_yield`, whose
     # denominator this shares.
     "payout_ratio",
-    # Average dollar volume is a liquidity floor, not a reading of the
-    # tape: `avgdvol > 20m` means "this name normally trades enough to get
-    # out of", and half a day of price drift does not change that answer.
-    # Today's `dollar_volume` is deliberately absent, for the same reason
-    # `volume` is — it only rises, so a stale value drops the very names
-    # that have since crossed the threshold.
-    "avg_dollar_volume",
+    # avg_dollar_volume is NOT here, though it looks like a standing fact:
+    # it multiplies avg_volume_3m (slow) by `price` (live), so prefiltering
+    # on a stale snapshot answers with yesterday's price baked into today's
+    # threshold — the same two-clocks-in-one-number mistake `pct_52w_range`
+    # avoids by staying off this set. It's in LIVE_COLUMNS instead, same as
+    # `dollar_volume`, and gets re-priced before a screen sees it.
     # Two rolling averages, same drift tolerance as `sma_spread`. Today's
     # raw `volume` is deliberately absent, same reasoning as the line above.
     "volume_trend",

tests/test_screen.py

@@ -760,17 +760,21 @@ def when_for(hour, minute=0):
           and "volume_trend" not in screen.LIVE_COLUMNS)
 
     # Today's dollar volume only rises, so prefiltering on a stale one would
-    # drop the names that have since crossed the line. The average is a
-    # standing fact about the name and narrows safely.
+    # drop the names that have since crossed the line. The average looks
+    # like a standing fact but multiplies a static volume by a live price,
+    # so it waits for a live quote too, same as the raw figure.
     static, live = screen.split_live("avgdvol > 20m and dvol > 100m")
-    check("average dollar volume narrows before re-pricing",
-          static == "avgdvol > 20000000.0", static)
-    check("today's dollar volume waits for live quotes",
-          live == "dvol > 100000000.0", live)
+    check("neither dollar-volume line narrows before re-pricing",
+          static == "", static)
+    check("both wait for live quotes",
+          live == "avgdvol > 20000000.0 and dvol > 100000000.0", live)
     check("day range position is live",
           "day_range_pct" in screen.LIVE_COLUMNS)
     check("today's dollar volume is never prefiltered",
           "dollar_volume" not in screen.STATIC_SAFE)
+    check("nor is its average, which bakes in today's price",
+          "avg_dollar_volume" in screen.LIVE_COLUMNS
+          and "avg_dollar_volume" not in screen.STATIC_SAFE)
 
     dvol_filter = screen.filters_to_expr({"dvol": "Over $10M"})
     check("the dropdown compiles", dvol_filter == "avgdvol > 10m", dvol_filter)