Stop one subject taking every notification slot

Connor noticed 5 of 5 alerts were Middle East. Measuring showed that was not
proportionate: Middle East blocks are ~11% of the eligible corpus while
software-systems alone is 25%, yet Middle East took 35% of the top 20 and
software took none.

Two mechanisms, both measured:

1. Five conflict-* blocks could each claim their own alert slot. Interest blocks
   now carry a `group`, and routing enforces max_alerts_per_group_per_day, so
   related blocks share one quota. This alone moved the mix from 5/5 Middle East
   to three distinct groups.

2. Z-scoring subtracts each block's MEAN, which silently encodes how much of the
   corpus that block matches. software-systems has the HIGHEST mean raw
   similarity of any block (0.537) and the LOWEST max z (2.51);
   conflict-mideast-israel has a lower mean (0.460) and reaches 3.47. A topic is
   penalised for being well covered in your own feed list.

Added `normalize = "top-anchored"` as an alternative that rescales each block by
its own upper tail instead of its mean, leaving `weight` to express preference
rather than corpus composition.

Also ruled out a plausible-sounding cause rather than assuming it: HN and
Lobsters items are title-only and much shorter than news articles, so terse
titles seemed likely to match prose probes poorly. Measured the opposite --
title-only HN items score 0.618 against the software block, ABOVE Phoronix/LWN
with summaries (0.610) and Ars Technica (0.540). Short text is not the problem;
prevalence is.

news/cli.py

@@ -113,7 +113,7 @@ def cmd_simulate(args) -> int:
         routing["important_threshold"] = args.important_threshold
 
     now = datetime.now(UTC)
-    decisions = route(con, scores, routing, cfg.delivery, now)
+    decisions = route(con, scores, routing, cfg.delivery, now, cfg.block_groups())
     by_lane = {}
     for d in decisions:
         by_lane.setdefault(d.lane, []).append(d)

news/config.py

@@ -106,6 +106,7 @@ class Config:
         r = self.section("routing")
         return {
             "interest_per_day": r.get("interest_per_day", 3),
+            "max_alerts_per_group_per_day": r.get("max_alerts_per_group_per_day", 1),
             "important_per_day": r.get("important_per_day", 2),
             "important_per_night": r.get("important_per_night", 1),
             "interest_threshold": r.get("interest_threshold", 2.0),
@@ -136,6 +137,15 @@ class Config:
     def profile(self) -> dict[str, Any]:
         return _load(self.profile_path)
 
+    def block_groups(self) -> dict[str, str]:
+        """block id -> quota group, defaulting to the block's own id."""
+        profile = self.profile()
+        out: dict[str, str] = {}
+        for kind in ("interest", "importance"):
+            for b in profile.get(kind, []):
+                out[b["id"]] = b.get("group", b["id"])
+        return out
+
     @property
     def delivery(self) -> dict[str, Any]:
         return self.profile().get("delivery", {})

news/route.py

@@ -64,6 +64,7 @@ def route(
     routing: dict,
     delivery: dict,
     now: datetime,
+    groups: dict[str, str] | None = None,
 ) -> list[Decision]:
     """Decide a lane for every scored cluster.
 
@@ -71,6 +72,7 @@ def route(
     budget is spent on the best candidates rather than on whatever arrived
     first -- the ordering bug that let the old system's budget go to noise.
     """
+    groups = groups or {}
     quiet = delivery.get("quiet_hours", {})
     quiet_now = in_quiet_hours(now, quiet)
     may_wake = bool(delivery.get("importance_may_wake", True))
@@ -89,6 +91,24 @@ def route(
         )
     }
 
+    # One subject must not take every slot. Measured before this existed: five of
+    # five alerts were Middle East, though Middle East blocks are only ~11% of
+    # the corpus and software-systems alone is 25%.
+    per_group = routing.get("max_alerts_per_group_per_day", 1)
+    group_used: dict[str, int] = {}
+
+    def group_of(s: ClusterScore, kind: str) -> str:
+        block = s.interest.block if kind == "interest" else s.importance.block
+        return groups.get(block, block)
+
+    def group_full(s: ClusterScore, kind: str) -> bool:
+        g = group_of(s, kind)
+        return group_used.get(g, 0) >= per_group
+
+    def take_group(s: ClusterScore, kind: str) -> None:
+        g = group_of(s, kind)
+        group_used[g] = group_used.get(g, 0) + 1
+
     decisions: dict[str, Decision] = {}
     # Deferred items are digest CANDIDATES, not digest members. Everything that
     # misses an alert lane competes for the same capped digest, so a quiet-hours
@@ -132,12 +152,16 @@ def route(
         if important_budget <= 0:
             defer(s, "important-budget-exhausted")
             continue
+        if group_full(s, "importance"):
+            defer(s, f"group-quota:{group_of(s, 'importance')}")
+            continue
         if quiet_now:
             if not may_wake or night_budget <= 0:
                 defer(s, "quiet-hours")
                 continue
             night_budget -= 1
         important_budget -= 1
+        take_group(s, "importance")
         settle(s, IMPORTANT, f"importance:{s.importance.block}", s.importance_score)
 
     # --- interest lane: never overrides quiet hours ---
@@ -155,7 +179,11 @@ def route(
         if interest_budget <= 0:
             defer(s, "interest-budget-exhausted")
             continue
+        if group_full(s, "interest"):
+            defer(s, f"group-quota:{group_of(s, 'interest')}")
+            continue
         interest_budget -= 1
+        take_group(s, "interest")
         settle(s, INTEREST, f"interest:{s.interest.block}", s.interest_score)
 
     # --- one capped, ranked digest ---

news/score.py

@@ -110,6 +110,29 @@ def _z(values: np.ndarray) -> np.ndarray:
     return (values - float(values.mean())) / std
 
 
+def _top_anchored(values: np.ndarray) -> np.ndarray:
+    """Rescale so each block's own 99th percentile lands at 1.0.
+
+    Z-scoring subtracts the block's MEAN, which silently encodes how much of the
+    corpus that block matches. Measured: `software-systems` matches ~25% of this
+    reader's corpus (Hacker News, Lobsters, Phoronix, LWN, GitHub releases,
+    reddit) and has the highest mean raw similarity of any block, 0.537 -- yet
+    the lowest maximum z, 2.51. `conflict-mideast-israel` matches 5% of the
+    corpus, has a lower mean of 0.460, and reaches z 3.47.
+
+    The effect is perverse: a topic is penalised for being well covered in your
+    own feed list. Anchoring on the upper tail instead makes "the best item this
+    block has seen" comparable across blocks, leaving `weight` to express actual
+    preference rather than corpus composition.
+    """
+    med = float(np.percentile(values, 50))
+    hi = float(np.percentile(values, 99))
+    return (values - med) / max(hi - med, 1e-9)
+
+
+NORMALIZERS = {"zscore-per-block": _z, "top-anchored": _top_anchored}
+
+
 def score_clusters(
     con: sqlite3.Connection,
     profile_vectors: ProfileVectors,
@@ -142,11 +165,16 @@ def score_clusters(
     veto_raw = profile_vectors.raw_matrix("veto", mat)
     demote_raw = profile_vectors.raw_matrix("demote", mat)
 
-    interest_z = {b: _z(v) * profile_vectors.weight("interest", b) for b, v in interest_raw.items()}
-    importance_z = {
-        b: _z(v) * profile_vectors.weight("importance", b) for b, v in importance_raw.items()
+    norm = NORMALIZERS.get(scoring_cfg.get("normalize", "zscore-per-block"), _z)
+    interest_z = {
+        b: norm(v) * profile_vectors.weight("interest", b) for b, v in interest_raw.items()
     }
-    veto_z = {b: _z(v) for b, v in veto_raw.items()}
+    importance_z = {
+        b: norm(v) * profile_vectors.weight("importance", b) for b, v in importance_raw.items()
+    }
+    # Veto stays on z-scores: it is a comparison against the interest signal,
+    # and both sides must be on the same scale.
+    veto_z = {b: norm(v) for b, v in veto_raw.items()}
 
     def argbest(per_block: dict[str, np.ndarray], i: int) -> tuple[str, float]:
         return max(((b, float(v[i])) for b, v in per_block.items()), key=lambda kv: kv[1])

profile/config.toml

@@ -72,6 +72,9 @@ corroboration_depth_weight = 0.2
 # Daily budgets. Deliberately small: the point of the system is a handful of
 # well-chosen interruptions, not a feed.
 interest_per_day = 3
+# Max alerts per quota group per day (see `group` in profile.toml). Stops one
+# subject taking every slot.
+max_alerts_per_group_per_day = 1
 important_per_day = 2
 # Hard cap on waking the reader during quiet hours, inside important_per_day.
 important_per_night = 1

profile/profile.toml

@@ -10,6 +10,10 @@ schema_version = 2
 
 # =========================================================== interests =======
 
+# `group` shares one notification quota across related blocks. Without it the
+# five conflict-* blocks could each claim an alert slot and fill a day with the
+# same subject -- which is what happened: 5 of 5 alerts were Middle East.
+#
 # --- foreign affairs, split from armed conflict ------------------------------
 # Connor: "my actual war interest is low, but unfortunately all wars are
 # important." The click history looked war-heavy only because world-news feeds
@@ -30,7 +34,8 @@ probes = [
 # The A-E ordering Connor gave, encoded as weights. Even the lowest band still
 # reaches him -- but through importance, on major developments only.
 [[interest]]
-id = "conflict-anti-imperial"          # (A) top of his ordering
+id = "conflict-anti-imperial"
+group = "conflict"          # (A) top of his ordering
 weight = 0.9
 probes = [
   "An armed liberation movement fights a colonial or occupying power for self-determination.",
@@ -39,7 +44,8 @@ probes = [
 ]
 
 [[interest]]
-id = "conflict-us-involved"            # (B)
+id = "conflict-us-involved"
+group = "conflict"            # (B)
 weight = 0.85
 probes = [
   "US forces carry out strikes and the administration defends the legal basis for the war.",
@@ -47,7 +53,8 @@ probes = [
 ]
 
 [[interest]]
-id = "conflict-mideast-israel"         # (C)
+id = "conflict-mideast-israel"
+group = "conflict"         # (C)
 weight = 0.8
 probes = [
   "Israeli forces strike targets in Gaza, Lebanon or Syria amid continuing escalation.",
@@ -56,7 +63,8 @@ probes = [
 ]
 
 [[interest]]
-id = "conflict-undercovered"           # (D)
+id = "conflict-undercovered"
+group = "conflict"           # (D)
 weight = 0.75
 probes = [
   "Fighting intensifies in Sudan as famine spreads and aid convoys are blocked.",
@@ -65,7 +73,8 @@ probes = [
 ]
 
 [[interest]]
-id = "conflict-general"                # (E) -- low interest, still important
+id = "conflict-general"
+group = "conflict"                # (E) -- low interest, still important
 weight = 0.45
 probes = [
   "Front-line fighting continues as both sides exchange drone and artillery strikes.",
@@ -137,6 +146,7 @@ probes = [
 
 [[interest]]
 id = "labor-inequality"
+group = "left-politics"
 weight = 0.85
 probes = [
   "Workers vote to unionize after a long organizing campaign; the employer contests the result.",
@@ -148,6 +158,7 @@ probes = [
 
 [[interest]]
 id = "left-movements"
+group = "left-politics"
 weight = 0.85
 probes = [
   "A socialist party wins a decisive election, shifting the governing coalition leftward.",
@@ -157,6 +168,7 @@ probes = [
 
 [[interest]]
 id = "anticolonial-sovereignty"
+group = "left-politics"
 weight = 0.8
 probes = [
   "Indigenous communities win recognition of land rights against extractive industry.",
@@ -176,6 +188,7 @@ probes = [
 
 [[interest]]
 id = "political-theory"
+group = "left-politics"
 weight = 0.7
 probes = [
   "An essay argues about socialism, imperialism and political economy, drawing on history rather than reporting a news event.",
@@ -204,6 +217,7 @@ probes = [
 
 [[interest]]
 id = "cincinnati"
+group = "local"
 weight = 0.9
 local = true
 probes = [
@@ -216,6 +230,7 @@ probes = [
 
 [[interest]]
 id = "kentucky-louisville"
+group = "local"
 weight = 0.6
 local = true
 probes = [
@@ -226,6 +241,7 @@ probes = [
 # coverage alongside Cincinnati city government.
 [[interest]]
 id = "ohio-state-politics"
+group = "local"
 weight = 0.75
 local = true
 probes = [