rss: provenance + score calibration in items

Item titles gain a rune-capped source-feed suffix so list views (which
only ever show the channel title) display provenance inline. Item bodies
open with feed, model score, and the batch-wide score median; the median
is computed over every scored entry at poll time and stored on the run
row as a nil-safe pointer, so old rows render without it. /api/status
exposes the median too.

1 files changed, 6 insertions, 1 deletions

diff --git a/core/runs.go b/core/runs.go
index 38e6710..0dd4a7f 100644
--- a/core/runs.go
+++ b/core/runs.go
@@ -47,7 +47,11 @@ type KeptItem struct {
 
 // RunRow is one poll batch: what was seen, what was kept, when, and how
 // old the model was. EntriesSeen counts every fetched entry including any
-// skipped before scoring (see RecFeedSkipped).
+// skipped before scoring (see RecFeedSkipped). ScoreMedian is the median
+// score across every scored entry this poll — the batch's background
+// calibration, rendered next to kept items' scores in the RSS body. It's
+// a pointer so old rows (and empty polls) read as "absent", never as
+// a misleading 0.00.
 type RunRow struct {
 	RanAt          time.Time  `json:"ran_at"`
 	EntriesSeen    int        `json:"entries_seen"`
@@ -55,6 +59,7 @@ type RunRow struct {
 	Kept           []KeptItem `json:"kept"`
 	DurationMs     int64      `json:"duration_ms"`
 	ModelCreatedAt time.Time  `json:"model_created_at"`
+	ScoreMedian    *float64   `json:"score_median,omitempty"`
 }
 
 // ReadRuns parses all of path strictly. A missing file is not an error —