From eff960c54620bfba47d40bc3792ba805dea3dab3 Mon Sep 17 00:00:00 2001
From: Sam Scholten
Date: Sun, 30 Aug 2026 20:34:38 +1000
Subject: 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.
---
 cmds/rss.go | 57 ++++++++++++++++++++++++++++++++++++++++++++-------------
 1 file changed, 44 insertions(+), 13 deletions(-)

(limited to 'cmds/rss.go')

diff --git a/cmds/rss.go b/cmds/rss.go
index 3c49513..8679f6f 100644
--- a/cmds/rss.go
+++ b/cmds/rss.go
@@ -8,10 +8,16 @@ import (
 	"encoding/xml"
 	"fmt"
 	"time"
+	"unicode/utf8"
 
 	"fluxrec/core"
 )
 
+// Cap on the source-feed name in item titles: feed titles can be
+// arbitrarily long, and the suffix must stay compact so list views keep
+// showing the article title itself.
+const maxFeedTitleRunes = 40
+
 const rssXMLHeader = `<?xml version="1.0" encoding="UTF-8"?>` + "\n"
 
 type rssGUID struct {
@@ -40,9 +46,11 @@ type rssDocument struct {
 }
 
 // RenderRSS renders runs (oldest to newest, as read from runs.jsonl) as an
-// RSS 2.0 document. The source feed title prefixes each item description so
-// provenance is visible in-reader; item content was already cleaned when
-// the run row was written.
+// RSS 2.0 document. Each item title gets the source feed as a rune-capped
+// suffix (readers show the channel title as the source of every item, so
+// this is the only in-list provenance), and the description body opens
+// with feed + model score + the batch-wide median score. Item content was
+// already cleaned when the run row was written.
 func RenderRSS(runs []core.RunRow) ([]byte, error) {
 	doc := rssDocument{
 		Version: "2.0",
@@ -54,11 +62,11 @@ func RenderRSS(runs []core.RunRow) ([]byte, error) {
 	for i := len(runs) - 1; i >= 0; i-- {
 		for _, k := range runs[i].Kept {
 			doc.Channel.Items = append(doc.Channel.Items, rssItem{
-				Title:       k.Title,
+				Title:       rssItemTitle(k),
 				Link:        k.URL,
 				GUID:        rssGUID{IsPermaLink: false, Value: fmt.Sprintf("mf:%d", k.EntryID)},
 				PubDate:     k.PublishedAt.Format(time.RFC1123Z),
-				Description: rssDescription(k),
+				Description: rssDescription(k, runs[i].ScoreMedian),
 			})
 		}
 	}
@@ -69,14 +77,37 @@ func RenderRSS(runs []core.RunRow) ([]byte, error) {
 	return append([]byte(rssXMLHeader), body...), nil
 }
 
-// rssDescription builds the item body: source feed name on the first line,
-// then the cleaned content (if any).
-func rssDescription(k core.KeptItem) string {
-	if k.Content == "" {
-		return k.FeedTitle
+// rssItemTitle builds the displayed title: article title first, then the
+// (rune-capped) source feed as a suffix so narrow list views truncate the
+// provenance, not the title itself.
+func rssItemTitle(k core.KeptItem) string {
+	feed := truncateRunes(k.FeedTitle, maxFeedTitleRunes)
+	if feed == "" {
+		return k.Title
+	}
+	return k.Title + " — " + feed
+}
+
+func truncateRunes(s string, max int) string {
+	if utf8.RuneCountInString(s) <= max {
+		return s
 	}
-	if k.FeedTitle == "" {
-		return k.Content
+	return string([]rune(s)[:max]) + "…"
+}
+
+// rssDescription builds the item body: first a headline line with the
+// source feed, model score, and batch-wide median (omitted for run rows
+// written before the median was logged), then the cleaned content (if any).
+func rssDescription(k core.KeptItem, median *float64) string {
+	head := fmt.Sprintf("score %.2f", k.Score)
+	if median != nil {
+		head += fmt.Sprintf(" · batch median %.2f", *median)
+	}
+	if k.FeedTitle != "" {
+		head = k.FeedTitle + " · " + head
+	}
+	if k.Content == "" {
+		return head
 	}
-	return k.FeedTitle + "\n\n" + k.Content
+	return head + "\n\n" + k.Content
 }
-- 
cgit v1.2.3