Yen Idhazh

A daily digest that checks its own work.

Console

The latest run on this page finished at 18:40 UTC on Monday 19 October. A run still going is not on this page yet.

Latest day

2026-10-19 ran 3 runs and published 195 of 240 planned items. 45 items failed.

3 runs are worth a look

Worst thing right now

26 feeds are resting, so nothing they carry reaches the digest. Each is asked again after 5 runs. On Pipelines.

Site size

45.7 MB of the 1 GB limit - room for about 386,000 more articles.

The days control above needs JavaScript, and so do the sections below - they draw rows a browser fetches. Their shape is 14 days.

10 summaries were marked "not sure" on this day. Summaries →

Stories the day merged

One column is one day: the stories that day folded behind another because we read them as the same story. The dot is the biggest group that day, counting the one we kept.

Stories merged against stories kept apart.

42 stories merged

024686 Oct 20269 Oct11 Oct14 Oct16 Oct19 OctStories
19 Oct, the newest published day
 Merged
 
1 of 195 published
Groups
 
1 group
 Biggest group
 
2 stories

Point at a column to read it. Left and Right step through them, Escape returns to the newest.

42 stories were folded into another over these 14 days. The biggest group held 4 stories, on 17 Oct. That is 1% of the 4,677 stories these 14 days published.

Where the merge line sits

The solid line is the nightly calculation's final score for grouping two stories as one, after limits on its change. The dotted line is the proposed score before those limits. The shaded band shows how far the calculated line was allowed to fall each day. A build may have used a different line. The tinted strip shows the part of the score range inside this plot for 4 pairs a person marked as two stories. Their scores run from 0.9343 to 0.9407. If used to group stories, a calculated line inside this strip would clear the score threshold for at least one of those pairs. This does not show that a build grouped them.

Each calculated line against its proposal and the line the newest day used.

The judge record is packed through 17 Oct 2026. Later days have no packed reading here.

The last packed period with rows in the judge record is 17 Oct 2026 to 17 Oct 2026. This is a period, not the date of the last run.

0.940 was the newest calculated line

the pairs marked two stories0.8800.9000.9200.9400.9600.9801.0006 Oct8 Oct10 Oct13 Oct15 Oct17 Oct
17 Oct, the newest recorded day shown
 Calculated line
 
0.940
 Proposed line
 
nothing was fitted
Change
 
stayed where it was

Point at a column to read it. Left and Right step through them, Escape returns to the newest.

The clamp has not held the line back in these 14 days. Nothing was fitted on 12 recorded days in this 14-day window.

Whether the judge agrees with itself

Every pair is read twice, with the two summaries swapped. A "disagreed" dot shows how often a pair's two readings disagreed. A "could not tell" dot shows how often the pairs whose two readings agreed could not tell. When a day's rate is past its own dashed mark, the run does not move the merge line that day.

The judge's two readings against each other, and each rate against its own limit.

The judge record is packed through 17 Oct 2026. Later days have no packed reading here.

The last packed period with rows in the judge record is 17 Oct 2026 to 17 Oct 2026. This is a period, not the date of the last run.

12% of 443 pairs disagreed with the judge's own second reading

0%10%20%30%40%15% - the "disagreed" mark35% - the "could not tell" mark6 Oct 20269 Oct11 Oct14 Oct16 Oct19 Oct
17 Oct, the newest day with numbers
 Disagreed with the second reading
 
just below 15% of 101 pairs
 Could not tell
 
0% of the 86 that agreed

Point at a column to read it. Left and Right step through them, Escape returns to the newest.

In these 14 days, 12% of 443 pairs disagreed with their own second reading, and 0% of the 392 that agreed could not tell. Both rates are inside the marks.

What the record still needs

Three counts have to be reached before the line may move at all. The squares are one a day: what the record did with that day.

Each gate's recorded count against the count it needs.

The judge record is packed through 17 Oct 2026. Later days have no packed reading here.

The last packed period with rows in the judge record is 17 Oct 2026 to 17 Oct 2026. This is a period, not the date of the last run.

2 of 3 gates passed

Readings the record holds 64

200 needed before a line may be fitted

Days the record has counted 12

10 needed, so one fortnight of one kind of news cannot set the line

Pairs judged at or above the line 150

30 needed - these are the whole precision reading

19 Oct 2026, the newest day
The record
 
no judge row was returned for this day.
 Square
 
silent

Point at a square to read its day. Left and Right step through the days, Escape returns to the newest.

The record has 64 readings, 12 days, and 150 pairs above the line; it needs at least 200 readings, 10 days, and 30 pairs above the line. No fitted line was returned for these 14 days.

What the judge said about the line

Every judged pair, split by whether its score cleared the line the newest day was built with and by what the judge said about it.

Pairs judged one story against pairs judged two stories, split at the merge line.

The judge record is packed through 17 Oct 2026. Later days have no packed reading here.

The last packed period with rows in the judge record is 17 Oct 2026 to 17 Oct 2026. This is a period, not the date of the last run.

206 pairs judged one story

146 Eligible to merge, and the judge agrees
4 Eligible to merge, and the judge disagrees
798 Not eligible, and the judge disagrees
60 Not eligible, and the judge agrees

The line would let a pair in the second cell through, and where the group forms it is a story the reader never sees. A pair in the cell below it is one the reader sees twice.

2.7% of the pairs above the line were called two stories. The fit aims for 3.0%.

called one story: 0.880 to 0.999, middle 0.908called two stories: 0.880 to 0.999, middle 0.8940.8800.9000.9200.9400.9600.9801.000
In the band
101

How many pairs were close enough to be worth judging.

Judged
101

How many the day's budget actually read. A smaller number means the cap cut the draw.

Usable
86

How many got two readings that agreed. Only these went into the record.

The record, slice by slice
ScoreOne storyTwo storiesCould not tell
0.880 to 0.88584160
0.885 to 0.8909190
0.890 to 0.8958980
0.895 to 0.9009550
0.900 to 0.9055720
0.905 to 0.9108250
0.910 to 0.9155780
0.915 to 0.9207260
0.920 to 0.9255310
0.925 to 0.9304400
0.930 to 0.9354700
0.935 to 0.9402700
0.940 to 0.9453900
0.945 to 0.9502200
0.950 to 0.9551500
0.955 to 0.960600
0.960 to 0.965700
0.965 to 0.970100
0.970 to 0.975200
0.995 to 1.0005440

The pairs a person marked apart

Every pair somebody read as two different stories sets a floor the merge line has to stay above. The dots are those pairs, the rule is the line, and the tinted zone is as far as the line may fall in one day.

Each hand-marked pair's score against the merge line.

The hand-mark record starts on 19 Sep 2026.

The hand-mark record is packed through 17 Oct 2026. Later days have no packed reading here.

The last packed period with rows in the hand-mark record is 19 Sep 2026 to 19 Sep 2026. This is a period, not the date of the last run.

The holdout-score record starts on 21 Sep 2026.

The holdout-score record is packed through 17 Oct 2026. Later days have no packed reading here.

The last packed period with rows in the holdout-score record is 21 Sep 2026 to 21 Sep 2026. This is a period, not the date of the last run.

0.0007 below the closest pair a person marked as two stories

One day's legal fall is up to 0.010, so 3 more marks are inside tomorrow's reach and 4 of 4 would be on the wrong side the day after.

one day's fall0.0007the line, 0.940read as two stories: 0.9343 to 0.94070.94070.9343read as one story: 0.7316 to 0.9914, middle 0.9377, 117 below the line0.9200.9300.9400.950
Google renames Lake Ontario to Lake America for US users following Trump order against Trump asks Apple to rename Lake Ontario to Lake America, the first mark
Scoring
 
0.9343
Against the line
 
below it, so the pair stays two stories
Marked
 
19 Sep 2026

Point at a dot, an edge chip or the strip below them to read it. Left and Right step through them, Escape returns to the first.

The line is 0.0007 BELOW a pair a person marked as two stories, so it would merge them. 1 hand-marked pair is on the wrong side of it.

196 pairs were read as one story, running 0.7316 to 0.9914. 117 of them score below the line, so the reader would see those stories twice.

Scored at 1 on the cosine, the weight on the newest fitted day, 2026-10-17.

Every pair a person marked as two stories
ScoreOne headlineThe otherMarked
0.9407Google Maps renames Lake Ontario to Lake America for US usersApple Maps renames Lake Ontario to Lake America for US users2026-09-19
0.9374Apple changes Lake Ontario name to Lake America after Trump pressureTrump asks Apple to rename Lake Ontario to Lake America2026-09-19
0.9352Trump ordered Apple to rename Lake Ontario to Lake AmericaApple renames Lake Ontario to Lake America following Trump executive order2026-09-19
0.9343Google renames Lake Ontario to Lake America for US users following Trump orderTrump asks Apple to rename Lake Ontario to Lake America2026-09-19

196 more marked pairs were read as one story. They set no floor, so they are drawn as the strip above and not listed.

Scored on 2026-09-21 at a line of 0.9400, against marks by claude-opus-4.6. It joined 1 of the 4 pairs marked as two stories, and left 117 pairs marked as one story apart.