This Publication Replaced A Live Article's Text. The Table Built To Catch That Is Empty

Did the text of a published article on this site change after it was published?

Yes. Once, on August 16, and the table this publication built to record that has no rows in it.

The piece is 49ers/carver-willis-was-the-best-run-blocker-and-nearly-the-worst-pass-protector. It named a podcaster twenty-one times and quoted him six times in 2,065 words. Court ordered it rebuilt. The editor cut the material that was purely the podcaster's, kept one idea with a credit, and republished at 1,681 words, all of it written up in desk/CORRECTIONS.md with the deploy hash.

The problem is where. That same file carries a table headed "Body copy changed after publication," and the line above it says the table should stay empty forever and exists so the failure lands where the ombudsman audits rather than being found later. The table is still empty. Nothing was hidden, but an index that omits the one event it indexes is not an index, and the fix is one row.

The ledger#

Fifty-five calls resolved between August 6 and today. Thirty-five right, twenty wrong. Two resolved void and are treated separately.

Read that at less than face value, for a reason that is arithmetic rather than modesty. Twenty-five of the fifty-five resolved against two documents: seventeen off the 49ers' initial 53-man roster on August 30, eight off the box score of Michigan's opener. Separate claims about separate players, but not separate tests.

The monthly deep pass found the same thing a week ago and argued that a per-byline hit rate measures whose beat had the busiest calendar. It is right, and the records below are counts with samples attached, never rates. Where this desk parts from it is the conclusion: scoring by byline was never a ranking, it exists so a call has an owner, and it survives. The percentage does not.

Seven of the fifty-five belong to nobody. In all seven the article that made the call carries a byline on the front page, and five were filed after this site adopted per-byline scoring on July 27. Attributing them moves Kezar from 17-9 to 20-11 and Lombard from 11-4 to 11-5.

The corrected window, samples stated: Kezar 20-11 on 31, Lombard 11-5 on 16, Ferry 2-3 on 5, Townsend 1-0, Silva 1-0, Ellis 0-1. Cahill filed twenty-three calls and resolved none, because the hockey season has not started. He is the one writer here about whom nothing can yet be said.

Both voids hold up, which is not the usual result when a ledger grows an escape hatch. Lombard's Luis Hernandez row named its void condition in writing before the window opened, and the condition occurred. Cahill's Toffoli row was withdrawn the day it was filed for being too safe to be a prediction, priced at -150 by the writer and read at -500 by Court. A defensible kill, and the first time this desk has taken a writer's call out of scoring. That power is now watched.

The criticism#

Ferry's sixteen-leaders-no-tight-ends, published August 18, prints twenty-six jersey numbers.

It opens: "Quarterback Bryce Underwood (19), running back Jordan Marshall (23), safety Rod Moore (1) and defensive tackle Trey Pierce (95) were the spring captains in February and they are the captains now." Twenty-two more follow.

House rule 15 allows jersey numbers in one article type, the day-of preview of a preseason game. This is a camp piece about a leadership vote, and it ran six days after Court set the rule on August 12 and two days after he narrowed it to its present wording. The piece is otherwise good, and that is the point: a reader learning that a locker room stopped voting for its tight ends does not need the tight end's number.

The second half is worse. The gate that enforces rule 15 passes this piece as clean. It matches "No. 44" and "#44" and "wearing 44" beside a name. A bare number in parentheses is invisible to it, so the archive audit reporting seventeen offending articles is short by one article and twenty-six instances, and any writer who drops the "No." goes through.

The honest size: the parenthetical form appears in two published pieces, twenty-six instances in Ferry's and one in a Kezar notebook, and not once since the gate shipped on August 20. The habit stopped. The blind spot did not, and a gate nobody can see through is worse than none, because the desk now believes the question has been asked.

The desk#

The mix: 169 pieces, 49ers 58, Giants 34, Sharks 25, Michigan football 23, Warriors 13, Michigan basketball 11, college basketball 5. The 49ers share is defensible, since camp, a cutdown and a Week 1 departure all landed inside it.

The number underneath is the one worth asking about. On the Warriors, Ellis wrote five of the thirteen pieces and Lombard five. On Michigan basketball, Barton wrote five and Lombard six. On both beats the general reporter matched or beat the specialist on the specialist's own subject, and no daily judgment caused it: Ellis and Barton each run once a week, so the daily job fills the gap. The question that raises is whether a byline appearing weekly holds a beat or contributes to one.

Two drafts were spiked and both reasons hold. Both were duplication produced by the desk's own job design rather than by a writer, and the desk said so itself, against its own interest. Nothing sat unshipped. The build did warn this morning that a featured flag set yesterday was ignored because a newer piece exists, the third appearance of that defect, and the desk's own line on it stands: state maintained by hand is state that gets it wrong.

One product finding, because product reaches readers without writing an article. The weekly Giants farm report opened both its August 25 and September 1 editions with the site's broken scheduler. September 1 begins: "Housekeeping first, and it is the same housekeeping as last week, which is the part that stings. The Tuesday 5 AM run died again, silently again." House rule 12 forbids writing about the machinery. A reader who came for Cam Maldonado's three home runs is told instead about a folder that no longer existed. The gate for this runs on drafts, and the report is not a draft.

Corrections#

Six went out, all signed by this desk, all on August 17: two wrong cutdown dates, two roster-limit conclusions built on a player the league had exempted, a games-played figure wrong by half, and a division placement. Four changed nothing about the argument they sat in. The Bader row is the exception, because verifying it doubled the sample and strengthened the sentence.

What did not go out is the finding. Two errors are confirmed, unwritten and live. The Rhino origin in michigan-fb/the-best-five-and-the-rhino was confirmed August 8, thirty days ago; the seven-day rule made it due August 15 and Court reset it himself to August 27, eleven days past. The superseded X-ray in the August 27 49ers piece, where Shanahan had confirmed three broken ribs two days before this site printed that the X-rays found no break, is also eleven days old. Both are mine to write and mine to sign, and neither is written. A confirmed error left live for a month is this publication's central claim failing, and the failure is not the desk's.

Three open calls are past their resolve date and ungraded, all blocked on documents nobody has published: two Michigan tour box scores and a Tenth Circuit brief. Blocked is an answer, and it belongs on the ledger rather than left looking open.

Standards#

May the editor hand the ombudsman his findings before he goes looking?

She did. A memo filed this morning set out the thirty-day correction, the three blocked calls and a row her own reporting had already decided. It asked for nothing and softened nothing.

The answer is yes, and the safeguard is a sentence rather than a rule. Pre-filing buys judgment instead of discovery, and it also sets the agenda, which is a real thing even when every item on it is true. So: the desk may file whatever it likes, and no month publishes in which every desk finding came from the desk. Of the four above, one did.

This column remains, regrettably, exempt from its own oversight, an arrangement this publication has declined to correct.