Corrections

Re-pull evolving disaster figures from that day's own cited source, every time. Three of this week's five corrections were Venezuela-earthquake casualty/aid figures (injured, missing, aid dollars) that were stale, transposed, or simply unsupported by the sources cited alongside them — in a fast-moving, daily-updating disaster, don't carry a number forward from memory or a prior day; verify it against the specific source linked in that day's item before publishing.

When an article covers multiple similar entities or cases, pin every figure and name to the specific one it belongs to. Three of this week's corrections (Iraq's cash-courier total, Bangladesh's gold-smuggling container, India's seizure total) misattributed a detail from an adjacent case, company name, or sub-total in the same source to the story's actual subject. Before publishing a figure, re-read the sentence it came from in the source, not just the surrounding paragraph.

State a trend's direction only as the cited source states it — verify "up" vs "down," "easing" vs "worsening" explicitly. This week's Panama vessel-detention item inverted its source's trend (called an accelerating detention campaign "easing"), echoing the same failure mode as the Taiwan PLA-sortie correction two weeks prior. Don't infer direction from narrative expectation; read the source's own characterization of the trend.

Don't widen a source's own qualifier into a bigger claim. Two corrections this week (Iran "acknowledging" a Bab el-Mandeb threshold that no source reports it acknowledging; DRC's Ebola outbreak called "the fastest-growing ever recorded" when the source said third-largest overall / fastest within its own or a regional trajectory) turned a narrower, hedged source claim into an unqualified, broader one. If the source scopes a superlative or an attribution, keep the brief's claim scoped the same way.

Every specific figure, actor, or detail in the headline and so_what must trace to one of the URLs actually cited for that item — not to background familiarity with the story. Six of this week's corrections (a Jazan refinery strike, DRC Ebola's "third-largest"/Europe-export cases/$518M response plan, a UN "repeat of El-Fasher" warning, Ecuador's 57% Guayaquil-Antwerp contamination rate, an IRGC tanker-strike attribution, a Serbian-flag/Balkan-network claim) were independently plausible or true but weren't supported by either source linked on that item, so a reader clicking through couldn't verify them. Before publishing a detail, confirm it appears in this item's own cited source(s), not just somewhere in the wider news cycle.

The "re-pull the evolving figure" lesson applies to fast-moving violence/attack tolls, not just disaster figures. Two corrections this week (a Cúcuta truck-bombing wounded count reported as 14 when both cited sources said 11; a Darfur drone-strike toll reported as "up to 37" when both cited sources said "at least 35," the 37 tracing only to an uncited RSF self-report) repeated the 2026-07-06 disaster-toll lesson's mechanism in a new context — an actively developing attack, not a natural disaster. Casualty/wounded/killed figures in any breaking violence story need the same discipline: pull the exact number from the sources actually cited on that item, never a nearby or higher figure from elsewhere in the news cycle.

The "don't widen a source's own qualifier" failure keeps recurring — this is now the third Reckoning to log an instance of it (2026-07-20's Iran-Bab-el-Mandeb and DRC-Ebola corrections, 2026-07-27's cluster of five in one week, and this week's UAE-Hormuz correction, which turned the cited source's explicit "not immediately clear whether the incident involved the same [already-counted] vessel" into a confirmed "fourth vessel" / "first non-ADNOC" claim). When a source expresses uncertainty about whether an event is new or a duplicate of one already reported, keep that uncertainty in the brief — don't round it up to a distinct, confirmed occurrence.

When a group of similarly-charged defendants or entities differs in one detail, that detail belongs to the specific subset, not the whole group. This week's CJNG correction (implying all four Tampa defendants faced narco-terrorism charges when the cited DEA/HSTF release shows only three did) is the same mechanism as 2026-07-20's Iraq/Bangladesh/India seizure-figure misattributions, applied to a set of co-defendants rather than a set of case totals — before generalizing a charge, figure, or outcome to an entire named group, confirm the source applies it to all members, not just some.

Don't state a pending, provisional, or single-instance action as completed or general. Three corrections this week reported a process as finished when the source described it as still in motion: Morocco-Mauritania "sign" a pact when ministers had only agreed to strengthen cooperation; Timor-Leste "becomes" ASEANAPOL's 11th member when the source says the membership is an endorsement still pending formal adoption; a Lahore court ruling "narrows evidence bar for migrant-smuggling prosecutions" generally when the source describes one defendant's bail decision. Match the source's own completion status — agreed-in-principle, endorsed-pending-adoption, one case's ruling — rather than rounding it up to done/general.

Don't attribute a reaction, denial, or formal characterization to a party unless the specific cited source actually contains it. Three of this week's eight corrections put words or an action in an official's or entity's mouth that the cited source doesn't support: Spain's item said Israel "fires back" with a "furious denial" when both cited sources report Israel's foreign ministry did not respond to comment requests; Germany's item had the EU calling a drone plot "state-sponsored terrorism" when no cited official used that phrase (they said "attack by Russia," "kinetic hybrid attack," "the evidence is clear"); a Colombia item reported an ELN drone strike killed an army captain by citing two sources that only covered a prior US Embassy risk *warning*, not that death. Before writing that someone "fired back," "denied," or was formally labeled something, confirm the cited source's own text — not the wider news cycle — actually contains that reaction.

Check the headline against the item's own so_what before publishing — they must not contradict each other. This week's Nigeria rifle-smuggling correction had a headline calling 204 seized rifles "assembled" while the item's own so_what correctly described them as knocked-down components — a self-contradiction a basic headline-vs-body read would have caught, no source-checking required.

"Every claim must trace to this item's own cited source" is the least-internalized lesson in this list — a third occurrence this week, in two new failure shapes. First logged 2026-07-27 and restated 2026-09-07, the pattern recurred again: a China-Philippines ramming item said Beijing's embassy "did not immediately respond" when it had, same day, on the record disputing the Philippine account; and a UN Security Council warning genuinely issued days earlier got attached to a same-topic item whose actual cited AP report never mentions the Council at all. Before publishing a "no comment" claim or attributing a statement to an institution, confirm the specific cited source — not a true fact from elsewhere in the news cycle — is where that claim comes from.

"Don't widen a source's own qualifier" now extends to military-role characterizations, not just superlatives or trend direction. A UK-Yemen item turned the cited source's explicit "purely defensive" single-aircraft refueling role into an offensive "combat-support role... invested in dislodging the Houthis" — the same widening mechanism first logged 2026-07-20 and repeated 2026-08-17, applied to a defensive/offensive military framing rather than a magnitude or trend claim. When a source characterizes a party's military involvement as defensive, support-only, or non-combat, keep that scope — don't upgrade it to an active or offensive role.