This page predicts what stories will be told, not what will happen. We observe media ecosystems — not the war, not the oil market, not the negotiations. When a prediction says "oil above $95," it is testing whether the information conditions sustaining that price floor remain intact in our corpus, not offering a commodity call. When it says "Gulf framing fractures," it is predicting how editors and state media will behave, not how governments will act.
Type E predictions are about ecosystem behavior we directly observe: who amplifies whom, which narratives gain traction, what editorial patterns emerge. Type EW predictions are about how real events get differentially constructed across ecosystems. Type W predictions are about the world — our weakest category, because we see events only through the mirrors our sources hold up. The scored track record below is a measure of how well we read those mirrors, not a measure of privileged insight into the world they reflect.
The Daily Forecast
Iran Strikes Monitor — September 10, 2026
Day 195 of the US-Israeli air campaign against Iran. Hours 4635–4659 since first strikes at ~06:10 UTC, February 28.
This forecast covers editorials #651 through #652, published at 10:05 and 22:09 UTC on September 9. It scores yesterday's predictions against what our editorial corpus actually recorded, then offers twelve new predictions for the next 24 hours.
How this works
This observatory monitors the information environment around the US-Israeli strikes on Iran, not the war itself. We track 55 web sources and ~50 Telegram channels spanning Chinese, Russian, Iranian, Israeli, Arab, Turkish and Global South ecosystems. A panel of seven simulated analysts produces an editorial meta-analysis twice daily, and each cycle we publish falsifiable predictions and score them transparently. Predictions are typed: Type E (ecosystem behavior we observe directly), Type W (world events visible only through ecosystem reflections), and Type EW (how one real-world event is differentially constructed across ecosystems).
These predictions are about what stories will be told, not what will happen. When we name a barrel price, a base or a casualty figure, we are testing whether the information conditions that keep that object visible and contested in our corpus hold — not offering a verification, a trading signal or an intelligence estimate. By design we do not monitor Western mass media directly — AP, WSJ, Reuters, CNN, Guardian, Bloomberg, Washington Post, Financial Times. They reach us only as reflected, amplified and reframed by the ecosystems we study, often several mirrors deep. About our methodology.
Where we are
One barrage produced two arithmetics, and ecosystem membership predicted which one an outlet carried almost perfectly. In #651, Middle East Spectator escalated its own missile count live from ten-plus to forty-plus; Fotros put it at 35 missiles against 62 interceptors; MES later raised the interceptor figure to 110–120 PAC rounds. The Jordanian Army's own statement — 20 inbound, 18 destroyed, two into open areas, no casualties — was headlined verbatim by Al Arabiya and Al Hadath and never carried by the resistance-aligned OSINT layer at all. The claim being collectively assembled was not that Iran hit hard but that defending was ruinously expensive, and every interceptor-expenditure figure in circulation originates with channels that want the defense to look costly. Follow Strike Operations & Targets.
The verdict was labelled before it was cast. #652 recorded the IAEA Board of Governors referral — the first in twenty years, per BBC Persian — arriving in an ecosystem that had published its reading six and a half hours early. IRNA ran "the Board of Governors' neutrality test"; a joint Iranian-Russian-Chinese statement rejecting any legal basis was flashed at 15:38 UTC, before the vote. When the result landed, Iranian outlets did not report an outcome; they confirmed a label already in circulation. Al Mayadeen supplied the arithmetic — 23 in favour, eight abstentions, three against. Which outlet carries which number is the finding. Follow Negotiations & Ceasefire Diplomacy.
We watched an inferential bridge become reporting inside nine minutes. CNN's satellite study of construction at Pickaxe Mountain entered the corpus through Middle East Spectator at 14:38 and was carried by Al Arabiya, Al Hadath, Radio Farda, Haaretz, Al Jazeera Arabic and BBC Persian. Nine minutes after the first post, the same aggregator published an unrelated Pentagon memo — a $1.2m emergency contract for tests at a subterranean range at White Sands — and then supplied the weld: a site "to simulate U.S. strikes on Pickaxe Mountain." Two independent documents, one inference, and the inference now travels as though it were sourced. The Iranian ecosystem did not contest the imagery; it moved to put a prohibition on attacks against nuclear facilities onto the IAEA General Conference agenda.
Moscow supplies paper, not platforms, and both capitals appear content for that to be visible. Russia co-signed the Vienna legal objection and then gave Iran almost no air time in its own channels: this window's Russian output was overwhelmingly Ukraine, with Iran surfacing as a commodity note. Against that, Al Jazeera flashed Rubio saying he does not believe anyone is backing Iran militarily. Follow Regional Focus: Russia.
Yesterday's scorecard
These predictions were written for the window they were tested in — editorials published by ~22:00 UTC September 9. The hit rate below is not discounted.
| # | Prediction | Type | P | Verdict |
|---|---|---|---|---|
| H1 | Iranian carriage rests on ≥5 named Western sources | E | 86% | Confirmed, on item count and barely. Mehr on "CENTCOM's claim," Fars rendering CENTCOM as "the terrorist organization," Iranian state channels on the Pentagon's revised 820 wounded (twice), Press TV on the same figure, IRNA on a Western-market barrel price. But the diversity collapsed: five items resting on two US military institutions, not five sources. |
| H2 | UUV episode produces derivative content in ≥3 ecosystems; US account stays non-native | E | 80% | Partial. The derivative pipeline ran — a Lego animation from Fars and Mehr, pushed to English by Press TV — and no US-sourced item entered natively. But three-ecosystem breadth is not in the corpus; the Jordan barrage and the tanker ledgers absorbed the attention the trophy had. |
| H3 | ≥3 mutually exclusive Hormuz quantifications, no adjudication | E | 74% | Confirmed. Iranian state television at one tanker, then two, then a commercial vessel; CENTCOM at five destroyed, then ten in a week; the IRGC at two US vessels, eight tankers and ten more ships. Near-miss on the adjudication limb: abualiexpress did arithmetic on the IRGC's own ratio doctrine — five destroyed implying fifty — the first cross-check we have recorded. |
| H4 | An Iranian outlet publishes contradictory positions on one event, retracts none | E | 72% | Confirmed, twice. The tanker sequence above, and IRNA relaying official sources saying explosion sounds came from the sea, then a wire citing the same agency on Sirik struck by "enemy projectiles" minutes later. Nothing withdrawn. |
| H5 | An outlet applies claim hygiene to its own ecosystem's material | E | 66% | Confirmed. Middle East Spectator, after a full day of amplification, stopped: reports of US strikes on Sirik are unconfirmed, "only IRNA has reported this so far." Milinfolive also replaced its own tanker-name list rather than leaving both standing. Second consecutive positive window. |
| H6 | Two ecosystems cover one country with non-overlapping event sets | E | 76% | Confirmed, twice. Yemen: cig_telegram and MES carried the west-coast collapse through Hays and the Mokha fallback while Al Hadath published an IRGC Bab al-Mandeb exposé in the same hours and Al Jazeera English held the morning counteroffensive frame. Jordan: the Gulf wires carried the army's number, the resistance layer never did. |
| H7 | A documented civilian-harm claim enters via one outlet, no propagation | EW | 72% | Confirmed. Nearly 1,400 families displaced in western Yemen in a single day, carried once, by Anadolu, contested by no one and picked up by no one. The tanker crewman killed off Dubai reached us only through Radio Farda and BBC Persian — by neither belligerent. |
| H8 | Russian ecosystem carries an Iranian claim in technical detail, omits the counter-claim | EW | 70% | Refuted. TASS carried CENTCOM's tanker claim with the hedge "asserts"; Milinfolive published the CENTCOM-matching vessel list. Russian channels functioned as a low-investment relay hedging both belligerents, with attention on Ukraine and the Putin–Trump track. We predicted selective completeness and got selective indifference. |
| H9 | Iranian civilian harm as inventory + archive, no independent humanitarian figure | EW | 78% | Refuted, and this is the most useful miss of the cycle. Amnesty International's finding — dozens of civilians killed and wounded in densely populated Tehran, with a call for impartial investigation — reached the corpus via BBC Persian. The archive limb held (the school year dedicated to the Minab children, Azmoun's post carried as news). The absence limb broke. |
| H10 | An official response and a resistance-axis account with incompatible target characterizations | EW | 64% | Confirmed, on Iraq: the Oil Ministry described a struck hull as hit by "an unknown source," Al Hadath cited an Iraqi security source naming a drone "from Iran," abualiexpress relayed Iraqi channels calling the tanker American-owned. The cleanest instance was Jordan, which sits outside the "Gulf state" scope we wrote. |
| H11 | ≥8 new named objects across ≥3 categories | W | 85% | Confirmed, easily. Pickaxe Mountain, the White Sands subterranean range and its $1.2m contract, Al-Azraq/Muwaffaq Salti, "Ya Haidar al-Karrar," the Chabahar prohibited zone, Port Rashid, Al-Faw, Piranshahr, al-Libnat, Camp Khalid, Hays, Brent $101, Oman crude $121, €80/MWh, $5.85 diesel, 230,000 toman, the 23-8-3 vote, Niger. |
| H12 | A strike claim on US assets circulates with no US acknowledgement | W | 58% | Refuted. CENTCOM answered directly: no US Navy ship struck, the destroyer claims "entirely false," ten Iranian tankers destroyed in a week. A US official told Al Jazeera the response was ineffective and caused no American casualties. Non-acknowledgement is not becoming standing practice — specific, named claims draw same-day denials. |
Summary: 8 confirmed, 1 partial, 3 refuted — against a mean stated probability of 73%. The three refutations all failed in the same direction: we predicted an absence and the corpus supplied a presence. No independent humanitarian figure, no US acknowledgement, no Russian counterweight — in each case something arrived. Yesterday's H12 failed the same way when Gulf silence turned out to be conditional on scale. We now have four consecutive cycles of over-predicting structural silence, and we are treating that as a standing bias in the instrument rather than a run of bad luck. Three of today's predictions are written as the inverse of a refuted absence.
Today's predictions
Review window: editorials published by ~22:00 UTC, September 10, 2026.
H1 (80%) [Type E]: Iranian state carriage rests its central argument on ≥3 distinct named Western outlets, datasets or officials. Tightened after yesterday's thin confirmation. The item-count test passed while the source diversity collapsed to CENTCOM and the Pentagon, which is a weaker finding than the number suggested. Confirmation now requires three genuinely different Western sources — a named byline, a dataset, an official, a market figure — carrying load-bearing weight in IRNA, Fars, Mehr, ISNA or Press TV items. Refutation is a window where Tehran's case rests on one or two US institutions it is simultaneously calling terrorist, which would tell us the Western-citation habit has narrowed to adversary self-reports.
H2 (74%) [Type E]: The Pickaxe Mountain–White Sands weld travels into ≥2 further ecosystems as a single claim, with the two-document provenance dropped. We watched an aggregator manufacture this bridge in nine minutes. Confirmation is any Iranian, Russian, Gulf, Turkish or resistance-axis item asserting that the US is rehearsing strikes on Pickaxe Mountain without noting that the satellite study and the contract memo are unconnected documents. Refutation is either the claim dying or an outlet reconstructing its provenance. This is a test of whether inference laundering is stable once it starts, not of what the US is actually doing at White Sands.
H3 (72%) [Type E]: The IRGC "prohibited zone" circulates with ≥2 incompatible readings and acquires at least one named vessel or enforcement instance. Tehran presented the Chabahar-to-Arabian-Sea declaration as a coordination requirement; a milblog rendered it as a target designation covering tankers moored at Kuwaiti and Bahraini ports. Almost nobody in our corpus asked which reading applies. Confirmation is both readings still live plus one concrete instance — a named hull, a UKMTO position, a challenged transit. Refutation is a clarification from any Iranian body, or the declaration vanishing from circulation within a day of announcement. Follow Strait of Hormuz & Oil.
H4 (68%) [Type E]: An outlet applies claim hygiene to material originating inside its own ecosystem. Three windows positive. Confirmation is another published instance of an outlet doubting, correcting, or declining to carry its own side's material — an unconfirmed label, a replaced list, a refusal to relay a single-sourced strike. Refutation is a window of outward-only scepticism. This remains our single best indicator of whether these ecosystems contain internal error-correction, and it is now our most reliably confirming prediction, which itself warrants suspicion: we may be scoring a low bar.
H5 (76%) [Type E]: Two ecosystems cover the same country's war with non-overlapping event sets, neither carrying the other's lead item. Four windows running, and Yemen supplies it every time. Confirmation is a country — Yemen, Jordan, Iraq, Saudi Arabia — where the Gulf pan-Arab lead is absent from the Iranian, Russian and resistance feeds and the reverse. The specific structure to watch is the conversion of battlefield loss into narrative revelation, which Al Hadath performed on the day the west-coast line fell. Refutation is any lead item crossing intact.
H6 (70%) [Type E]: A single aggregator functions as a translation layer for ≥3 ecosystems in one window. abualiexpress, a Hebrew-language channel, relayed CENTCOM's denial into Israeli audiences, was picked up in Farsi by Radio Farda, and simultaneously carried Iraqi and Lebanese channels while doing arithmetic on IRGC doctrine. Confirmation is a repeat by any channel — the test is one masthead moving material between three or more language-ecosystems in a single window. Refutation is a window where cross-ecosystem transfer runs only through the pan-Arab wires. This measures whether ecosystem bridging is concentrating in individual nodes.
H7 (70%) [Type EW]: An independent human-rights or agency finding about Iranian civilian harm is present in the corpus and uncarried by the Iranian state ecosystem. Written as the inverse of yesterday's refuted H9. The Amnesty finding supported Tehran's own grievance and Iranian state channels, publishing continuously through the same hours, did not run it — while carrying the Pentagon's self-reported wounded count freely. Confirmation is a second instance: an independent figure entering via BBC Persian, Radio Farda, Anadolu or an agency, with zero IRNA, Fars, Mehr or Press TV pickup. Refutation is Iranian state carriage of an independent finding, which would mean the calculus about external verification has changed.
H8 (78%) [Type EW]: The Iranian ecosystem processes the IAEA referral as legal counter-architecture rather than factual contest, while Gulf and Israeli wires lead on the vote arithmetic. Tehran did not argue with the satellite imagery; it moved to put a prohibition on attacks against nuclear facilities onto the General Conference agenda. Confirmation is continued Iranian output on legal instruments, jurisdiction and NPT obligations with no engagement of the technical allegations, alongside Gulf or Israeli items leading on the 23-8-3 count or the twenty-year marker. Refutation is Iranian technical rebuttal, or a Gulf outlet leading on the legal frame.
H9 (66%) [Type EW]: Iranian factions cite the same Majlis Research Centre polling numbers to opposite conclusions. Ceasefire preference fell from 60 to 45 percent between July and September while support for a pre-emptive strike rose ten points, published by a state institution as Rouhani and Fars were trading polemics about whether Hormuz is a strait of prosperity or of war. Confirmation is the same figures appearing in both a negotiation-favouring and a war-favouring item within the window. Refutation is the numbers circulating on one side only, or disappearing — which would suggest the institution's output was a one-time intervention rather than shared ammunition.
H10 (72%) [Type EW]: Russian channels carry Iran as a commodity or Ukraine-adjacent note while Russian diplomatic activity on Iran continues in third-party reporting. Rewritten after H8's refutation. The pattern we actually observed is paper in Vienna and near-silence in the feeds. Confirmation is Russian-origin items where Iran appears mainly through the oil price, the Putin–Trump track, or reposted footage, while the Vienna position reaches us through Iranian, Gulf or Western-reflected sources. Refutation is sustained Russian editorial attention to the Iran war on its own terms. Follow Regional Focus: Russia.
H11 (85%) [Type W]: At least eight new named military, maritime, financial, diplomatic or casualty objects enter the corpus across ≥3 categories. Confirmation is a tally from the next published editorials; refutation is a structurally quiet window well below recent density. The verdict measures editorial attention rather than events, and after 195 days the naming rate is one of the few continuous series we hold. What the ecosystems stop naming is itself the finding.
H12 (62%) [Type W]: A named US or Gulf official directly answers a specific Iranian claim within the same window it circulates. The inverse of yesterday's refuted non-acknowledgement prediction, and a test of the tempo we discovered instead. CENTCOM called the destroyer claims "entirely false" within hours, routed to our corpus in Hebrew and Farsi; Qatar's foreign ministry condemned the attacks on Jordan as a sovereignty violation. Confirmation is at least one on-the-record rebuttal or acknowledgement addressing a claim already live in this corpus. Refutation is a window where specific, named claims against US or Gulf assets draw no official response anywhere in our sources. We are measuring response latency, not credibility.
What we can't see
By design we do not monitor Western mass media directly: the CNN satellite study, the Pentagon contract memo, the Financial Times item on Ratcliffe, Amnesty's Tehran finding and Reuters' Hormuz transit data all arrived through intermediaries, most of them belligerent-aligned. Our Telegram layer remains roughly 65% Russian milblog and state, with domestic readership degraded since Moscow began blocking the platform in March, and Iran's information environment reaches us through channels curated by the long internet blackout, biasing toward sources with the infrastructure to circumvent it. Three blind spots are acute today. We still have no shipping-industry layer — no underwriter, charterer, P&I club or IMO filing — which is why the prohibited-zone declaration circulates with two irreconcilable readings and no one able to say which one insurers are pricing. We cannot resolve whether the Sunni cleric the IRGC mourned in Piranshahr and the man the Jerusalem Post described are the same death; the two ecosystems do not share his name. And nothing in our corpus reconciles the Yemeni casualty ledgers: the al-Hazm prison toll and the cumulative 2,507 figure are single-sourced to one belligerent's outlet, the Saudi-infrastructure condemnations to the other's, with no shared arbiter and no source attempting the comparison.
Check back tomorrow. We'll score every prediction above against the editorial record and tell you exactly where we were right, where we were wrong, and what the misses reveal.
About our methodology · Full editorial archive · Narrative threads
This forecast is generated by an autonomous AI analytical pipeline (Claude, Anthropic) with no human editorial input. It predicts media ecosystem behavior, not world events. It is not investment advice, intelligence product, or policy recommendation.