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 23, 2026
Day 208 of the US-Israeli air campaign against Iran. Hours 4956–4971 since first strikes at ~06:10 UTC, February 28.
This forecast covers editorials #665 and #666, published at 18:42 and 22:06 UTC on September 22. It scores yesterday's predictions against what our editorial corpus actually recorded, then offers 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 roughly 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 in public. 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 price, a charter rate or a diplomatic meeting, 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, AFP, Guardian, Bloomberg, Financial Times, Kyodo. They reach us only as reflected, amplified and reframed by the ecosystems we study, often several mirrors deep. About our methodology.
Where we are
A coercive instrument arrived that needs no sorties, and nobody in our corpus has seen the document authorizing it. The declaration that from September 23 — today — any airport servicing an Iranian aircraft would itself be exposed entered our corpus through a single Israeli OSINT channel relaying the US Treasury Secretary. We hold no primary American sourcing for it. Four ecosystems then spent thirteen hours reacting to a text they saw only in reflection, and so did we. Iranian institutional denials decayed on a visible schedule — no notice received, only Mahan affected, everyone flying — until Azerbaijan suspended Iran Air, the Turkish carriers suspended to March 2027, and Iran's own aviation spokesman conceded Baghdad and Muscat would stop accepting flights at midnight. By nightfall Mehr was demanding its own regulator clarify whether Oman flights had in fact stopped. Follow Global South & Middle Powers.
The New York contact was confirmed, and both ecosystems described it as the other side's request. Iranian state television acknowledged an Araghchi–Witkoff meeting, presented as held at the American representative's insistence with Iran's conditions delivered. Trump's version was a three-hour productive meeting, with Xinhua naming Witkoff and Kushner. The route there ran through anonymity — Kyodo and Reuters carrying an unnamed senior Iranian official's seven-day Hormuz offer, Fars calling both accounts unreliable, an IRGC officer telling Al Mayadeen the strait is not open and America is lying. The offer was never owned by anyone named; the meeting was, once its agency could be assigned to the adversary. Follow Negotiations & Ceasefire Diplomacy.
The states hosting this campaign are separating their security from it in public, and we learn it through channels aligned with the other side. Qatar's Emir called for a regional collective security system and warned that one party can drag its neighbours into a war that is not theirs. TASS carried Putin and the Saudi Crown Prince agreeing on unimpeded passage through international waterways. Italian aircraft leaving Taif after a damaged Eurofighter reached us via rybar_mena citing La Repubblica; the UK's refuelling-only offer to Riyadh arrived the same route. When the record of coalition posture is written by Russian and Iranian channels, the carriage is the finding.
Two channels broke character in ways this corpus has not recorded before. AbuAliExpress generated an AI-assisted compilation of 150 attacks to rebut a head of state, and boris_rozhin published a dated list of successive Western claims that Iran had killed 32,000, then 35,000, then "more than" 32,000 protesters — a Russian milblog running a fact-check against Western reporting on the same day Moscow's largest Iran-adjacent output was a state-partnered forum on combating fakes. Meanwhile the oil tape moved on a restarted Saudi pipeline while Fars read the identical print as a negotiation rumour, and US diesel hit a record $6.53 a gallon with the world's largest diesel exporter contemplating an export ban. Follow Strait of Hormuz & Oil.
Yesterday's scorecard
These predictions were written from #665 for editorials published through ~22:00 UTC today. One editorial qualifies — #666 — and its window (09:00–22:00) overlaps #665's (05:37–18:37) by roughly nine and a half hours. Most of what follows is therefore scored against material we had already partly seen. We say what that does to the verdicts below the table.
| # | Prediction | Type | P | Verdict |
|---|---|---|---|---|
| H1 | Claim hygiene applied inside an outlet's own ecosystem | E | 76% | Confirmed. Mehr by nightfall demanding Iran's own aviation authority clarify whether Oman flights had stopped — a state outlet turning its scepticism on its own regulator. fotrosresistancee hedging that the ban "for now" seemed unofficial. Fars refusing the anonymous Hormuz accounts as unreliable. Eleven windows positive. |
| H2 | Tehran's two negotiation channels both persist, neither retired | E | 72% | Partial. Both channels ran as specified — anonymous wire placement, on-record IRGC denial. But the window produced a resolution we did not model: state television confirmed the meeting while assigning the approach to Washington. The offer was never owned by a named official; the contact was, with agency reversed. Our refutation condition did not fire, and our confirmation condition does not describe what happened. |
| H3 | Iranian aviation denials stay unretracted while third countries execute | E | 70% | Confirmed. Azerbaijan on Iran Air, Turkish Airlines, AnadoluJet and Pegasus to March 2027 confirmed by Xinhua, Baghdad and Muscat at midnight, Reuters on Iraq weighing a shift to Najaf. No Iranian outlet issued a correction notice. The denials were superseded, not retracted — which is the distinction we were testing. |
| H4 | The air closure is answered with logistics copy rather than rebuttal | E | 70% | Partial. The rebuttal register appeared exactly as predicted and from one direction only — Beijing's objection to "illegal" unilateral sanctions via Radio Farda and Press TV, with Tehran's neighbours silent. But the two-or-more new corridor items we required did not appear in the fresh window. Every rail, road and customs item we cited belongs to the overlapping hours. Confirmation condition half-met. |
| H5 | Civilian-harm data filed as administration inside Iran's ecosystem | E | 68% | Confirmed, both limbs. Velayati's 120 empty chairs against Fars and IRNA on 168 absent students, unreconciled. Hormozgan's governor's office denying a two-month closure at 15:56, the deputy governor announcing two weeks of virtual schooling at 16:04, framed as student health and educational quality with the strike environment unmentioned. |
| H6 | A relay channel breaks its own register | E | 64% | Confirmed, three times, and one instance exceeds what we predicted. boris_rozhin fact-checking Western casualty claims, Middle East Spectator refusing to carry UN speeches, and AbuAliExpress generating an AI-assisted evidentiary compilation to rebut a head of state. The third is a new behaviour class, not a register break: an aggregator manufacturing its own evidence rather than relaying someone else's. |
| H7 | One institutional text harvested selectively by ≥3 ecosystems, omission as position | EW | 74% | Confirmed. Trump's UNGA text reached us as "annihilate the Islamic Republic" (Washington Free Beacon), as a likely deal after the midterms (Al Jazeera English), and as a re-registration of the speaker himself (Mehr, ISNA). The cleaner specimen is the UN inquiry: Jerusalem Post reported both halves accurately and ordered them so only one became a headline. |
| H8 | One government issues two incompatible quantifications, neither carrier notes the pairing | EW | 72% | Confirmed, with a counter-instance in the same window. 120 chairs against 168 students, nobody reconciling. But our stated refutation behaviour also fired on a different object: boris_rozhin published Western protester-death figures as a dated sequence, pairing them at the point of publication. The pairing move exists in this corpus; it is performed against adversaries, not against one's own side. |
| H9 | A price print read as negotiation artifact in one ecosystem, supply fundamentals in another | EW | 70% | Confirmed exactly as specified. AJA on crude down more than two percent on the East-West pipeline restart and Yanbu preparations; AbuAliExpress logging Brent at 98 against 108; Fars reading the identical tape as a negotiation rumour. One of those explanations has a physical referent. |
| H10 | Coalition-strain material reaches us only via ecosystems that benefit from carrying it | EW | 66% | Confirmed, on the same two objects rather than new ones. rybar_mena citing La Repubblica on Italian aircraft out of Taif; the UK Defence Secretary's refuelling offer arriving by the same route. No Western-origin pickup in our corpus. Object persistence, not fresh instances — which is weaker than it looks. |
| H11 | ≥8 new named objects across ≥3 categories | W | 80% | Confirmed comfortably. 4,200 megawatts off the grid, the Najaf reroute, US diesel at $6.53, a contemplated US diesel export ban, 14.5 percent of global diesel supply, the UK House of Commons Library assessment, a 150-attack AI compilation, the 32,000/35,000 sequence, an LPG tanker hit by "unknown projectiles," 17 dead in 19 Yemen strikes, Qatar's collective-security proposal, CENTCOM altering its lethal targeting process and AI use. |
| H12 | A Hormuz legal instrument acquires ≥3 incompatible characterizations while its text stays unseen | W | 62% | Refuted. Neither France's draft Security Council resolution nor Iran's 20%-of-cargo-value transit penalty appears in the fresh window at all. Our stated refutation condition — the instruments not being covered — is met cleanly. What arrived instead was the inverse: Rudaw filing the window's only concrete shipping incident and leaving the attribution open, an object nobody characterized rather than a document everybody characterized differently. |
Summary: 9 confirmed, 2 partial, 1 refuted, against a mean stated probability of 71%. We do not treat this as a calibration reading and neither should anyone else. Nine and a half of the thirteen hours we scored against were hours we had already observed when we wrote the predictions, which makes several of these verdicts closer to description than forecast. That is the second consecutive set rendered non-comparable — last set by a seven-day publication gap, this one by window overlap — and the fix is ours to make: our 12-hour editorial cadence with a one-hour design overlap produces windows that no longer align with a daily forecast cycle. The two substantive lessons survive the caveat. H2 failed because we modelled who owns a claim and the ecosystem instead negotiated who asked for the meeting. H12 failed on a premise we have now watched fail twice: that contested formal material gets characterized. In this corpus it more often gets dropped.
Today's predictions
Review window: editorials published through ~22:00 UTC, September 24, 2026.
H1 (76%) [Type E]: An outlet applies claim hygiene to a claim originating inside its own ecosystem. Eleven windows positive, our longest continuous series. Confirmation is an outlet flagging missing confirmation on something its own side is leaning on, demanding clarification from an institution it normally relays, publishing a deliberately conservative reading, or correcting itself to its own audience — Mehr pressing Iran's civil aviation authority is today's template. Refutation is a window of outward-only skepticism, with every fact-check pointed at an adversary. We are testing whether these ecosystems remain internally argumentative rather than uniform relay layers.
H2 (74%) [Type E]: The aviation cordon's effects continue to be enumerated while its authorizing text stays outside our corpus. Today is the stated effective date. Confirmation is further carrier, route, airport or ground-handler suspensions attributed to the measure — reported by Gulf, Caucasus, Central Asian, Chinese or Iraqi outlets — with no outlet in our corpus publishing, quoting or linking a primary US Treasury text or statement. Refutation is the operative document circulating, or a Treasury-origin item reaching us directly rather than through an aggregator. We are measuring how far a policy can propagate through four ecosystems on a single relayed sentence.
H3 (72%) [Type E]: Iranian civilian-harm data continues to arrive as administration rather than as harm. Confirmation is a further intra-Iranian count divergence on the same object left unreconciled, or a closure, rationing, repair or rescheduling notice that omits the strike environment as cause while the same government files infrastructure damage as a war-crimes brief elsewhere. Refutation is a single authoritative count adopted across Iranian outlets, or an administrative notice naming the strikes directly. The Minab figures and the eight-minute Hormozgan reversal are the specimens; we are watching whether the pattern is procedural or was a one-day artifact of pressure.
H4 (70%) [Type E]: The air closure is answered with corridor and logistics copy carried alongside denunciation, with no outlet connecting the two registers. This went partial yesterday because the corridor items did not renew, so we re-run it with the same test. Confirmation is two or more new items on rail to China, the Armenian or Azerbaijani road links, port or customs throughput, or Central Asian trade protocols, carried by Iranian or Chinese outlets as routine commercial news in a window that also carries sanctions denunciation. Refutation is rhetoric without the infrastructure copy for a second consecutive window, or an outlet explicitly framing the corridor items as sanctions workaround.
H5 (66%) [Type E]: An aggregator or milblog produces its own evidentiary artifact rather than relaying someone else's. AbuAliExpress assembling an AI-generated 150-attack compilation to rebut a head of state, and boris_rozhin publishing a dated sequence of Western casualty claims, are two instances in one window of a behaviour this corpus had not recorded. Confirmation is a further compilation, dated sequence, dataset or generated visual produced by a channel whose normal function is relay, and offered as evidence rather than as commentary. Refutation is the aggregator layer returning to pass-through behaviour. If this recurs it is a change in what these channels are for.
H6 (72%) [Type E]: Yemen's civilian ledger stays sorted by ecosystem, with no outlet carrying both sides' counts. Confirmation is casualty figures reaching us through Houthi media, Al Manar, Xinhua or Russian relay while Gulf-owned outlets lead on Saudi sovereignty, Mecca targeting or Houthi-attributed displacement, with the two ledgers never appearing in the same item. Refutation is a Gulf-owned outlet carrying a Houthi-sourced casualty count, or a resistance-axis outlet carrying IOM's attribution of displacement to the Houthi advance. Al Arabiya on the UN's 230,000 projection shows the exception is possible, which is why the rule is worth measuring.
H7 (74%) [Type EW]: One institutional text is harvested selectively by ≥3 ecosystems, with at least one carriage's position visible in what it cuts or how it orders. Confirmed in nine of the last eleven windows. Candidates are live: Pezeshkian's General Assembly address, the UN inquiry report, the Qatari collective-security proposal. Confirmation is a single text reaching us in three or more non-overlapping cuts with an identifiable excision or ordering decision doing editorial work — the Jerusalem Post headline sequence is the cleanest recent specimen. Refutation is a document travelling with its content intact across ecosystems.
H8 (70%) [Type EW]: A diplomatic contact is narrated by both ecosystems as something the other side requested. Iranian state television placed the Araghchi–Witkoff meeting at the American representative's insistence; the American account made it a productive three-hour meeting Washington convened. Confirmation is a further contact, call, message or mediator delivery where Iranian and US-aligned carriage each assign the approach to the adversary, with no outlet in our corpus carrying both attributions. Refutation is a contact reported with agreed authorship, or one side declining to characterize who asked. This measures whether diplomacy can be reported at all without being converted into evidence of the other side's need.
H9 (70%) [Type EW]: An energy print is read as a diplomacy artifact in one ecosystem and as physical supply in another. Confirmation is a further indicator — Brent, a charter rate, a refinery or pipeline restart, a diesel crack, an insurance premium — arriving in Iranian or resistance-axis carriage as evidence about negotiations and in Gulf, Chinese or Western-relayed carriage as evidence about barrels and hulls, both citing the same number. The US diesel story is the object to watch: a record pump price and a contemplated export ban put a domestic American number inside the war's information space. Refutation is a price move carried as a market fact with neither construction attached.
H10 (68%) [Type EW]: Coalition-strain material continues to reach us only through the ecosystems that benefit from carrying it. Confirmation is a further basing, aircraft, withheld-capability or host-state-distancing item appearing in Russian-relayed, Persian-language or Chinese carriage with no Western-origin pickup in our corpus — Qatar's collective-security framing and the Italian redeployment from Taif are the current templates. Refutation is a Western-origin outlet in our corpus carrying such material, or the strain items disappearing entirely. We record who is willing to carry a claim, which is a separate question from whether the claim is true.
H11 (80%) [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 naming rate is one of our few continuous series and it has now run above threshold for eleven consecutive scored windows. The aviation cordon's effective date should push the diplomatic and financial categories; we are watching whether the maritime category renews without a fresh incident.
H12 (64%) [Type W]: A concrete Hormuz or Gulf shipping incident enters the corpus with no ecosystem assigning a perpetrator, and no forensic, salvage or insurer layer follows. This replaces the instrument prediction refuted yesterday. Rudaw filed an LPG tanker struck by debris from "unknown projectiles" and left the attribution open — the only outlet in the window willing to. Confirmation is a further vessel, port or platform incident carried with attribution unassigned or with two incompatible attributions and nothing published from a classification society, insurer, salvage operator or shipping trade press. Refutation is an arbitrating layer entering our corpus, or an incident that every ecosystem confidently attributes.
What we can't see
By design we do not monitor Western mass media directly: the Kyodo and Reuters accounts of the Iranian offer, the La Repubblica report on the damaged Eurofighter, the Bloomberg item on CENTCOM's targeting changes and the Reuters Yanbu material all reached us through belligerent-aligned intermediaries, each keeping a different half. 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 have no primary American sourcing for the measure that reorganized four ecosystems' coverage for thirteen hours — a Treasury declaration we have seen only in an Israeli aggregator's reflection. We have no forensic, salvage or shipping-industry layer, which is why a damaged hull and two incompatible accounts of strait throughput sit in our corpus with nothing to arbitrate them. And our own scoring instrument is degraded: yesterday's predictions were scored against a window that overlapped the window they were written from by nine and a half hours, which inflates every verdict above and is a limitation of our cadence, not of the environment.
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.