Timestamped & publicly verifiable
Prediction Audit Trail
Every M3I prediction with the date it was made, the original call, the outcome, and links to the source posts. No edits, no hindsight.
The record
#1 — Negotiated deal by ~April 14
Made March 10, 2026 · Day 11
The call
Assigned 42% probability to a negotiated deal / ceasefire by ~April 14.
Outcome
Ceasefire confirmed April 7 (Day 38) — 7 days ahead of target.
#2 — Resolution window April 6–13
Made April 1, 2026 · Day 32
The call
Narrowed the resolution window to April 6–13.
Outcome
Ceasefire landed April 7 — inside the predicted window.
#3 — Ceasefire by ~April 14
Made March 10, 2026 · Day 11
The call
42% probability of a ceasefire by ~April 14 (Path B).
Outcome
Confirmed Day 38 (April 7) — 7 days ahead of the outer target.
#4 — Oil above $100 as escalation signal #1
Made March 30, 2026 · Day 30
The call
Flagged oil above $100 as the primary Path A escalation trigger.
Outcome
Brent hit $109 on April 2 — as modeled.
#5 — 81% ceasefire-collapse risk
Made April 9, 2026 · Day 40
The call
Model assigned 81% headline-driven collapse risk despite a 57% extension probability.
Outcome
Vance left Pakistan after 21h with no deal (Day 43) — the 5th failed cycle; limited naval blockade announced.
#6 — Framework deal by ~May 14–15
Made April 14, 2026 · Day 45
The call
Composite model predicted a framework-deal resolution by ~May 14–15 with 60% deal probability (Pattern Engine first targeted ~May 5).
Outcome
Missed. No deal by mid-May — the model was too aggressive on timing, and the resolution window had to be pushed out.
What it still got right
The trajectory: ceasefire → failed talks → escalation → de-escalation → structured diplomatic process — exactly what played out. Correctly classified a resource war (not ideological) and called the blockade as escalate-to-de-escalate, not terminal.
Why it missed
The date. Underestimated Iran's pain tolerance behind five preconditions, did not model the Lebanon front as a separate blocking variable, and treated enrichment as resolvable rather than a standalone deadlock.
#7 — War settled within the revised ~July 31 window
Revised call May 15, 2026 · CONFIRMED on signature Jun 17, 2026
The call
After the May miss, the model re-anchored the resolution window to ~July 31 and held a resolution bias (deal more likely than collapse) through the escalate-to-de-escalate cycle that followed.
Outcome — CONFIRMED on signature, ~6 weeks ahead of the revised window
SIGNED 2026-06-17 at Versailles. President Trump signed a hard copy of the US–Iran 14-point agreement at the Palace of Versailles during a dinner with French President Emmanuel Macron after the close of the G7 summit —
two days earlier than the previously announced Jun 19 Bürgenstock ceremony in Switzerland. Iranian President Masoud Pezeshkian's signed counter-copy was published by Iranian state media IRNA. At Iran's insistence the document was signed in
both English and Farsi. The G7 joint statement praised the deal. Trump set a public
60-day implementation redline:
"If it doesn't get done in 60 days, we go back to bombing." The Hormuz reopening commitment stands; the US naval blockade has been lifted. Resolved roughly
six weeks ahead of the ~July 31 window.Sources:
CNN live updates — Trump signs hard copy of US-Iran agreement (Jun 17) ·
NBC live updates — G7 summit / US-Iran agreement (Jun 17) ·
CNN Politics — Dinner at Versailles, alpine air and an Iran agreement ·
NPR — Trump administration releases agreement with IranEarlier track (now superseded by signature): Jun 11 — Trump announced
"We just made a great settlement of the war with Iran," with documents in final form. Jun 15 — Trump, Vance and Iran's Ghalibaf digitally signed the 60-day MoU. Jun 17 — Hard-copy signature at Versailles closes the prediction.
Annotation (Jul 19): Signing ≠ implementation. The MoU was declared "over" by Trump on Jul 8 after the Hormuz tanker strikes and renewed US strikes. This prediction scored the signature event, which occurred as called; the implementation phase is tracked separately in predictions #8–#9 below.
#8 — M3I outcome call: frozen armistice, active phase ends ~Nov 2026–Feb 2027
Made July 19, 2026 · Day 141 · OPEN — falsifiable
The call
No decisive military outcome and no formal treaty. Resolution via a frozen armistice brokered by a guarantor state (Pakistan/Qatar/Oman/Egypt). Active kinetic phase ends ~Nov 2026–Feb 2027 (9–12 months from the Feb 28 start). Sequence: Aug 16 redline passes without full compliance → renewed US strikes late Aug/Sep → Iran absorbs and sustains low-grade Hormuz pressure → mutual economic exhaustion by Q4 → second MoU or de facto quiet neither side calls a treaty. Terms likely deliberately ambiguous: Iran suspends attacks on US bases/Gulf infrastructure; US pauses strikes and relaxes naval pressure; monitored commercial passage through Hormuz; nuclear negotiations and IAEA access deferred to phase two; both sides claim victory. Key variable: the Mojtaba Khamenei succession — a fractured decision structure in Tehran means no faction can deliver compliance (the Vietnam-'73 problem), the main argument against a fast ceasefire.
#9 — External model call: face-saving ceasefire by Aug–early Sep 2026
External model — ChatGPT · Dated July 19, 2026 · OPEN — falsifiable
The call
An ugly, face-saving ceasefire by August–early September 2026, followed by months of tense negotiations and occasional violations; no stable military route to victory for either side. Same armistice structure as the M3I call — the divergence is timing only: this call treats the current flare-up as the final escalation cycle; the M3I call treats it as the penultimate one.
Resolution criteria — set in advance, scored as written, no retro-fitting
(a) A ceasefire — formal or de facto quiet, 30+ consecutive days without major strikes — beginning before Sept 15, 2026 → the external call scores correct and the M3I timeline is wrong by a quarter-plus. (b) The Aug 16 redline passes → renewed US strike campaign → ceasefire materializing Nov 2026 or later → M3I sequencing scores correct. (c) A pause that begins in September but breaks within 30 days counts for neither — logged unresolved and re-opened. Whichever resolves gets scored; the other gets scored against.
Methodology revision — reference class misapplied
Logged July 19, 2026 · scored against the model
What was wrong
At authoring time the engine classified the US–Iran conflict as a resource war and applied the ~103-day resolution median. That was a reference-class error: the market-facing variable (oil/Brent) was allowed to define the conflict taxonomy instead of the war aims. The war opened Feb 28 with regime-change strikes and nuclear-program targets — not resource seizure; Hormuz is leverage, not a prize. Correct comparables: the 1984–88 Tanker War, Korea-style armistice conflicts, and compellence campaigns — median durations in years, not ~100 days.
What survives
The behavioral read (strike-for-leverage, then deal) was correct — the Jun 17 signing call stands CONFIRMED as the record shows. What failed was the duration base rate, which is why the signed deal did not implement. The forward call now uses the compellence-war reference class. The original reasoning is preserved unedited in "How the model called it" below, with this revision logged against it.
How the model called itAI Predictive Intelligence
The War-Pattern Engine — resource wars vs ideological wars
The call wasn't a hot take — it was a base rate. The Pattern Engine analysed every US conflict since 1861 (the Civil War to today) and sorted each into one of three types: resource, ideological, or occupation. The split that drives everything: wars fought over resources — oil, ports, shipping lanes — resolve dramatically faster than wars fought over ideology.
Resource wars · oil · ports · shipping lanes
Independent verification
M3I Research is for informational and educational purposes only. Nothing on this site constitutes financial, investment, legal, or tax advice, an offer or solicitation to buy or sell any security, or a recommendation of any investment strategy. The scores, signals, gauges and interpretations are outputs of a proprietary research framework and should not be relied upon as the sole basis for any investment decision. References to historical patterns, valuations and market levels describe past observations and do not guarantee future results. All investing involves risk, including the possible loss of principal. M3I Research is not a registered investment adviser or broker-dealer. Always do your own research and consult a qualified, licensed financial professional before making any investment decision.