Our one locked, falsifiable bet missed: the deal signed modified, and a day past our window — outcomes our matrix had put at zero. Then the part the May-26 read got right — clause for clause it tracked the signed text, and the single hardest clause, Iran's enriched uranium, it called a month early. We publish the miss as plainly as the match; the lock is what makes both checkable.
P-02 locked 100% on "signed as drafted, by June 16." It signed June 17 as a modified text — landing in outcomes the matrix had assigned zero probability. This is, by our own pre-commitment, Branch 2 — the post-mortem we titled in advance "the locked prediction was wrong." What follows is a self-scored, clause-by-clause read of where the May reasoning did and did not track — reported separately from the locked bet, never as a substitute for it.
Two scores, kept apart: the locked, falsifiable bet (missed) and the self-scored document read. We never let the second stand in for the first.
The "~88%" is a document/clause match (9 full + 3 partial of 12 rows, partials at half), explicitly NOT the locked outcome-class probability — which is reported here as a miss. The two numbers measure two different things, and the brief keeps them apart on purpose.
RAOSCAFF locked prediction P-02 on the US–Iran "14-point Memorandum of Understanding" on 2026-05-26 at 20:58 IST and published it the same evening at 21:38 IST (commit 2ca4d9b on origin/main) — before the agreed text was public. The brief has been live, timestamped, and unedited since.
We forecast the deal would form on one spine — end of war · Strait of Hormuz reopening · a 60-day negotiating window · Pakistan-mediated — and flagged the two clauses most likely to fracture it: the US naval-blockade snap-back and HEU (highly-enriched-uranium) removal, ranked #1 most likely to slip (reworded or deferred).
The 14-point MoU was signed electronically by Presidents Trump and Pezeshkian (Trump also initialled a physical copy); the US military (CENTCOM) said it lifted the naval blockade on June 18; the MoU opens a 60-day window for the final deal. (The first round of US-Iran implementation talks scheduled for Jun 19 at Bürgenstock, Switzerland was postponed (Swiss Foreign Ministry; NBC/CBS/CNBC, Jun 19) — VP Vance cancelled his trip and Tehran sought Lebanon-ceasefire guarantees amid fresh Israeli strikes. The deal's first setback; it does not unwind the signed MoU or the 60-day window.) Published terms (full text carried by NPR/CNN/CBS):
| What P-02 locked (May 26) | The signed MoU (Jun 17) | Match |
|---|---|---|
| Deal signs; war ends | Pt 1: permanent end of operations, incl. Lebanon | ✅ |
| Strait of Hormuz reopens | Pt 5: safe passage begins immediately | ✅ |
| Phased / gradual easing | Pt 4: blockade fully removed within 30 days; Hormuz toll-free 60-day window (not permanent) | ◑ reopening hit, but the 30-day clock attaches to the blockade lift (Pt 4), not Hormuz; Hormuz is a 60-day toll-free window, not permanent — direction right, terms diverge |
| 60-day negotiating window | Pt 3: final deal in max 60 days | ✅ |
| Pakistan-mediated | Pakistan + Qatar mediated; not named in text | ◑ real, not in-text |
| Enrichment moratorium unsettled | Deferred to the 60-day talks | ✅ |
| Sanctions relief conditional | Oil waivers immediate; full relief deferred | ✅ |
| Frozen-asset release flagged; no figure locked | Pt 11: freed on implementation, no figure | ✅ no figure either side |
| No-nuclear-weapon pledge | Pt 8: "shall not develop nuclear weapons" | ✅ |
| IAEA inspections unsettled | IAEA supervises down-blending; full regime deferred | ◑ partial |
| Naval-blockade snap-back = friction | Blockade lifted, no snap-back in text | ✅ friction called |
| HEU removal = #1 clause to slip (reworded or deferred) | Down-blended on-site under IAEA; disposition deferred | ✅✅ bullseye |
Each signed-term claim above traces to the Tier-1 sources named in §Sources: the full 14-point MoU text carried by NPR (Jun 18), CNN (Jun 17), CBS News, and Military Times; the signing and blockade-lift by NBC, CNBC (Jun 18), Al Jazeera (Jun 17–18), The Hill, and NPR.
A month before the document was public, we named HEU removal as the clause most likely to slip — to be reworded or deferred — and wrote it would most likely be "reworded as supervised storage rather than export, or deferred."
Signed Point 8: the uranium is down-blended on site under the IAEA (the minimum methodology), with final disposition deferred to the 60-day talks. We named the family — supervised or deferred rather than removal — and the deferral landed exactly; the specific method (on-site down-blending under IAEA) was within that family, not one we named by name. On removal-vs-not, the decisive axis, we were right. This was our secondary modal scorecard's #1 pick — and it was right. The primary lock, by our own Branch 2 pre-commitment, was not.
We flagged HEU removal a month early as the one most likely to slip — to be reworded or deferred. The signed text down-blends it on site, disposition deferred. The approach we named — on removal-vs-not, the decisive axis, we were right.
We said (May 26): "HEU removal — #1 most likely to slip (reworded or deferred)," most likely reworded as supervised storage rather than export, or deferred. · They signed (Jun 17, Pt 8): down-blended on site under IAEA, final disposition deferred to the talks. · This was our secondary modal scorecard's #1 pick. The primary lock, by our own Branch 2 pre-commitment, was not.
P-02 came from RAOSCAFF's position-desk simulation — locked and timestamped before the event, then scored against the published terms. We publish the primary miss and the secondary match. That is the model.
The lock is what makes the scoreboard checkable. Without the cryptographic timestamp, any prediction-series methodology could claim retroactive accuracy by editing the matrix after the fact. P-02 was locked in writing on 2026-05-26 at 20:58 IST and published the same evening at 21:38 IST (commit 2ca4d9b on origin/main) — before the agreed text was public — and this resolution scores that unedited lock against the signed MoU of 2026-06-17.
2ca4d9b on origin/main · before the agreed text was public/intel/p-02-us-iran-mou-memorandum-decode/report.html · commit 2ca4d9b on origin/main
The locked matrix cannot be modified after the original commit landed on origin/main. This resolution publishes as a separate document, citing the original brief by URL. The compounding asset is the doctrine, the Position Desk library, and the lock-and-score cadence — bounded by the world's disclosures, never by RAOSCAFF's.
P-02 came from RAOSCAFF's position-desk simulation — locked and timestamped before the event, then scored against the published terms. The original 8-Position-Desk orchestrated panel, the clause-by-clause matrix, and the structural caveats are preserved unedited in the original brief. This resolution does not revise the lock; it scores it.
We report two distinct scores because they measure two distinct things. The primary, locked probability bet (100% on "signed as drafted, by June 16") resolves against the literal outcome class — and it missed: signed June 17 as a modified text, both outcomes our matrix had assigned zero probability. By our own pre-commitment, this is Branch 2. The document match (~88%) measures, clause for clause, how the May-26 read tracks the signed terms (9 full + 3 partial of 12 rows, partials at half = 10.5/12 = 87.5%). The "~88%" is a document/clause match, explicitly NOT the locked outcome-class probability. We keep them apart on purpose.
The "We said / they signed" table places each locked spine item and flagged clause against the corresponding signed point. Full matches (✅) cover the deal itself, end-of-war + Lebanon, Hormuz reopening, the 60-day window, the enrichment-moratorium deferral, conditional sanctions relief, frozen-asset release (no figure locked; no figure in text — a match), the no-nuclear-weapon pledge, and the snap-back-as-friction call. Partial matches (◑) cover the phased-easing terms (direction right, Hormuz/blockade clock structure diverges), the Pakistan-mediator role (real, not in-text), and IAEA inspections (partial). The HEU-removal clause is scored a bullseye (✅✅, counted as 1 full): the supervised-or-deferred family we named is the approach that was signed. Match arithmetic: 9 full + 3 partial of 12 rows, partials at half = 10.5/12 = ~88%.
One divergence is recorded in the honesty ledger: we placed Pakistan's role as central where it was true in the room but not written into the text. The frozen-asset figure is not a divergence — we flagged frozen-asset release without locking a figure, and the signed text (Pt 11) is figure-less too; this is a match. The primary outcome-class miss is recorded as a miss: signed June 17 as a modified text, both outcomes the locked matrix had assigned zero probability.
Each signed-term claim carries a Tier-1 source named in §Sources. The full 14-point MoU text is carried by NPR (Jun 18), CNN (Jun 17), CBS News, and Military Times; the signing and blockade lift by NBC, CNBC (Jun 18), Al Jazeera (Jun 17–18), The Hill, and NPR; the Switzerland implementation-talks postponement (Swiss FM; NBC/CBS/CNBC, Jun 19). The locked call traces to the original P-02 brief (locked 2026-05-26).