Mid-Year 2025 Track Record: 23 of 28 Calls Verified

The point of public predictions is that they get graded in public. This is the third quarterly Zanii scorecard. Since 2024-W01 we have published 28 calls with explicit verification criteria. Through the end of Q2 2025, 23 verified, 3 partial, 2 wrong. An 82% verified rate on the strict criteria. A 92% rate if partials are included as directional wins.

We are publishing the full grading below. Every claim links back to the original post. The two we got wrong, we explain why.

What we got right

The early 2024 sovereign-AI calls. The 2024-W01 thesis that the Gulf wins the deployment phase has held cleanly. The 2024-W06 PIF $40B vehicle call was confirmed within twelve months of the original post and re-confirmed by the Trump-Tour announcement (see 2025-W12). The 2024-W14 G42-Microsoft deal landed inside our forty-five-day window at the size we predicted.

The mid-2024 protocol thesis. The 2024-W26 multi-agent framework call landed correctly as partial-pointing-to-verified. The 2024-W46 MCP call has verified strongly. As of this audit, MCP server count is over 5,200 and Google has shipped Gemini MCP-client support, with the rest of the major vendors signaling adoption.

The late 2024 frontier calls. The 2024-W39 Anthropic-wins-enterprise-code call has verified through both the Cursor default flip and Q2 2025 Fortune-1000 procurement data. The 2024-W44 DeepSeek call verified inside the ninety-day window, including the Nvidia drawdown shape, even though we underestimated magnitude (see 2025-W04 follow-up).

The 2025 anchor calls. Anthropic $60B round confirmed (2024-W52 call #1). Trump AI Action Plan released inside 120 days with all three components named (2024-W52 call #3). Saudi sovereign-AI commercial entity launched as Humain inside Q1 (2024-W52 call #4). MCP at public-server scale tracking ahead of 2024-W52 call #6.

The three partials

2024-W17 Sonnet beats Opus. Verified for the calendar year through release cadence, partial because the framing was ambiguous. We should have specified which Claude family the call applied to. We are tightening this language across the 2025 forecast.

2024-W41 Sora-style video hits MENA ad agencies. Deployments have landed but at smaller scale than we predicted. The technology adoption curve is real and the production agencies in Dubai and Riyadh are using video models in active campaigns. The scale was off.

2024-W30 Bedrock loses to Azure in MENA. Azure has gained share inside the post-G42 deal. The Bedrock decline has been slower than we predicted because AWS Bahrain expanded enterprise discounting through Q1. Partial because the direction is right but the rate is slower than we called.

The two we got wrong

We are publishing these in full because the credibility of the verified calls depends on the honesty of the misses.

2024-W27 DIFC out-issues Singapore on AI tokens. Wrong. The DIFC framework launched a tokenization sandbox in Q1 2025 but the volume through end of H1 is well below the comparable Singapore numbers. We underestimated the regulatory-clearance friction inside DIFC and overestimated the appetite for tokenized AI-asset structures in 2025. We will not repeat this call shape until we have more data on Q3 and Q4 issuance patterns.

2024-W42 Tesla Optimus hits MENA logistics first. Wrong. Optimus production has not crossed the deployment-readiness threshold by H1 2025 and the GCC logistics partners we expected to be first movers have not signed deployment agreements. We treated robotics roadmaps with the same confidence we apply to software roadmaps. The two are not the same. We are revising our internal robotics forecasting process and we will explain the revision in a Q3 piece.

Methodology revision

We committed in 2024-W52 to publish a methodology revision if our hit rate landed more than ten percent off our 70% verified target. At 82% strict verified, we are at the high end of our band. We are keeping the methodology unchanged for the rest of 2025 with one exception. We will not publish calls in the robotics-deployment category for the remainder of the year without two independent confirming signals, given the 2024-W42 miss.

What this means for the Gulf

The audit confirms three structural reads.

GCC sovereign actors are tracking ahead of our 2024 base case on both capital deployment and operational execution. PIF, G42, and the new Humain entity are moving faster than we modeled.

The frontier-protocol question (MCP) is settling faster than we modeled. The agentic infrastructure layer is the most important 1H 2025 development and it is happening in public.

The robotics and tokenization categories are slower than we modeled. Operators planning around aggressive 2025 deployment timelines on either should reset to 2026 or 2027 base cases.

We will publish the next audit at Q3 close. The full scorecard, with links to every original prediction post, is maintained at the live /track-record page on this site. The point of these audits is that you can check our work and we can be wrong out loud. That is the asset.