FORGE Operational Report: 143 Signals, 15 Opportunities, 8 Products in 21 Days

One of the most surprising findings from this report is that 97.2% of signals came from PyTrends.

The FORGE system collects data through signals from PyTrends and CoinGecko, which are then used to detect opportunities. These opportunities are then evaluated by the DebateOrchestrator, and products are generated based on the outcome. This report provides a snapshot of the system's operational metrics as of 2026-06-06.

Signal Collection: Which Sources Actually Produced

PyTrends dominated signal collection with 97.2% of signals, while CoinGecko accounted for only 2.8%. The top categories for signals were crypto and ai_tools, which each had an average score of 0.727 and 0.68, respectively.

Opportunity Detection: From Signal to Candidate

Of the total 143 signals, 15 opportunities were detected, with only 2 being published as products. The crypto category had the highest average score for opportunities at 0.586, while ai_tools had an average score of 0.573.

Content Quality Gate: What the Debate Saw

The DebateOrchestrator evaluated a total of 28 signals, with 1 outcome passing and 9 outcomes being rejected. The revised outcomes accounted for 18 signals, with an average final score of 2.324.

Model Routing: Real Cost and Latency

The Model Router made a total of 8 LLM calls, with no cost logged. Both the gemini-1.5-flash and gemini-2.5-flash models had zero latency and cost.

What Is NOT Yet Measured

As of this report's snapshot date, sales are at 0, reward events are also at 0, and SEO posts total 27.

Conclusion

This report marks the first in a series of operational reports for FORGE. As this is a small sample size, we will compare these numbers to a later snapshot to track progress. The next report will be generated at [unmeasured].