The system's capability boundaries: ten honesty clauses written into the product
This system is best at recognition, not prediction. Ten capability boundaries from engine research and measurement are written into the product. Here they are, one by one, unvarnished.
The one-line conclusion: the system is a detector, not a predictor. The ten boundaries below are part of its genetics, and no page or article will ever paper over them.
The ten boundaries, one by one
- A detector, not a predictor. Long-range forecasts like "which coin will pump within 60 days" have been falsified by two rounds of blind testing. That is why "60-day rally probability of similar coins" is always labeled experimental and always says "historical base rate of similar coins, not a prediction for this one".
- Ignition confirmation lags. The measured median delay is about 11 hours, and by confirmation time price had already risen a median of +27.4% (historical sample medians, not per-event promises). So ignition cards must show "already up by confirmation", and the near-edge layer must say "eyes early, engine unconfirmed". See Ignition and dual-track confirmation.
- Slow stair-step grinds may never trigger ignition. A coin that climbs in steps without abnormal volume (the ARB case in our research) may never satisfy either track. The checkup report's ignition section flags this blind spot.
- The dump-risk score is a 0–100 signature score, not a probability. The product always says "points", never "% probability". See The five dimensions of the distribution signature.
- Stage duration has no predictive power. Stage history is a historical description: it records how long each act lasted, but it never counts down "days left in this act".
- Jump-style C1 ignitions give no near-edge lead time. Violent launches often happen in one step, faster than the warning layer — the near-edge disclaimer says so.
- Hyperliquid-sourced coins have short daily K-line history (about 419 bars at most), and the HL ignition book's record is poor. Reports on HL-sourced coins are labeled "shorter history, one notch lower confidence".
- The AI juror has zero discriminative power. So the product uses no LLM for any judgment; a model may only polish wording under template constraints. Every number in a report comes from rules and data, never from generation.
- Paper books exclude part of funding and borrow costs. Paper results therefore differ from live methodology and are not directly comparable; the signals and paper pages carry this in their methodology bars.
- The Kelly criterion for the leverage books is currently negative — the mathematical answer is "do not size up". Hence live trading defaults to 1–3x leverage, mandatory exchange-side stop-losses, and leverage amplification off by default.
Why publish the shortcomings out loud
The industry's habit is to hide the shortcomings and show the screenshots. Our bet is the opposite: in a market full of exaggeration, verifiable honesty is itself the product. These ten boundaries appear on the relevant interfaces as info hints or methodology bars; the scorecard and proof verification (How to verify the scorecard and proofs) are their evidence layer.
If you remember only one line, make it the first: what we sell is a tool for seeing the present clearly, not a service that claims to see the future.