The tracker, and how to read it
The Capability & Cost Tracker is the standing evidence board behind every Systemic Script issue. It tracks the markers that actually move engineering decisions across AI systems, quantum, biotech, and compute hardware — with the number that defines each and its status. Figures are vendor- or lab-stated unless marked independently reproduced; we label which is which. Status moves are flagged in each Thursday issue's Scoreboard.
AI systems — capability markers
| Marker | What it measures | How to read it | Status |
| Frontier model releases | Architecture + scale of flagship systems | Mechanism over marketing: routing, context, training method | Continuous |
| Benchmark suites (MMLU-class, SWE-bench-class) | General capability / task performance | Only with methodology; watch saturation and contamination | Continuous |
| Open-weight ecosystem | Capability available for self-hosting | Gap vs frontier in months; quantization viability | Continuous |
AI economics — cost markers
| Marker | Why it matters | Status |
| API inference $/M tokens | The build-vs-buy input for every AI feature | Falling steeply across 2024–26 |
| Self-host break-even | Volume at which owned GPUs beat API pricing | Rising bar as API prices fall |
| Training compute of frontier runs | Sets the capital floor for frontier participation | Rising |
Quantum — hardware markers
| Marker | What it measures | How to read it | Status |
| Physical qubit counts | Raw hardware scale | Meaningless without error rates — read together | Continuous |
| Logical qubits / error-correction overhead | Usable compute after correction | The number that matters; overhead ratio is the story | Early but real |
| Two-qubit gate error rates | Operation fidelity | Sub-0.1% is the threshold territory that changes the math | Continuous |
Biotech — cost & capability markers
| Marker | Why it matters | Status |
| Sequencing cost per genome | The curve that built modern genomics | Long tail below the $100 headline |
| DNA synthesis cost per base | The write-side counterpart — enables engineering biology | Falling, slower than sequencing fell |
| Editing precision/delivery methods | What can actually be edited in practice | Delivery is the bottleneck, not the editor |
Independent technical analysis — not investment advice or engineering counsel.