Primary-source data
Reference Data
Engine-computed reference grids. Each surface sweeps one deterministic engine across a two-axis input space and publishes every cell as real engine output. CI recomputes the whole grid on every build, so the page cannot drift from the math.
Kelly Fraction Reference Grid
Full-Kelly bet fraction across a 7×8 grid of win rate (0.40–0.70) by win/loss payoff ratio (0.5–5.0). 56 cells, every value computed live from the kelly-sizer engine and recomputed in CI. Bundled CSV download.
- 56 cells, win rate × win/loss ratio
- Full-Kelly fraction headline metric
- Deterministic — seeded, recompute-verified
- Downloadable CSV + Dataset JSON-LD
Options Greeks Delta Reference Grid
Black–Scholes call delta across an 8×6 grid of strike ($80–$130 on a $100 spot) by implied volatility (10%–60%). 48 cells, every value computed live from the options-greeks-explorer engine and recomputed in CI. Bundled CSV download.
Edge Variance Kelly Reference Grid
Variance-adjusted Bayesian fractional Kelly across a 6×5 grid of estimated edge mean (1%–12%) by edge-estimate uncertainty (0.5%–8% stddev). 30 cells, every value computed live from the position-sizing-edge-variance engine and recomputed in CI. Bundled CSV download.
Token Cost vs Cache Hit Reference Grid
Monthly LLM research-loop cost across a 6×5 grid of input tokens per call (2k–64k) by prompt-cache hit rate (0%–90%), on Claude Sonnet 4.6. 30 cells, every value computed live from the token-cost-optimizer engine and recomputed in CI. Bundled CSV download.
Batch vs Realtime Savings Reference Grid
Monthly savings from the batch tier vs realtime inference across a 6×5 grid of jobs per day (1k–100k) by output tokens per job (200–4000), on Claude Sonnet 4.6 at a 24h deadline. 30 cells, every value computed live from the batch-vs-realtime-cost-calculator engine and recomputed in CI. Bundled CSV download.
Agent Cost Envelope Reference Grid
Projected monthly agent cost across a 6×6 grid of steps per loop (2–16) by markets analyzed per day (5–200), on Claude Sonnet 4.6. 36 cells, every value computed live from the agent-cost-envelope-calculator engine and recomputed in CI. Bundled CSV download.
How these are built
Every reference surface follows the same rules
-
Engine-computed
Every number on a reference surface is the live output of a public /engines/{slug}.js module, never typed by hand.
-
Recompute-verified
CI re-imports the engine bundle and recomputes every cell on each build. A mismatch fails the build.
-
Deterministic by construction
Surfaces use seeded, pure engines. The same input always returns the same output, so the grid is reproducible offline.
-
Raw data shipped
Each surface includes a CSV download and a schema.org/Dataset record for independent verification.
-
Education only
Reference grids are planning aids, not financial, tax, or investment advice. No suitability claim is made for any cell.