Methodology
The Robot Ledger exists to be citable. That requires rules stricter than journalism's and disclosure cleaner than a vendor's. These are the rules; they apply to every figure we publish.
Observed Only
No estimates, no interpolation, no figure from memory. Absence is reported, never filled.
Receipts
Every figure traces to a dated, checksummed archive entry captured from public sources.
Corrections in Public
Prints are never edited. Corrections run in the next print with both values preserved.
Observed prices only
Every stored figure comes from an observed source with a document reference and a capture timestamp. Unobserved is unobserved: we never estimate, interpolate, or infer a price. When a price disappears from public view, we report the last observed figure with its date — the tractor that went dark in 2025-01 is reported exactly that way, not modeled around.
Non-price absence is data too. The price-visibility tracker records, category by category, whether public prices exist at all — and every time one appears or vanishes.
An append-only, checksummed archive
Every observation is captured from archived public sources and stored with a checksum at the moment of capture. Raw archive entries are never modified or deleted; corrections are new entries. This is what lets us say every figure has a receipt — the archive predates the claim, and the claim traces back to it.
Versioned index bases
Frontier lines are computed against versioned unit-of-work definitions. The v1 basis is per-unit entry price: the lowest observed price of a shipping unit in each category. When the normalization basis improves — for example, capability-normalized lines once both price and specification series are observable for the same units — history is restated under the new version, openly. Never silently.
Shipping units only
Machines enter the index when they ship at an observable price — not when they are announced, demoed, or crowdfunded. The announced-vs-shipping watch tracks headline programs that remain announcements, with evidence dates.
Corrections in public
Published prints are never edited. When a figure needs correcting, the correction runs in the next print with both values preserved — the record shows what we published and when we fixed it.
No forecasts
This publication reports observed market state only. No fitted curves, no residual predictions, no speculation dressed as data. Where an adjacent market offers an observed preview (as used-GPU asking prices do for fast-frontier depreciation), we publish the observations and their dates — and stop there.