Flagship dossier · No. III
The Evidence Ledger — environmental accountability, from orbit
Alberta's oil and gas footprint is one of the most surveyed landscapes on Earth — and almost none of that surveying is public. The Ledger watches it from orbit: 99 assessed sites across 13 licensees, each with a machine-readable evidence packet. And its best feature is a mistake it caught in itself.
The first version of the assessor measured change with seasonal medians — and reported dramatic "recovery" at site after site. An audit caught the flaw: the medians were comparing May/June greenness against July brownness. The instrument was measuring the calendar, not the land. The assessor was rebuilt around month-matched median-of-deltas — each month compared only against the same month in other years — and the phantom recoveries vanished. The arc is the point: a system that catches its own false positives in public is worth more than a system that never admits one.
Comparing a May image against a July baseline reads spring green-up as "improvement." Twenty-seven sites of phantom recovery.
A review of the inputs showed the month mismatch. The method's own falsifier — re-run with matched months — killed the finding.
Each month is compared only against the same month across years, requiring at least two matched months — and the assessment refuses to run otherwise.
Four stages, each with stated limits. Detection is allowed; identification is allowed; attribution to a cause is not — the Ledger says what changed, never why, and never who is to blame. That restraint is structural, not stylistic.
21,892 geocoded sites across Alberta, located to approximately ±300 m DLS centroids with 500 m assessment buffers. The registry is the population; everything downstream samples from it.
Sentinel-2, 10 m pixels. Freely available, revisit every few days, honest resolution. Ten metres cannot resolve a wellhead — the Ledger never claims to see one.
NDVI/NDMI change signals, month-matched. 99 sites assessed across 13 licensees: 28 with an identified change signal, 71 with no significant change, 3 high-confidence identifications. Below four matched months, no confidence interval is printed at all.
Machine-readable evidence packets, public dashboard. Every assessment ships as a packet — inputs, method version, matched months, signal, confidence — rendered on a live public dashboard anyone can audit.
The current recorded state: 99 assessed sites across 13 licensees, 28 identified change signals, 71 with no significant change, 3 high-confidence identifications, 12 sites where no confidence interval is printed. The dashboard is live and public.