No number is hand-entered, and none is guessed. Each one is derived by code from the filings, then checked against the filings’ own totals before it ships.
Conceptual, deliberately: the joins are the work, and their internals stay off the page.
Twenty-four datasets are extracted from the CTUIL and CEA filings: the Annex-IV(3) headroom book, the per-application and node-level annexures, the effective-grants list and its archive, the connectivity and GNA application registers, revocation lists, the granted-connectivity registers going back to late 2023, STU one-pagers, the Reg 17.1 consumption book, and CEA’s under-construction register. Extraction is scripted; nothing is retyped.
The joins are where the value sits. Developer names arrive in dozens of spellings, so 720 SPVs are resolved into 256 verified corporate groups. Substation names drift between editions, so nodes are matched across registers. Grant dates move between editions, so every archived vintage is compared grant-by-grant — that archive is where the platform’s slippage calibration (a 123-day median) comes from, measured rather than assumed.
Before anything renders, 47 automated reconciliations run: register totals against sheet totals, grant counts against stated counts, MW against MW. Today 45 pass and 2 are disclosed can’t-verify limits — stated openly, not papered over. A FAIL anywhere stops the build. The ledger is the chip in the header; click it any time.
Closure timing is a bracket, not a forecast. The floor needs no dates at all: where the published queue alone covers the remaining margin, the node is closed by queue. The ceiling assumes every approved grant achieves effectivity on its stated date — and the platform’s own archive says dates slip, so both edges of the bracket are drawn. The file-by clock subtracts the measured filed-to-grant latency (176 days median) from each node’s crossing year. Every modelled figure carries its basis next to it.