# Reproduce the headline findings

Unzip the handoff and run from its top directory with Python 3:

```sh
python3 research/verify_accounting.py
python3 research/alternatives_2m/independent_verifier.py
python3 research/continuation_5m/binary7_independent_verifier.py
```

The first command recalculates all ten accounting rows and the token sums. The other two replay the K3-derived and binary-matrix contradictions with the original independent verifiers and retained inputs. They write their verification results beside the source files. No external packages or paid services are required.

The reports in `data/` describe the broader original campaign. Their relative links refer to its original workspace. This compact handoff includes runnable code for the two headline exclusions; it does not contain every historical columnar solver or raw certificate. The complete finite-ledger summaries are included so its arithmetic can be reproduced independently.

## Counting units

The finite ledger totals model-labelled parameter assignments. Each family label remains part of the identity, so the sum includes overlapping cipher functions. It is not the number of unique plaintexts, plausible English solutions, or a percentage of all K4 methods. The column-reset extension to all positive key lengths has no finite labelled total and is omitted. Replications, capped results, and the surviving width-31 model are also omitted.

The K3-derived exact count is 4,524,029,526,974,464. An earlier conversation printed 4,524,050,360,758,272 beside the correct formula; the website corrects that arithmetic.

## Token provenance

Six distinct completed native goals total 8,403,684 tokens. The original interrupted goal has a last observed checkpoint of 366,344 tokens, recorded 14 September 2026 at 04:37:50 UTC. Its final usage was not captured. Total documented usage is therefore at least 8,770,028 tokens. The main campaign completion records 3,417,704 tokens at 09:35:50 UTC. The five other completion records are in `usage/`. These records were reconciled against the original research task, with each goal counted once by its creation timestamp. Planning, unbudgeted follow-up and website creation are outside this count.
