Appendix
Framework attribution, the external source register, and everything we added beyond the upstream specification — each number labeled with where it came from, or wearing its UNVERIFIED badge visibly if nothing was found.
This engine implements the public Token Compute specification: 176 equations across 8 modules, published by The Token Compute Project (publisher: Switch, tokencompute.org).
The Token Compute Project also operates a public, unauthenticated MCP endpoint at mcp.tokencompute.org — nine reference tools (model_version, search_equations, get_equation, search_claims, get_claim, get_algorithm, get_analysis, get_scenario, equation_dependencies). It is a librarian for the specification: search, retrieve, and trace dependencies. It does not compute a scenario, rank binding constraints, or issue refusals.
The spec snapshot used to build and verify this engine is dated 2026-07-23 and was harvested 2026-07-26.
Our engine implements 176 of 176 upstream equations, verified 2026-07-26 (mission B-02 closed the prior 162/176 gap).
The equation-by-equation diff against the live upstream registry is recorded in research/tokencompute-org/ENGINE-DIFF.md.
Everything beyond the 176 upstream equations — the constraint-classification layer, demand-frame analysis, and related additions — is documented in the additions register and labeled ours.
research/tokencompute-org/ENGINE-DIFF.md68 catalog parameters checked against external sources — every row kept, including 10 UNVERIFIED and 10 MODEL-INTERNAL.
View the register →24 parameters, equations, heuristics, and analyses beyond the 176 upstream equations — 2 still planned, the rest implemented and labeled ours.
View the additions →This appendix renders the same engine, registry, and narratives as every other surface — the difference is posture, never truth. Nothing here is a verdict.