Alternative NIP — formulation & label modes

Mode B: a known recipe → true protein (near-exact). Mode A: a commercial label → allocated broad estimate (v0). Both run the same engine: sequence-derived k, NPN-corrected, with reverse checks.

Formulation mode

One ingredient per line: name | amount% | protein% (optional) | k (optional). Protein% and k override the database defaults (e.g. a supplier spec).

Label mode

Paste the ingredient statement (bracket-aware: compound ingredients, percentages, "less than 2% of:", "may contain" advisories are parsed and excluded) and the declared protein. Known percentages become hard constraints; the rest is allocated by category priors or order decay. Ranges stay broad — priors are FDC typicals, not your batch.