nishi code wiki / research / food science, chemistry and peptides
Food science, chemistry and peptide analysis: the lane and its measured blind spot
Built and measured 2026-07-23 (waves 1–2, 82/82 gate assertions GREEN); corpus brief 124,685B; transcribed 2026-08-01. Corpus slug: project-nishi-food-chem-peptide-supplement-2026-07-23. Public companion surface: nishifamily.com/stability. Numbers here are OURS (computed by our organs) anchored against published literature values.
The finding. The existing adulterant screen matched observed peaks against [M+H]⁺ only — but peptides electrospray onto a CHARGE LADDER. BPC-157 (1418.70 Da) has its [M+H]⁺ at m/z 1419.7, outside a typical 100–1000 screen window; its real signals are [M+2H]²⁺ 710.4 and [M+3H]³⁺ 473.9. So a single-charge screen reports CLEAN on a spiked sample. Made mechanical rather than rhetorical:
sup_single_charge_blind_spot() shows 3 of 5 catalog entries invisible — and a control proves the finding is specific, not blanket (TB-500 and GHRP-6 are NOT blind). Filed as debt against the screen with the fix named: iterate z=1..4 in the ppm match loop.Forks off — the lineage
| parent | inherited |
|---|---|
| the 67 pre-existing chem/food organs (tree census, not a corpus search) | ★The method law this lane produced: a corpus search for “ice cream formulation” returned pure ranking noise, which reads as “nothing here” — while a tree glob found 67 organs. Never conclude a capability is absent from a corpus search alone; glob the tree. (The same law later saved a duplicate build in the evidence lane.) |
| research_codequality · sibling | Non-vacuity and negative-control discipline — applied here as literature anchors that cannot pass by accident. |
What the lane measured (each number computed, then anchored)
| organ | the measurement that proves it |
|---|---|
| nx_peptide (18/18) | Sequence→mass, [M+nH]ⁿ⁺, b/y ladder, pI by bisection, GRAVY. Angiotensin II 1045.5345 Da EXACT vs literature; bradykinin off by 1 Q4 unit (0.0001 Da of accumulated rounding) — and the tolerance budget is a testable claim, not a promise. No float, no pow: a 10-entry decade table ⇒ pI good to ~±0.05 pH, stated in the header, not hidden. T17 MEASURES the bisection's precondition (charge monotone over pH 0–14) instead of assuming it. |
| nx_icecream (20/20) | The whole file hangs off one derivation: frozen_fraction = 1 − FPD/|T|. Validated vs published freezing curves: 506‰ frozen at −5°C (lit ~500), 810‰ at −13°C (lit ~800), 863‰ at −18°C (lit ~900) — and the ~4-point deviation in the deep tail (where ideal-solution theory stops holding) is STATED. PAC is derivable (molar-mass ratio) so it is computed; POD is psychophysical so it stays a cited table — confusing them is the classic formulation error. A gate found a real bug: integer truncation lost a whole PAC unit on dextrose; fixed to round-to-nearest, after which salt reads 586 not the trade table's 585 — the discrepancy is in the TABLE, not the code, documented rather than matched. |
| nx_supplement_screen (13/13) | The blind-spot finding above, plus label-claim dose verification in BOTH directions (superpotent is a finding too). Catalog masses are DERIVED from sequence, never tabulated. Terminal modifications modelled (C-term amide −0.98402, N-term acetyl +42.01056): hunting the free-acid ion for an amidated peptide misses by ~1130 ppm — a guaranteed miss at a 20 ppm window. |
| nx_peptide_ext (13/13) | Non-standard residues incl. Hyp (the collagen marker), disulfide and lactam bridges. Double literature anchor that cannot pass by accident: oxytocin 1006.4366 vs lit 1006.4367 and vasopressin 1083.4379 vs lit 1083.4380 — each lands ONLY with amide AND disulfide applied, with no-amide and no-disulfide run as NEGATIVE CONTROLS. Encodes the near-isobar trap: Leu/Hyp differ by 0.0364 Da (same nominal peak), and pex_resolving_power_needed returns 0 for Leu/Nle — truly identical, no resolution separates them. |
| nx_pow10 (shared primitive) | Peptide ionisation and thermal-death kinetics are the same decade-log physics, so the table was extracted and interpolated with a SIGNED error: 10ⁿ is convex so a chord lies above ⇒ negative exponents come out under-stated = the safe direction, stated in the header so callers carry margin. Nearest-entry lookup had been up to 12% error — far too coarse for a lethality calculation. |
Declared UNVERIFIED / caveats
| 1 | Freezing-curve agreement degrades in the deep tail (−18°C: ours 863‰ vs lit ~900‰) where ideal-solution theory fails; the deviation is documented, not corrected away. |
| 2 | pI precision is ~±0.05 pH by construction (0.1-log table granularity), which is adequate for ordering and inadequate for fine electrophoretic prediction. |
| 3 | Literature reference values are cited from the corpus brief and were not re-fetched at transcription; the NIST SRDs named there (SRD 1c peptide MS libraries, SRD 69 WebBook, SRM 2378, SRM 3532) are the external anchors. |
| 4 | This is analytical chemistry tooling, not clinical guidance. Nothing here is a health claim, a dosage recommendation, or a substitute for accredited laboratory testing. |
| 5 | Individual literature citations are not yet archived per rule A; queue on next touch. |