# METHODOLOGY ADDENDUM — Stage 2

Extends `stage1/report/METHODOLOGY.md`. Same core discipline (validated ligand series `stage1/ligands_series/`; canonical within-complex MDE estimator `stage1/dock_setup/mde_estimator.py`, imported verbatim; primary-source Kd oracle `stage1/dock_ctrl/kd_primary_source.json`; Signal A/B; 720-permutation multiple-comparison guard; Kd read-only/post-hoc; honest-negative-is-a-pass; fine-ordering-never-claimed; outer-sphere rigid-body docking with Ir→Fe surrogate + xTB ligand charges). Numbering canonical Q3V6T2 (paper = canonical−1); Gαi3 = P08753.

## Protocol A — rigid real-cavity competitive-block docking (item 1)
1. **Receptor:** experimental PDB 6MHF chain A (Gαi3), GDP retained (distal, stabilizes Switch-II), Girdin chain C + solvent stripped. Box on the Switch-II/α3 GBA-binding groove (center ≈[-31.6,-18.9,-9.8], 22.5 Å cube = authors' AD4 grid volume), same volume on all compared surfaces.
2. **Dock** the six complexes (≥3 seeds) into the groove; compute the within-complex MDE on this rigid cavity and CONTRAST with the Stage-1 IDR band. **MDE-collapse guard:** a rigid single-conformer cavity collapses the seed-reproducibility floor far below the ~1 kcal/mol physical scoring resolution → do NOT read a sub-0.1 kcal/mol gap as signal; the label-permutation null is the operative guard.
3. **Pocket-specificity / size-confound control (decisive):** dock the same six into (i) the Girdin-GBA-helix control receptor (SLiM, no cavity) and (ii) a decoy/off-groove box on the same Gαi3 surface. A positive is claimable as pocket-specific ONLY if the signal is absent on both controls; else it is a non-pocket-specific ligand-size effect (the Stage-1 SH2-pocket confound).
4. **Cross-surface labeling (mandatory):** the Kd series is complex↔Girdin, not ↔Gαi3 — every Gαi3-vs-Kd correlation is labeled cross-surface; the same-surface 1a→Gαi3 15 µM is a single point (no series).
5. **Competitive-block read-out:** does the best pose occupy the GBA-binding groove residues (would it block Girdin)?

## Protocol B — phospho apo/phospho + charge-sensitive rescoring (item 4)
1. **Ensembles:** phospho region reuses the Stage-1 members spanning ~1705–1755; the extreme C-terminus 1815–1871 is a fresh build (AF-full-length slice + ANM/MD collapse to compact, box-tractable variants — a docking-tractability device, not a stable-fold claim). Apo + phospho forms; UniProt marks pS1717/pS1820/pS1837, plus pY1744 (paper-only, labeled secondary). SEP/PTR grafted as the physiological dianion (−2).
2. **Blind SAR docking (apo):** same as Stage-1 VAL-BLIND (whole-surface, cluster→per-site MDE-gated Signal A/B + permutation guard).
3. **Charge-sensitive electrostatic test:** because smina/Vina scoring is charge-INDEPENDENT (apo and phospho dock identically), the phospho effect is invisible to the docking score. Probe it with **GEOMETRY-MATCHED** apo-vs-phospho GFN2-xTB ΔΔE_int: identical docked geometry (0.0 Å shared-atom deviation), phosphate/charge toggled, so ΔΔE isolates charge (not a geometry artifact). Report the DIRECTION (attraction) + its docking-blindness; the gas-phase magnitude is an unscreened upper bound (aqueous screening ~1–2 orders lower) and is NOT claimed; a buried-dianion truncated-pocket ALPB(water) number is unreliable and reported only as a sensitivity probe.

## Why more correct than the source paper
The authors placed 1a in a single rigid homology "SH2" pocket and never tested it against the measured Kd series. This methodology (i) never assumes a fold, (ii) searches blindly, (iii) makes every site stand or fall against the fixed experimental SAR under a noise-gated, multiple-comparison-guarded, size-confound-controlled test, (iv) adds a rigid-real-cavity control to distinguish "SAR below resolution" from "IDR-noise-swamped", and (v) uses a charge-sensitive method where the docking score is structurally blind (electrostatic/phospho). A rigorous negative is a valid, falsifiable result.

## Limitations
Cross-surface oracle mismatch (item 1); single-point same-surface Gαi3 anchor; unscreened phospho magnitude (direction only); 1815–1871 low-pLDDT (compact variants are a sampling device); pY1744 paper-only; all binding modes are computational hypotheses (no experimental Ir+Girdin/Gαi3 co-structure). The phospho-xTB (VAL-PHOS-003) analysis is the worker's committed on-disk output; its Zenith finalization was blocked by an org-budget limit and the synthesis+publish were completed by the orchestrator.