Method zoo: 36 per-method cards (M1–M14) + key references

[GiggleLiu]

Method zoo — the method-side parallel of the model zoo (#58)

Adds a method zoo: 36 per-method cards under .knowledge/methods/<name>/METHOD.md, one per distinct method in method-survey.md, each with a structured M1–M14 property table, cost & accuracy classes, recommended models, one key reference, and verified benchmarks.

Stacked on #58 (model-zoo). Based on model-zoo because the method cards cross-link the model cards and the prerequisite knowledge files committed there, and main doesn't yet have them. Retarget to main once #58 merges. The diff here is only the method-zoo work.

What's new

• .knowledge/method-property-checklist.md — the new M1–M14 axis schema (M-I Capability, M-II Applicability gates keyed to model A1–D16, M-III Cost, M-IV Limits), the method-side companion to model-property-checklist.md.
• .knowledge/methods/_TEMPLATE.md — the card template.
• 36 cards in 7 families: Exact (ed-full, ed-lanczos, ftlm-tpq, kpm, fci) · Classical-MC/RG (metropolis-mc, cluster-mc, mcrg) · Ground-state TN (dmrg, peps-ipeps, trg-hotrg, mera) · TN dynamics/finite-T (tebd, tdvp, mps-finite-t, ltrg-xtrg, dmrg-x) · Quantum MC (vmc-nqs, sse, dqmc, afqmc-cpmc, dmc-gfmc, pimc, diagmc) · Embedding/perturbative (dmft, dmet, nrg, frg, coupled-cluster, mbpt-gw) · Mean-field/circuit/certified (hartree-fock, spin-wave, slave-particle, vca-cluster-mf, circuit-sim, sdp-nctssos).
• .knowledge/methods/INDEX.md + shared ref.bib (55 entries).

References — discover-first

Each card's key reference was reused from the existing .knowledge/literature/ tree where present (20 cards: cite keys copied verbatim, full text linked in place) and downloaded only when absent (the rest). 4 honest bib stubs for paywalled/pre-arXiv classics (fci, cluster-mc, coupled-cluster, slave-particle). The pre-existing literature/ tree is untouched.

Quality gates

• Each of the 7 waves passed an independent physics-accuracy review (subagent) + fixes; a final whole-zoo cross-card consistency review (APPROVED) checked every M9 cost against the survey master table, taxonomy consistency, and cross-card agreement.
• Reviewers caught real errors that were fixed: ED dense-vs-Lanczos 48–50-site frontier attribution, AFQMC constrained-path bound direction (E_CP ≥ E_0), fRG Γ⁽⁴⁾ vertex, DMFT Mott U_c2/D units, MERA S=(c/3)lnL prefactor, HOTRG O(χ^{4d−1}) (not ATRG), stabilizer-rank α≈0.396, TEBD quench entanglement linear growth, plus 3 wrong arXiv IDs caught and corrected.

Test plan

• 36 cards, each with all 14 axes M1–M14
• every [@key] resolves in methods/ref.bib; no duplicate keys
• every Rendered: path exists (downloaded next to card, or linked into literature/)
• no .raw//.figures/ blobs staged
• literature/ tree unmodified (download-ref collateral damage reverted)
• all 7 per-wave physics reviews + final cross-card review APPROVED

🤖 Generated with Claude Code