LMM Candidate Decision Matrix
feeds D1 · prototype
live · debate-ready
click scores to cycle · drag weights
read-only
spec = lmm-candidate-matrix.md
6 candidates × 8 weighted axes
Which LMM vehicle do we target? Six candidates scored across eight weighted axes. Every weight and every score is live-editable — drag a weight to test a thesis, click a score to argue a cell. The composite, ranking, and pick readout recompute instantly. Use this to converge (live, in the room) on what crosses into the recommendation.
⚠
Universal caveat (load-bearing): every LMM candidate scores low on demand evidence — 0/108 YouTrack, 0 HubSpot LMM signal. Fake-door (TSK-19) is the gate for the whole field. No candidate has an evidence edge; differences here are structural, not demand-backed. Treat all composites as provisional pending fake-door signal. lmm-thesis.md
matrix
Candidates × weighted axes
click a score to cycle (shift-click = down) · drag a weight to re-test a thesis
Live leaderboard (composite · scale 0–5)
The pick (recomputed live)
What crosses to Eric (recommendation-first · M06/M14)
Weight presets — stress-test a thesis
load-bearing
Score notes — the cells worth debating
why each load-bearing score is what it is (per spec §3)
Show / hide the score-note ledger
Dotted-corner cells in the matrix carry a note — hover any score to see it as a tooltip. C4 ↔ C2 bidirectional upsell (decided): C4 stands alone as a wedge and integrates into C2 once built — mutually reinforcing, not either/or (spec §1).
Provenance:
lmm-thesis.md NotebookLM-derived ideation (market-unvalidated) ·
BTDR-0004 G2 vehicle fork = LOCKED, not pre-decided (appetite call for Eric/Thom) ·
fundamentals-sizing-brief 32–53 dev-week fundamentals (the A7 constraint) ·
Carve-out (blue) = reuse EMA modules via DevenConnect, frontend-only. Net-new IP (purple) = greenfield build (GAAP/VDR/marketplace); cost + timeline behave very differently across that line.