$$ % Define your custom commands here \newcommand{\bmat}[1]{\begin{bmatrix}#1\end{bmatrix}} \newcommand{\E}{\mathbb{E}} \newcommand{\P}{\mathbb{P}} \newcommand{\S}{\mathbb{S}} \newcommand{\R}{\mathbb{R}} \newcommand{\S}{\mathbb{S}} \newcommand{\norm}[2]{\|{#1}\|_{{}_{#2}}} \newcommand{\pd}[2]{\frac{\partial #1}{\partial #2}} \newcommand{\pdd}[2]{\frac{\partial^2 #1}{\partial #2^2}} \newcommand{\vectornorm}[1]{\left|\left|#1\right|\right|} \newcommand{\abs}[1]{\left|{#1}\right|} \newcommand{\mbf}[1]{\mathbf{#1}} \newcommand{\mc}[1]{\mathcal{#1}} \newcommand{\bm}[1]{\boldsymbol{#1}} \newcommand{\nicefrac}[2]{{}^{#1}\!/_{\!#2}} \newcommand{\argmin}{\operatorname*{arg\,min}} \newcommand{\argmax}{\operatorname*{arg\,max}} \newcommand{\dd}{\operatorname{d}\!} $$

Planar Build

The critical path, one step at a time: what each step built, the evidence that it works, and the decisions still open.

The code walkthrough studies cmg as Angelo left it on phase1/dev. This section follows the work that builds on it: a planar (x–z) version of the box-over-a-step problem, on the branch phase1/planar of Symplectomorphism/c-gnn. Each page reports one step of the critical path (Decision 5 of the discussion compendium). It gives the results, figures you can inspect, and the choices that still need a decision. Pages are updated when a decision changes.

Step	What it delivers	Status
1. The planar contact layer	Planar scene, principal-contact on-set (one bit per body pair), contact sets, hold test on contact sets	Done (2026-11-05); all refinements settled; known issue: ties under deep penetration (step 2)
2. Valid, transfers and a certified hold gate	Valid, transfer and pass-through in Cand; H by persistent contacts, decided exactly with certificates and margins; hold-solver benchmark; tied poses rejected	Done (2026-11-05); four choices for review; εpen moved to step 3
3. The local controller’s cost	The compiled cost of the paper against the current hand-tuned MPPI cost; success criteria; εpen	Next
4. The ranker	Attention over contact templates with a goal direction; LTS/PHS search	Planned

Background tutorials explain a method used in a step, from first principles:

• Deciding support exactly (step 2): how the hold test turns “can these contacts hold the box?” into a cone in three dimensions, and answers with a witness, a certificate and a margin.
• The hold test as a linear program: the same question as LPs that serve the planar and the 3D builds alike, with force limits and margins against a set of disturbances.

Known issues lists every open defect we know of in the code and the paper, with where it is documented and which step fixes it.