Roadmap: what is next, in the order it unblocks other work.
Ordered by what unblocks the most, not by size. Every item names the work that owns it, so this page cannot drift away from the backlog.
In order of what it unblocks
tracked in team/TEAM-ROADMAP.md
NEXT / 01
A public home for the benchmark corpus
The corpus is packaged so that it can be cited: stable identifiers, stated integer units, a licence and a valid citation, byte-identical on rerun. What it does not have is an address someone else can cite from. Publishing it sits with the launch content and waits on the planned release waves, because an announcement lands once and should land after the thing it announces.
LATER / 02
Beyond convex and compressible
Convex hulls and compressible items ship in all four engines. What is still not expressible is a bag that changes shape, the other eighteen orientations a hull has, and hull coordinates past 6.25 m - each a contract change rather than an engine fix.
LATER / 03
One quality level everywhere
Rust and the JavaScript fallback are held to validity and an objective floor, not to the reference engines' quality. Closing that gap matters more than any new capability.
HOW WORK LANDS
How a request field becomes a supported capability.
The sequence exists so the gate can see the change. Skipping a step does not make the work smaller, it makes it unverifiable.
01Add the field to the request schema
02Reject it explicitly in every engine that does not implement it
03Implement it, one engine per change
04Recompute it independently in the validator
05Add the fixture and its field-matrix row
06Document it, then move the row to supported
STATUS
Status: 1.11, contract frozen at 1.0.
The suite is at 1.11, and every change it made to the contract was additive: a request that follows the 1.0 schema gets the answer it got before, unless that answer was a bug fixed since, and breaking that costs a major version. Two exceptions are named rather than hidden: a configuration key the schema never declared, once ignored, is refused from 1.5, and from 1.6 an item left out because the inventory ran out or an axle would be overloaded carries that reason instead of no_feasible_placement or search_exhausted; from 1.7 an output unit of ticks, which the schema never listed, is refused; and from 1.8 keep_upright, stackable, must_be_on_floor and require_placement_coordinates must be booleans, and optimal means the answer reaches the engine's lower bound rather than that the search ran out; from 1.9 a side of zero, a stray obstacle and a size or weight past the admission ceiling are refused, and the JavaScript packages return a weight past 2^53 - 1 ticks as a BigInt; and from 1.10 a units key other than length is refused, and a result must carry placement_id to validate; and from 1.11 a money amount past 2^53 - 1, a measure written in a spelling only one engine read (1e2, .5, 0x64) and more than 32 hull_vertices are refused, and equally ranked container types open in the code-point order of their ids. Coverage and the performance audit stay open, and the known gaps are published by name rather than discovered by users.