Repository navigation
Description
This tracks the distance between what optifab delivers and the vision: a fully end-to-end formally verified, recursively improving intelligence system that evolves its hardware and software together, as one unit. The full account is docs/VISION.md, and the proposals are docs/PROPOSALS.md. Prior work that set the idea: #1.
Where it stands
v0.2.0 is on main (#2, merged). v0.3.0 is in #13, a draft. Both are claimed work and no person has reviewed either. The "Exists" column below describes v0.3.0.
| Part of the vision | Exists | Missing |
|---|---|---|
| Formally verified | Fifteen theorems in Lean, accepted by lake comparator --paranoid under six kernels. Eight are about the gate. Seven are about designs: the plant model, the bounds, the objective and the step are inside the proved code. Proven Commons bar items 2 to 6 pass. |
A reviewer's approval of the specification (bar item 1, #8). A branch rule on main (bar item 7). Any proof that the model describes a real plant. Five known defects in the model (#14). |
| End to end | One command runs certificates → design → margins and objective → step → ledger. A second works 3,071 decisions out again in Lean from the designs, and runs each ledger in one piece through the proved loop, with no mismatch. | Both ends: no sensor feeds it (#4) and nothing obeys it (#5). 4 of 10 loop interfaces are checked. |
| Recursively improving | loop_never_worse: whatever is proposed, the objective of the design held cannot fall. The notes one proposer left were given to the next, which needed 4 proposals where the first needed 11. |
The hand-over was done once, by hand, in prose. Nothing improves the gate, the model or the proposer's own code. A design that ties is taken up, so "never worse" is not "always better" (#14). The objective drives the design onto a limit, where it can no longer change (#6). |
| Intelligence | A language model proposed through the loop's interface, seeing only the printed state, and reached the best objective a direct search finds. | Two recorded runs of one model. Neither Weft nor Exo proposes anything (#6). |
| Hardware and software | One design: 3 hardware genes, 2 software genes, one step, one ledger. The model's proposals changed both in the same step. | The hardware is three numbers in a thermal model (#7). Nothing structural moves and nothing is built (#10). |
| One unit | One predicate, one 64-byte frame, one margin vector, one record. 32 of 41 repositories placed. | No shared interface in code between the four organisations yet (#4, #9). |
| A network | A first look: the load floor depends on the site, and placement between campuses is degenerate in the current plant model. | A model with a cost that differs by site (#11). |
The theorems that carry the weight: loop_safe (from a start within its limits, whatever designs are proposed, the margins the model computes stay within the limit and the budget is never overdrawn), loop_never_worse (the objective never falls), loop_bounded (the number of changes is at most the budget a person set) and loop_keeps_physics (a margin within its limit means an exact inequality about heat or storage). They are about the model. None says the model's constants are true, and an independent read found mistakes in the model itself (#14).
Proposals and open work
Each is a sub-issue with a first slice and a test that can fail.
- Proposal 1: plant sense, two sensors and a frame emitter #4 Plant sense: two sensors and a frame emitter
- Proposal 2: load control with an energy reading #5 Load control with an energy reading
- Proposal 3: a model as the proposer #6 A model as the proposer. First slice done twice in optifab v0.3.0: plant model and loop proved in Lean; a model proposes through it and hands its notes on #13. See the comment there for what is open.
- Proposal 4: structural genes, so the loop can change the structure #7 Structural genes
- Proposal 5: a reviewer for the gate's specification, so it can be called verified #8 A reviewer for the gate's specification. The specification grows in optifab v0.3.0: plant model and loop proved in Lean; a model proposes through it and hands its notes on #13.
- Proposal 6: a signed ledger and command channel #9 Signed ledger and command channel
- Proposal 7: a first fabrication cell under the gate #10 A first fabrication cell under the gate
- Proposal 8: a network of campuses #11 A network of campuses. The first slice needs to change. See the comment there.
- Five known defects in the plant model, found by an independent read #14 Five known defects in the plant model, found by an independent read. Disclosed in optifab v0.3.0: plant model and loop proved in Lean; a model proposes through it and hands its notes on #13, not fixed.
Decisions waiting on a person
- v0.1.1: report unlocked compute in IT power; add the sorption re-cooling limit funi-frame#4: whether 30 °C is the intended design recooling temperature, and whether the compute value is per MW of IT load. The reviewer asked for both before merge.
- Five known defects in the plant model, found by an independent read #14: fix the model and re-run the recorded runs, or fix it and keep them with a note. And whether to do it inside optifab v0.3.0: plant model and loop proved in Lean; a model proposes through it and hands its notes on #13 or after.
- A branch rule on
mainrequiring thebar / barcheck (bar item 7). - Who reviews
formal/SPEC.md(Proposal 5: a reviewer for the gate's specification, so it can be called verified #8, Decide: name the first spec reviewers Proven-Commons/mission#8). - Which way to change the first slice of Proposal 8: a network of campuses #11.
- Whether the objective should give headroom a value (Proposal 3: a model as the proposer #6, Five known defects in the plant model, found by an independent read #14).
Open in other repositories
- v0.1.1: report unlocked compute in IT power; add the sorption re-cooling limit funi-frame#4: the corrections optifab found in the plant certificate. Review answered, waiting on the two rulings above.
- Index optifab v0.2.0 in the organisation profile; repair two rows .github#1: optifab in the organisation profile. Ready for review. Its row says v0.2.0.
- C2 gate raises OverflowError on a locked value beyond the float32 range, instead of refusing Moissanite-Labs/weft#220: an edge case in Weft's gate, found by the differential test here.
- Decide: name the first spec reviewers Proven-Commons/mission#8: the reviewer decision that bar item 1 waits on.
Boundaries
The gate refuses; it does not command. A person sets the budget and authorises. Nothing here moves a valve, a load or a machine, and nothing may be called verified until a person who did not write the specification approves it.
Filed by an AI agent (Claude Code) working for the owner of aska-physical.
Activity