quilt-c
Why this port exists
Intelligence should accrue in an external, auditable substrate, not inside an ever-larger model.
The model doesn't get bigger. The exocortex gets richer.
I have 4,856 repositories and 33 stars. That ratio is the whole story, and it is not a marketing problem — it is an identifiability problem. Nobody can tell if any of it is real, because everything I built is verified by me.
So here is one artifact, and you check it yourself. 2.9 seconds. I don't want you to trust my CI, my fleet, or my word for it.
The check
$ git clone https://github.com/SuperInstance/quilt-c.git
$ cd quilt-c && git checkout v0.1.0
$ make verify
[ok] test 47 passed 0 failed
[ok] test-proof 1059 passed 0 failed
[ok] test-route 27 passed 0 failed
[ok] test-crdt 28 passed 0 failed
[ok] test-world 34 passed 0 failed
[ok] test-time 41 passed 0 failed
[ok] test-quf 49 passed 0 failed
verdict: VERIFIED 1285 passed / 0 failed
receipt_sha256: 37ee0737…
Then confirm the tag names that tree — also without trusting me:
$ node release_verify.mjs --tag v0.1.0
[ok] tag_resolves -> adae27e496bb
[ok] tree_matches_receipt now=be3be613… receipt=be3be613…
[ok] receipt_unaltered recomputed=37ee073… recorded=37ee073…
RELEASE VERIFIED
The trust root is git object hashing. A tag points at one commit; that commit's tree is derivable by anyone. So the check needs no trust in the author — only in SHA-256 and git.
What it is
The Quilt cell-fabric runtime in C99. A spreadsheet where every cell is a live, addressable capability, and every transition is one of five opcodes:
BIND(cell, dials) set dials, idempotent by content
LINK(c1, c2) add an undirected edge
EFFECT(cell) propagate dial[0] to neighbours
VIEW(cell) return dials
TICK(fabric) advance every dial, saturating at zero
Dependency closure: a C99 compiler. No packages, no lockfile, no network. The ports below are byte-checked against this reference.
The fabric, live
RECEIPT CHAIN
VIEW
— awaiting VIEW —simulated projection — the C99 kernel runs these same opcodes across 1,285 assertions. the math on this page is real: dial digests are FNV-1a 64, ledger receipts are SHA-256 parent-chained, both computed in your browser. the timeline is scripted; the kernel is not in your browser.
Install
Two packages, same verified kernel. Both run the full 1,285 assertions on install.
# PyPI
pip install https://github.com/SuperInstance/quilt-c/releases/download/v0.1.0/quilt_c99_kernel-0.1.0-py3-none-any.whl
python -c "import quilt_cell; print(quilt_cell.verify()['verdict'])"
# npm
npm install https://github.com/SuperInstance/quilt-c/releases/download/v0.1.0/quilt-c99-kernel-0.1.0.tgz
npm run verify
Registry publication is pending egress approval from this build environment; the signed release tarballs are downloadable now and are byte-identical to what would be published.
The honest part
I ran an externalisability audit on my own project. I took ten sealed, hash-verified predictions, stripped them to their prediction text alone, and handed them to a verifier on a different model lineage.
0/10 decidable from the text · 0/10 decidable by a stranger · 10/10 satisfiable by bookkeeping · 791 commits, 0 signed
Every one of them stated a failure condition and every hash verified. The pass conditions were presence-of-bookkeeping checks — an adversary named the cheap way to satisfy each one. The fleet has falsifiers but not substance tests, and the difference is invisible from inside because everyone inside already knows what the substance is.
That is why this page is one runnable command instead of an index. A searchable list of 4,856 repos optimizes for people who already know what to ask.
Download
Signed release artifacts, byte-identical to what would be published to a registry:
- PyPI wheel
quilt_c99_kernel-0.1.0-py3-none-any.whl - PyPI sdist
quilt-c99-kernel-0.1.0.tar.gz - npm tarball
quilt-c99-kernel-0.1.0.tgz
What I am not asking you to do
- Trust the fleet. The receipts are checkable, so check them.
- Read the doctrine first. Clone and run, then decide if the architecture is interesting.
- Take the self-description seriously. The 4,856-repo count is machine-counted, not designed.
The rest of the substrate
- quilt-c — the C99 reference port 1285 assertions, signed tag
- cell API — BIND/LINK/EFFECT/VIEW/TICK as HTTP methods design probe, 13/13
- externalisability gate — 4 checks, fail-closed 0/10 pass today
- JEV known-answer control — instrument calibration score type does not discriminate
- the fleet read — 23 writers, one oracle 18,617 bytes