StarForth-v4.0.0 into master: 122 commits since 268e5593. Most of them carry out #189: the kernel's
boot of the v4 fabric, built to spec (Captain Bob: "Build it to proper spec").
The mesh boot (046ce0be). Hera is born empty at place 0 and given the nucleus as a capsule. POST
runs in the fabric. Nodes are born and killed through her. The fabric is lockstep: every node that is
not blocked executes one instruction word at each step. It is not a scheduler (ENGINE.md 3d; MESH.md
ruling 7).
The word accounts at call, and the loops (#189 Phases 2 and 7). v4's POST is v3's run_all_tests, as written: 990 cases.
The one adaptive heartbeat (docs/v4.0.0/plans/2026-11-17-one-adaptive-heartbeat.md). The
kernel's heartbeat paces itself. Its only input is the Stadium's resident heat. Every VM ticks on it in
lockstep, active or not. It is held through POST.
Mechanism A (edbb79e2). Each step walks only the places that hold a node. The host mesh boot's
median fell from 11.695 s to 7.999 s over five alternating runs. aarch64 had stopped in POST at 570 s
since 37469f6c; it now boots inside it. The direct loop before 046ce0be was architecturally incorrect;
the fabric wiring removed it and exposed the slowdowns, and mechanism A is the first speed-up the
instruments found (FABRIC-3 XLVIII.25-26).
The fault dump (v4/system/dump.c). When a node stops, in every build, it prints the fault,
P/A/B, both stacks, the return stack named by word, and the running line.
scripts/nightly.sh diagnose. After any failed stage, the EMERGENCY build is made and its
artifacts kept; the workflows run it on failure. protocol-report now writes a report when a boot
stops before POST.
The documents brought up to date, each by Captain Bob's ruling: README, v4/README, DATASHEET-F18
section 13, capsules/README, .claude/CLAUDE.md.
Acceptance
logs/20261117-235006/PROTOCOL.md for the head: the protocol on c30e4a0a, PASS.
amd64, aarch64, riscv64 and the PC stick loaded high each reach [zuse@Hera] ok> with PARITY:V4_POST tests=990 pass=990 fail=0 and one dict_hash, 0x627f392ccbaa96f1.
test-32, the proofs, the images and the docs pass.
Logs from the other runs made by hand are committed in logs/: logs/20261117-mechanism-a/
(timing, sanitizers, tests), logs/20261117-heartbeat/, logs/20261117-dump/, logs/20261117-195259/fabric-cost/.
Hardware this was tried on: none (QEMU only).
The record
docs/fabric/FABRIC-3.md has the rulings and what was built: XLVIII.24-26 and the entries
before them.
CHANGELOG.md: not updated in this branch.
Nothing in this PR is a fix nobody asked for. Each change was ruled by Captain Bob.
## What and why
`StarForth-v4.0.0` into `master`: 122 commits since 268e5593. Most of them carry out #189: the kernel's
boot of the v4 fabric, built to spec (Captain Bob: "Build it to proper spec").
- **The mesh boot** (046ce0be). Hera is born empty at place 0 and given the nucleus as a capsule. POST
runs in the fabric. Nodes are born and killed through her. The fabric is lockstep: every node that is
not blocked executes one instruction word at each step. It is not a scheduler (ENGINE.md 3d; MESH.md
ruling 7).
- **The word accounts at `call`, and the loops** (#189 Phases 2 and 7). v4's POST is v3's
`run_all_tests`, as written: 990 cases.
- **The one adaptive heartbeat** (`docs/v4.0.0/plans/2026-11-17-one-adaptive-heartbeat.md`). The
kernel's heartbeat paces itself. Its only input is the Stadium's resident heat. Every VM ticks on it in
lockstep, active or not. It is held through POST.
- **Mechanism A** (edbb79e2). Each step walks only the places that hold a node. The host mesh boot's
median fell from 11.695 s to 7.999 s over five alternating runs. aarch64 had stopped in POST at 570 s
since 37469f6c; it now boots inside it. The direct loop before 046ce0be was architecturally incorrect;
the fabric wiring removed it and exposed the slowdowns, and mechanism A is the first speed-up the
instruments found (FABRIC-3 XLVIII.25-26).
- **The fault dump** (`v4/system/dump.c`). When a node stops, in every build, it prints the fault,
P/A/B, both stacks, the return stack named by word, and the running line.
- **`scripts/nightly.sh diagnose`.** After any failed stage, the EMERGENCY build is made and its
artifacts kept; the workflows run it on failure. `protocol-report` now writes a report when a boot
stops before POST.
- **The documents brought up to date**, each by Captain Bob's ruling: README, v4/README, DATASHEET-F18
section 13, capsules/README, .claude/CLAUDE.md.
## Acceptance
- [x] `logs/20261117-235006/PROTOCOL.md` for the head: the protocol on c30e4a0a, **PASS**.
- amd64, aarch64, riscv64 and the PC stick loaded high each reach `[zuse@Hera] ok>` with
`PARITY:V4_POST tests=990 pass=990 fail=0` and one `dict_hash`, `0x627f392ccbaa96f1`.
- test-32, the proofs, the images and the docs pass.
- `scripts/check_protocol.sh` accepts 58d0e887.
- [x] Logs from the other runs made by hand are committed in `logs/`: `logs/20261117-mechanism-a/`
(timing, sanitizers, tests), `logs/20261117-heartbeat/`, `logs/20261117-dump/`,
`logs/20261117-195259/fabric-cost/`.
- [ ] Hardware this was tried on: none (QEMU only).
## The record
- [x] `docs/fabric/FABRIC-3.md` has the rulings and what was built: XLVIII.24-26 and the entries
before them.
- [ ] `CHANGELOG.md`: not updated in this branch.
- [x] Nothing in this PR is a fix nobody asked for. Each change was ruled by Captain Bob.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
These logs/ files were committed as plain text even though
.gitattributes routes logs/**/*.log through the LFS filter, so every
fresh clone showed them all as modified. Renormalize them so they are
stored as LFS pointers. File contents are unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Their pointers were committed in a5ed8c3d (2026-08-01) and the objects were
never uploaded: the server answers 404 for all of them, and no copy exists
(Captain Bob, 2026-11-11: "they can be trimmed out").
Removed: 3,827 phase-1 logs (docs/working/archive/legacy/phase-1), 3 in
docs/working/archive/historical, 106 DoE CSVs in experiments/bare_metal/runs,
QEMU_BASELINE.log, docs/SYSTEM_NARRATIVE.log, docs/ontology.log.
Kept: the 5 LaTeX build logs under docs/patent (legal hold) -- excluded from
fetch instead. .lfsconfig excludes every one of these paths, so older commits
check out and LFS fetches run without 404s.
Every LFS pointer left in the tree (1,195) has its object on the server.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
v3/include/version.h is written by every build (kernel/Makefile and
v3/Makefile, both FORCE), with a timestamp and the last build's ARCH; the
committed copy was stale (LITHOS_VERSION 2.1.0). Untracked, and .gitignore's
entry corrected: it said include/version.h, the path before the move to v3/,
and matched nothing.
capsules/BLOCK_MAP.md differed on every build only in mkcapsule's
"Generated ... <timestamp>" header. The timestamp is dropped, so the manifest
changes only when the capsules do; two runs are byte-identical.
mkcapsule's warnings unchanged (5, #143).
Captain Bob, 2026-11-11: untrack version.h and fix the ignore; drop the
manifest timestamp.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
amd64, aarch64, riscv64 (clean qemu), stick-high, test-32, run one at a
time in the foreground through scripts/nightly.sh:
PARITY:V4_POST tests=538 pass=538 fail=0, POST: PASSED, [zuse@Hera] ok>
on every boot; PARITY:V4_SYSTEM word_count=318
dict_hash=0x8c92508f748f99cf on all three; test-32 39 programs, 0 failures.
No tracked file changed across the five builds.
20261111-170703 amd64 PASS
20261111-170909 aarch64 PASS
20261111-171235 riscv64 PASS
20261111-171410 stick-high PASS
Also kept as evidence:
20261111-170134 amd64 on aec7ddb5 (before the clean-tree fix), PASS
20261111-170544 amd64 on 90c5c140, stopped by me during boot: it had been
started in the background, against the protocol, and was
rerun in the foreground as 20261111-170703
DoE CSVs for every completed boot (experiments/bare_metal/runs).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
90c5c140 untracked v3/include/version.h, but six places still put it back
with git checkout. Git refuses the whole checkout when one path is unknown:
- scripts/nightly.sh images (set -e): the stage ended there, so the nightly
(images, publish, proofs) and release.yml (images, docs, gitea-release)
would fail. Acceptance does not run images, so it passed.
- scripts/check_images.sh (no -e), demo/build-iso.sh, demo/mint-zuse.sh
(|| true): silently put nothing back, BLOCK_MAP.md and the SBOM included.
- scripts/qemu_session.py: an error after every boot, otherwise harmless.
- boards/BRINGUP.md: the step said to put it back.
BLOCK_MAP.md stays in every put-back: a signed build (CI's) still rewrites
its Signed column. Found while setting up alpha.5; Captain Bob: fix it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Run one stage at a time in the foreground through scripts/nightly.sh:
20261111-185352 amd64 PASS
20261111-185600 aarch64 PASS
20261111-185950 riscv64 PASS
20261111-190128 stick-high PASS
PARITY:V4_POST tests=538 pass=538 fail=0, POST: PASSED, [zuse@Hera] ok>
on every boot; word_count=318 dict_hash=0x8c92508f748f99cf on all three;
test-32 39 programs, 0 failures.
scripts/nightly.sh images -- the stage the version.h put-back had broken --
built PC, RASPI and MILKV and left the tree as committed.
DoE CSVs for every boot (empty, as every v4 acceptance boot's: #141).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
XLVIII.18 records 2026-11-11: the LFS trim (Captain Bob: "they can be
trimmed out"), version.h untracked and BLOCK_MAP.md without its timestamp
(#145), my put-back error and its fix, the acceptance rulings (never in the
background; all logs and CSVs are evidence), the acceptance of 9415d7ed,
and alpha.4's Gitea release marked final.
README: alpha.4's release is no longer a pre-release.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
logs/20261111-193825/v4-test: make -C v4 test on feb740c5 -- 79 programs,
3,740,484 checks, 0 failures, at V4_CELL_BITS=32 and 64. CI's v4-tests job
(Build run 247) was cancelled at 0 s by the shared qemu concurrency group
(#140), so this is the run of record for that job.
logs/20261111-185352/nightly: what scripts/nightly.sh wrote only under
build/nightly for 9415d7ed's acceptance (d6846b34) -- the four accept
summaries, test-32 (39 programs, 0 failures) and the images stage (PC,
RASPI, MILKV). The test-32 output of the earlier 90c5c140 run was
overwritten in build/nightly before it was kept, and is not recoverable.
Captain Bob, 2026-11-11: all logs are evidence.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Every workflow shared the concurrency group qemu. A group keeps one running
and one waiting run, and a run that joins cancels the one waiting: Build
runs 245 and 247 were cancelled at 0 s by PR 164's Acceptance runs.
Captain Bob, 2026-11-11: "It should be global"; the runner has capacity 1.
So the concurrency block is removed from all six workflows (a comment in
each says why): with one runner at capacity 1 every job waits its turn in
Gitea's queue and none is cancelled, and a runner-image rebuild still
blocks the rest. CLAUDE.md's CI section corrected; FABRIC-3 XLVIII.19.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob, 2026-11-11: "I'm thinking about adding a CAPA & FMEA GATE to our
build pipelines", then yes to writing it up. A nightly.sh quality stage run in
the acceptance gate: every kind/escaped issue has a CAPA record whose
preventive control exists and runs in a stage; every FMEA row over the
threshold has a control in the gate. Seeded with five CAPAs from 2026-11-11.
Five points to rule (section 7).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
scripts/check_protocol.sh [<rev>]: refused (exit 1) unless <rev> carries a
logs/<TS>/PROTOCOL.md naming a commit whose protocol passed on a clean tree,
and that commit differs from <rev> only in the evidence (logs/ and
experiments/bare_metal/runs/, committed after the run). Any other change since
makes the evidence stale: the protocol is run again.
scripts/test_check_protocol.sh, written first and seen to fail (8 of 8, and
tightened so a missing script is not taken for a refusal): no protocol, stale,
failed, dirty, unknown commit -- refused; fresh, fresh among stale ones, other
files in logs/ -- passes. 8 of 8 pass.
Not yet wired into acceptance.yml: that comes with nightly.sh protocol.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
scripts/nightly.sh protocol: every stage a PR, the nightly and a release run --
accept amd64, aarch64, riscv64, stick-high, test-32, proofs, images, docs --
in order, never publish; the first failure ends it. protocol <stage>... runs
those only, so each can be its own foreground command (the QEMU rule);
accept-amd64 begins a protocol, recording the commit and whether the tree is
clean. protocol-report writes logs/<TS>/PROTOCOL.md (the commit, the tree,
each stage's result and output, POST and dict_hash per architecture compared,
test-32's totals, the proofs' log, the images' and documents' SHA-256) and
keeps every stage's output in logs/<TS>/protocol/ -- nothing is left only in
build/nightly. PASS only if every stage passed on the commit it began on and
the tree is unchanged but for the evidence.
acceptance.yml: scripts/check_protocol.sh on the PR's head, first, before the
boots. The PR template asks for PROTOCOL.md; .claude/CLAUDE.md names the
protocol as the acceptance a developer runs.
Captain Bob, 2026-11-07 (#135); 2026-11-12: "Keep working on alpha.5".
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
logs/20261112-001304/PROTOCOL.md: scripts/nightly.sh protocol, one stage per
foreground command: accept amd64, aarch64, riscv64, stick-high -- 538/538,
word_count=318, dict_hash 0x8c92508f748f99cf on all three; test-32 39 programs,
1,869,197 checks, 0 failures; the proofs (StarForth, 36 s); the images (PC,
RASPI, MILKV); the documents (book, fabric, dictionary, vol1-3, api). The tree
clean before and after. Every stage's output in logs/20261112-001304/protocol/,
the boots' serial logs and DoE CSVs beside it.
The first protocol run: the evidence the PR's Acceptance check now asks for.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob, 2026-11-12: "Is it possible to include the FMEA in the pr
review? That way the items are forced into a manual review beforehand?" --
"Yes, add it to the plan". Each FMEA row names its paths; the PR carries a
review record with a decision for every row its diff touches; a
check_fmea_review.sh beside check_protocol.sh refuses the PR without one, or
with one for another diff. A sixth point to rule. Nothing built.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Each cleanup #129 set up (Captain Bob, 2026-11-07), and what proves it ran:
the Docker prune's and the package cleanup's logs written in the last 8 days;
no image older than 8 days that no container uses; at most KEEP (5)
lithosananke-runner tags; no artifact older than ARTIFACT_DAYS (30) + 1; no
Nexus snapshot asset older than 8 days (with NEXUS_USER/NEXUS_PASSWORD,
skipped and said so without); a dated disk line appended to HISTORY each run,
so "disk use flat" is a comparison of weeks. PASS:/FAIL:/SKIP:/INFO:,
RESULT:, exit 0 only if nothing fails.
Tested here against fixtures (a stale log, an old artifact, no docker -- each
FAIL; clean -- PASS) and a stand-in docker (one stale unused image of three --
FAIL, the in-use and the recent ones kept). The registry check ran against
the real registry: 2 tags. Not run here: the real Docker, Nexus.
Captain Bob, 2026-11-12: "Correct the issue text and write the check script".
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
logs/20261112-114545/PROTOCOL.md: one stage per foreground command (paused
between stages 2 and 3 for Captain Bob's emergency, then resumed on the same
commit): the four boots 538/538, word_count=318, dict_hash
0x8c92508f748f99cf on all three; test-32 39 programs, 1,869,197 checks, 0
failures; the proofs (StarForth, 44 s); the images; the documents. The tree
clean before and after.
The evidence for the held batch: plan section 8 (the FMEA in the PR review),
the notes, and scripts/ci_litter_check.sh (#129).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
failed() printed "MINT-ZUSE: failed -- nothing changed" on every failure
after YES, though by then gpt_write_one() had rewritten the stick's partition
table and the identity partition's first blocks had been zeroed.
gpt_write_one() now tells the two apart: GPT_EIO when its first write failed
(nothing written), GPT_PART when a later write or flush failed (the disk is
changed). MINT-ZUSE keeps "failed -- nothing changed" for GPT_EIO only, and
says "failed -- the stick was erased; Zuse not minted" for every failure
after the table is written. The marker is still written last.
Tested first, seen to fail: qemu_session.py mint_part_fail (writes fail only
at the identity partition's first sector, LBA 2048) -- "nothing changed",
logs/20261112-122132. After: mint_part_fail PASS; mint_fail (every write
fails) still "nothing changed", PASS; replace (the success path) PASS. Their
logs here. FABRIC-3 XLVIII.20.
Captain Bob, 2026-11-12: "start #168".
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
logs/20261112-124144/PROTOCOL.md: the four boots 538/538, word_count=318,
dict_hash 0x8c92508f748f99cf on all three; test-32 39 programs, 1,869,197
checks, 0 failures; the proofs (StarForth, 37 s); the images; the documents.
The tree clean before and after.
Stage 3 (riscv64) was stopped while booting, when Captain Bob interrupted to
mean the build pipeline; its partial serial log and console output are kept,
labelled, in logs/20261112-124831/. It was rerun on the same commit and
passed (logs/20261112-130801).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
The stick-high stage passed when one distinct dict_hash appeared: an
architecture whose boot left no accept file, no LOG: line, no serial log or no
PARITY:V4_SYSTEM line gave nothing, and two that agreed passed for three.
It is now a stage of its own, scripts/nightly.sh dict-hash, which stick-high
runs: one hash from each of amd64, aarch64 and riscv64, all found and one value;
otherwise refused, naming the architectures with none or the values that
differ. NIGHTLY_OUT points the stages at another directory, for the tests.
scripts/test_nightly_dict_hash.sh, written first and seen to fail (6 of 6, the
stage not existing), then 7 cases: agree -- pass; differ, no PARITY line, no
LOG: line, not booted, no serial log, none at all -- refused. Two of them
caught lookups that ended the script silently under set -e; guarded.
Captain Bob, 2026-11-12: "172 then finish alpha.5".
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
v4_manage_serve took its arguments with v4_dstack_pop and never checked the
stack's depth: a request with too few values acted on stale ring values, and
one whose results did not fit dropped the oldest silently -- around the D-16
guards. Now each request's arity (manage.h) is checked first: fewer values
than it takes, or no room for what it leaves, is error 12 with the stack as it
was -- as a block request is refused (blocks.c).
test_manage.c, written first and seen to fail (12 of 13 requests at both cell
widths; NODE-BORN passed only because the stale value read was below 1): every
request with one value fewer than it takes, and NODE-ME and CAPSULE-CELL on a
stack with no room -- refused, the stack unchanged. After: 78 checks, 0
failures at 32 and 64 bits; all v4 tests at 64 bits, 40 programs, 1,871,313
checks, 0 failures (logs/20261112-152602/v4-test).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
kernel/Makefile gave capsules/v4/sys/map-64.f18 and map-32.f18 one rule with
two plain targets -- in GNU make, two rules sharing a recipe. Under -j both
ran at once: two make -C v4 sysmap built and ran the same v4/build/mkmap-<w>
side by side (Build run 255's RASPI job: "mkmap-32: Permission denied").
Now one grouped target (&:, GNU make 4.3+, as here and in the CI image).
Seen to fail first: the two targets with -j8, mkmap removed each time, 12
runs -- 6 failed ("Text file busy", "Permission denied"). After: 0 of 12.
Both sets of runs in logs/20261112-152711/race-185/. The map files are unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
The header said "M6: Infrastructure present (validation deferred)" and "M7:
Not started (VM integration pending)", and "(LithosAnanke branch)" -- there
is no such branch since the split. .claude/CLAUDE.md had warned since
2026-08 not to propagate it instead of it being corrected. Now: M0-M6 and M7
complete, M7.1 in progress (as CLAUDE.md's Architecture section gives them),
v4 the system booted by default; the correction dated in place. The note in
CLAUDE.md's Gotchas goes, as #144 asks. Comments only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
The kernel build's only warnings were five -Wformat-truncation in mkcapsule's
--resolve: two snprintf copies of a capsule name into the claim registry, and
the capsule's path built from base_dir and its name.
- The names: both are char[CAPSULE_NAME_MAX], and names are checked to fit when
read (line 335) -- copied whole with memcpy, nothing to truncate.
- The path: base_dir may be up to 4095 bytes, so it can be cut, and a cut path
names another file -- now refused, the capsule left unchanged and counted
rejected, with the reason.
Seen to fail first: built with -Werror=format-truncation, 5 errors. After: 0,
and 0 warnings with the kernel's flags; the manifest it writes is
byte-identical, --lint passes. The too-long-path refusal is on --resolve's
interactive conflict path and was not driven by a test. CLAUDE.md's count of
open warnings corrected.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
- heat.c call_index(): the comment read as an unfinished thought ("hash-mod or
just take low bits?", "ignore for now? Or clamp?"). It now states the rule
the code follows: indexed by the target, a target past the table not counted,
-1 telling the callers so; the table D-6's stand-in, which ENGINE step 4
Part A1 plans to remove.
- stack.h points 1 and 2 said there was no overflow detection "and there never
will be" and that a depth exceeded corrupted silently -- D-2, which D-16
revised. Now: guarded before every opcode, a mesh node 10 and 9, the host node
32 and 32 (D-17); C code that serves a node checks the depth itself.
- stack.c named an 8-ring and depths 10 and 9 only, and said D-2's rule still
held.
- DATASHEET-F18.md section 13 item 4 (this gap) struck, resolved in place.
Comments only: test_stack and test_heat unchanged at both widths (730,137 and
8,272 checks, 0 failures).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
A data stick hot-plugged with every read at its identity partition's first
sector (LBA 2048) failing, after a boot with no drive; it expects a message
naming the identity partition, Artemis let go (both partitions or neither,
#179), no prompt, and the same stick without the fault going on to genesis.
add_stick() takes fail_read_sector for it.
It FAILS today, and is kept so (Captain Bob, 2026-11-12: "keep the test"):
the partition attaches anyway and Zuse is minted onto it, because
blk_format_or_load_disk() ignores the header read's result and takes an
unreadable disk as blank -- filed as #187. #179 is parked: its half-attached
state cannot be reached from a disk fault. The run: logs/20261112-160050.
Not part of acceptance.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
gpt_read() stopped at max used entries (16 for the identity code, 4 in the
mint): a disk whose StarshipOS partitions came after the first max read as
having none, and the entries after max were never range-checked.
Now every entry is checked and every used one counted; a good table with more
used entries than max is GPT_MANY (the first max in parts), and the identity
code says "GPT: a USB drive has more than 16 partitions -- not used". Other
callers treat anything but GPT_OK as before.
Tested first, seen to fail: qemu_session.py gpt_many -- a data stick with its
two partitions as entries 17 and 18 after 16 others, and the same with an entry
19 past the usable range: both "not this system's stick" before (silently cut,
entry 19 never checked; logs/20261112-162236). After: "more than 16
partitions" and "a damaged partition table" (logs/20261112-162921).
Regressions: gpt_bad_entries, gpt_foreign PASS. gpt_zuse FAILS, on a stale
expectation unrelated to this change -- an identity-only GPT drive with no
system stick has been held silently as a possible real Zuse since #92 Task 4;
its log is kept here. The gpt_many images are made before the boot: a first
version made them after, failed in sgdisk and left its QEMU running, stopped
by hand (logs/20261112-162003).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
logs/20261112-164108/PROTOCOL.md, for the batch #172, #169, #185, #144, #143,
#181, #180 and the #187 test: the four boots 538/538, word_count=318,
dict_hash 0x8c92508f748f99cf on all three -- the new dict-hash stage (#172)
naming each; test-32 39 programs, 1,869,223 checks, 0 failures (the manage
test's 78 among them, #169); the proofs (43 s); the images, no race (#185);
the documents. The tree clean before and after.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob, 2026-11-12: "Build an appropriate test harness! Love the idea."
The kernel had no tests below the boots, so defects QEMU cannot provoke could
not be tested first. Plan: docs/v4.0.0/plans/2026-11-12-kernel-host-tests.md.
kernel/tests/: one host program per kernel source, built with that source as
it is, the kernel's flags (-std=c99 -Wall -Wextra -Werror), and fakes --
the console captured line by line, a deterministic RNG that can fail, and a RAM
disk whose reads, writes or flushes fail at chosen 1 KiB blocks.
make -C kernel/tests test runs them; SRC_<test>= runs a test against other
code, which is how a test is seen to fail against the code before a fix.
test_gpt.c, 12 checks: a table written and read back; 16, 17 and 18 used
entries against max 16 and an entry out of range after the 16th (#180); the
RNG, the first write, a later write and a flush failing (#168). Against gpt.c
before #168 and #180: 4 failures, each defect caught. Now: 0. Both runs in
logs/.../kernel-tests/. Not yet a protocol stage: to rule (the plan, step 5).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
xhci.c's XHCI_XFER_CONFIG_DESC_FULL case walked the descriptor set inline, in
the event loop, where nothing could test it. The walk is now
usb_config_parse() (kernel/src/usb/usb_config.c), returning what it found;
the case acts on it as before -- the same messages, the same decisions: the
first Interface descriptor only, a BOT interface's bulk endpoints (the last of
each direction), a boot keyboard's first interrupt IN endpoint, bLength 0
ending the walk, nothing read past len. One difference, in debug output only:
the bulk endpoints are logged IN then OUT, not in descriptor order.
kernel/tests/test_usb_config.c, 7 checks, pins the walk as moved: a BOT stick
as QEMU presents it, a boot keyboard with its HID descriptor, a mouse before a
drive (the first interface only), a drive with bulk IN only, bLength 0, an
endpoint cut by len, an interface cut before its protocol byte. All pass.
Built for amd64, aarch64, riscv64; the boots come with the protocol (a data
stick enumerates through this path). The fixes for #170 and #171 follow.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
After the configuration descriptor's first 9 bytes the event loop asked for
min(wTotalLength, 128) whatever it was. A wTotalLength of 9 or less was then
asked for again as "short" -- for ever, that slot never leaving enumeration
(#170); 0 also issued a zero-length data stage. A set longer than the 128-byte
buffer was cut without a word, its later interfaces never seen (#171).
The decision is usb_config_request_length() (moved out of the event loop,
then fixed): 9 or less is 0 -- "xhci: a configuration of N bytes has no
interface -- not used", and nothing more is asked; longer than the buffer is
cut and said -- "xhci: config of N bytes cut to 128 -- rest unseen" (at most
53 characters). The buffer stays 128 bytes; its size is #171's to rule.
kernel/tests/test_usb_config.c, seen to fail first against the decision as it
was (10 failures: wTotalLength 1 to 9 asked for again, the cut unsaid), now
20 checks, 0 failures. Both runs in logs/20261112-180140/kernel-tests/. Built for amd64,
aarch64 and riscv64; the boots come with the protocol. #171's other box, a
composite device on the SER5, is Captain Bob's.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
a8824655's evidence log shows its "before" run failing to compile: the copy
of usb_config.c with the old decision was cut at the wrong brace by my script.
The failure the commit describes was real (seen in the session), but the log
did not show it. Rerun with a correct copy:
test_usb_config-before-rerun.log -- 20 checks, 10 failures (wTotalLength 1 to
9 asked for again; the cut unsaid). The first log is kept, as what it is.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
sk_v4_run's three stops -- "not started", "did not let go of POST's
blocks", "stopped on a fault" -- ended in for (;;) { }: a core spinning at
100%, the thing #151 removed from the prompt. Each now calls sk_v4_stop()
(kernel/src/v4/sk_v4_stop.c): interrupts off, then arch_halt() for ever. Each
still says why first. sk_v4_stop.c added to the loader's and the kernel's v4
sources.
kernel/tests/test_sk_v4_stop.c, with arch_halt and arch_disable_interrupts
faked (the halt jumps back out after three calls; an alarm fails a spin):
against the stop as it was -- "spun for 2 s without halting"; now -- 2
checks, 0 failures: halts in a loop, interrupts off before the first halt.
Built for amd64, aarch64 and riscv64. #177's "seen on the three
architectures": no boot reaches these stops, so it is shown on the host only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
sk_v4_idle allocated a controller's slot array again whenever max_slots + 1
was more than it held -- the old array dropped, neither copied nor freed --
while the identity code holds pointers into it (g_offered, g_spare,
g_pending_real, g_real_drive, g_zuse_dev). Its two slot loops then ran to
max_slots whatever the array held.
sk_v4_slots_get() (kernel/src/v4/sk_v4_slots.c): allocated once, zeroed, at the
size first asked for; never replaced; the slots that may be used are those it
has. sk_v4_idle uses it, and its loops stop at the array's end.
kernel/tests/test_sk_v4_slots.c, kmalloc_aligned faked, seen to fail first
against the allocation as it was (4 failures: replaced when more were asked
for, 17 said usable where 9 existed, a held pointer left behind, an allocation
for 0); now 6 checks, 0 failures. Built for amd64, aarch64, riscv64; latent
today (max_slots does not grow), so the boots show no change.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
sk_v4_grow.c's dev_rw4k computes 1 KiB block numbers as (uint32_t)devblock * 4,
which wraps past 4 TiB; Artemis is capped at 256 GiB (STICK_ARTEMIS_CAP_GIB,
kernel/Makefile), so it is latent -- but nothing said so where the cast is. Now
a compile-time check there (the C99 negative-array pattern the tree uses): a
cap past 4 TiB stops the build, and a cap not defined at all is an #error, not
a skipped check. gpt.c's casts are bounded by the disk's own size, itself a
32-bit count of 1 KiB blocks, checked before each: each now says so.
kernel/tests' check_cap, seen to fail first (a 4096 GiB cap built); now 3
checks, 0 failures -- 256 GiB builds, 4096 GiB and none refused. Built for
amd64, aarch64, riscv64. The plan's increments 2 and 3 marked.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
sk_v4_mint_zuse wiped seed and rseed only when the real Zuse was minted; its
refusals and every failure after YES left them in stack memory.
Option 1 (Captain Bob, 2026-11-12: "1 is great"): the seeds are in a 64-byte
buffer the caller owns. sk_v4_mint_zuse_with(seeds) runs the mint (now a
private mint()) and then wipes the buffer, whichever way it returned -- one
place, every exit; sk_v4_mint_zuse() gives it a buffer of its own.
kernel/tests/test_sk_v4_mint.c, 17 checks: the real gpt.c on the RAM disk, the
identity code, key mint, signature checks, marker and keyboard faked; each of
the mint's eight exits run with the buffer filled with 0xEE first -- no stick,
under 64 MiB, not YES, the key mint failing, the read-back not hers, no
marker, the marker write failing, and minted. Against the mint as it was: 7
failures (every exit but success). Now: 0. Both runs in logs/20261112-181719/. Built for
amd64, aarch64, riscv64. The identity code's two (sk_v4_genesis,
sk_v4_real_zuse_arrived) follow.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
sk_v4_genesis never wiped its seed or its read-back; sk_v4_drive_arrived
never wiped the proxy's seed after her login; sk_v4_real_zuse_arrived wiped
the seed only on a match. (The logged-in Zuse's own key stays in g_zuse_seed,
by design.)
Option 1: the three use one buffer the identity module owns, g_seeds (seed,
then its read-back), and sk_v4_drive_arrived -- which every drive goes through,
and which alone calls the other two -- runs the drive's handling (now
handle_drive()) and then wipes it, whichever way it went.
sk_v4_identity_seeds() lets the host tests read it.
kernel/tests/test_sk_v4_identity.c, 10 checks, through the real drive-arrival
path: a fake xHCI controller marks a slot attach-pending, the identity code's
idle beat handles it, the real gpt.c reads the stick's table from the RAM disk;
the block layer, signatures, key mint, marker and console faked. Genesis; a
genesis not read back; the proxy logging in; a real Zuse named, then her stick
logging in. Against the code before the wipe: 5 failures (every case's seeds
left). Now: 0. The harness gains fake_disk_bind. Built for amd64, aarch64,
riscv64. #176 done with 2bac2ac8 (the mint).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
logs/20261112-182015/PROTOCOL.md, for the kernel host-test chunk: the harness
and #170, #171, #176, #177, #178, #182. The four boots 538/538 -- a data stick
enumerating through the moved USB descriptor walk, genesis through the identity
code's new seed handling -- word_count=318, dict_hash 0x8c92508f748f99cf on all
three; test-32 39 programs, 1,869,223 checks, 0 failures; the proofs (37 s);
the images, no race; the documents. The tree clean before and after.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob, 2026-11-12: "correct deficiencies". heartbeat_capture_tick_snapshot
wrote cache_hits_delta and bucket_hits_delta as 0 ("TODO: wire ..."), though
physics_hotwords_cache.c counts both. Now each is read from the VM's hot-words
cache as a delta since the previous tick, the way word_executions_delta is
(and after the cache's stats are reset, the count since then); a VM with no
cache has 0. word_executions_delta was measured already: doe_log.c's "TODO: 0"
for it was stale, and its three column notes are corrected.
kernel/tests/test_heartbeat_export.c, 4 checks: a VM in memory with a cache
whose counters the test sets, captured tick by tick. Against the code as it
was: 3 failures (the hits 0 where the cache counted 5 and 3, then 7 and 7).
Now: 0. Built for amd64, aarch64, riscv64 and the hosted v3. Shared v3 code,
under the ruling. v3 is not part of acceptance; no DoE has run since 2026-09-16.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob, 2026-11-12: "Yes". Acceptance checked POST: PASSED, so a POST
that lost cases still passed. Now scripts/nightly.sh post-count, which the
stick-high stage runs after dict-hash: each of the four boots must report
PARITY:V4_POST with every test passed and at least the floor's tests --
v4/system/POST_FLOOR, 538 today, raised when POST grows and never lowered
without Captain Bob. A boot with no POST line, not run, or no floor file is
refused, not skipped.
scripts/test_nightly_post_count.sh, written first and seen to fail (the stage
not existing), then 7 cases: 538 and 600 pass; fewer than the floor, a
failure, no POST line, a boot not run, no floor -- refused. Run on the boots
of the last protocol (d7b2f6bf): all four 538 of 538. logs/20261112-185620/.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
4738aee4's log has a placeholder where the run against the code before #173
belongs ("see the session") -- not evidence. Rerun against
scripts/nightly.sh as it was (4738aee4~1): every one of the 7 cases fails,
the stage not existing. post-count-before-rerun.log; the first log is kept.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob, 2026-11-12: "Yes, reject unsigned builds". acceptance.yml's
"The capsule-signing key" step warned when CAPSULE_SIGN_KEY was missing and
the PR's acceptance went on, unsigned. Now it is an ::error:: and the step
fails, so the gate does too. A local or protocol build may still be unsigned
(kernel/Makefile, v4/Makefile): the refusal is the gate's. build.yml,
nightly.yml and release.yml still warn -- release.yml is Captain Bob's to rule.
scripts/test_acceptance_signing.sh takes the step from acceptance.yml as CI
runs it and runs it with and without a key. Against the workflow as it was:
the unsigned case passed, no error -- 2 failures. Now: no key fails with an
::error::, a key is handed to the build as SIGN_KEY. logs/20261112-185739/.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
logs/20261112-185800/PROTOCOL.md, for #142, #173, #174 and the notes: the four
boots 538/538, word_count=318, dict_hash 0x8c92508f748f99cf on all three; the
new post-count stage (#173) on the real boots: all four at or above the floor,
all passed; test-32 39 programs, 1,869,223 checks, 0 failures; the proofs
(40 s); the images; the documents. The tree clean before and after.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob, 2026-11-12: "release MUST refuse unsigned builds". release.yml's
"The capsule-signing key" step warned when CAPSULE_SIGN_KEY was missing and the
release went on, unsigned. Now it is an ::error:: and the step fails, as the
gate into master's does (#174). build.yml and nightly.yml still warn.
scripts/test_acceptance_signing.sh now takes the step from both acceptance.yml
and release.yml as CI runs them. Against release.yml as it was: the release's
unsigned case passed, no error -- 2 failures (acceptance's 3 checks passed).
Now 6 checks, all pass. logs/20261112-205446/.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob, 2026-11-12: "Wire L2, keep L7 on". The table held the 128
cards of the 7-bit L1-L7 space; nothing read L7, and the rolling window
recorded whatever the card said, so the 128 were 32 behaviours.
- compudynamics.c: cd_tuning_word() sizes the table to 64 (L1-L6).
- ssm_jacquard.c: card_bits() makes a card's 7-bit mask, L7 set, where a
card is read, seeded and started (C4: 17 -> card 8);
ssm_l8_force_config() still takes the 7-bit index, its L7 bit implied.
- physics_execution_hooks.c, kernel vm_core.c: the rolling window records
only while L2 is on (or before there is a config).
- ssm_jacquard.h says what is wired, bit by bit (it said L7 was);
v3/CLAUDE.md and the L8 reference corrected in place; FABRIC-3 XLVIII.21.
Tests first, seen to fail against HEAD (logs/20261112-210510/kernel-tests):
test_ssm_jacquard 4 of 7, test_physics_l2 1 of 4; all pass now. v4 reads
no rolling window: acceptance is unchanged; the protocol runs next.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
#188 (the Jacquard selector) and 7e0728e4 (a release refuses an unsigned
build). POST 538/538 on all four boots, dict_hash 0x8c92508f748f99cf on the
three, test-32 39 programs 0 failures, proofs, images, documents.
logs/20261112-210531/PROTOCOL.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Plan 2026-10-20 (ENGINE step 4), Task A1, for #189 -- Captain Bob,
2026-11-12: the physics blocks alpha.5; order A, 1, 2, then 7.
- exec.h/exec.c: v4_call_hook, es->on_call/on_call_ctx; called at `call`
once the return address is pushed, before P moves. NULL: as before.
- heat.h/heat.c: the per-call-target table (4096 slots, no capsule word
counted), its freeze mask and v4_heat_call_* go (ENGINE.md step 3: "The
per-call-target count goes"); the kernel will keep a word's accounts.
- tests: test_call_hook_sees_every_call, seen to fail first (no on_call);
test_heat's per-call tests go with the table.
- the plan records today's rulings (order; HEAT-OP: both, a full array).
Suite (logs/20261112-225951/v4-tests): test, test-32, sanitize, 0
failures; hosted-check, the three alike. amd64 boot: POST 538/538,
dict_hash 0x8c92508f748f99cf, unchanged -- nothing sets the hook yet.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob, 2026-11-13: "I merge master in". PR 164 could not merge:
master's six comment-only commits of 2026-11-11 (751bfe26..268e5593,
"record runner.capacity: 1 decision ... (FABRIC-3, #140)") edit the
concurrency blocks the branch deletes for #140 ("It should be global").
Resolved in every workflow for the branch -- no concurrency group --
with master's fact kept as a line of the #140 comment: the capacity is
explicit in the runner's config.yaml, runner.capacity: 1. Each file
parses; scripts/test_acceptance_signing.sh passes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
#189 -- Captain Bob, 2026-11-12: "Add HEAT-OP. If it's a huge performance
hit, we'll revert it. If not ... it stands"; a full 33-cell array, as
DECOMPOSITION.md section 7 draws it; the Stadium feed is Phase 1's.
- node.h/node.c: v4_node_heat_regs_attach; a data fetch from base+k gives
opcode k's count, base+32 the anti-clock (the low cell at 32 bits); a
store changes nothing. Reset detaches them.
- host_map.h: HEAT_REG at word 13824, in what was free between the
dictionary space and SEND_TEXT; the compile-time check covers it.
Nothing moved. map-*.f18: HEAT-OP 13824, ANTICLOCK 13856.
- image.h/image.c/mkmap.c: the image carries the address; v4_image_born
attaches the node's own counts.
- system.f18: HEAT-OP@ ( op -- n ), op mod 32; ANTICLOCK@ ( -- n ).
word_count 318 -> 320 (hosted 317 -> 319): dict_hash changes.
- v4-dictionary.tsv: both words; DECOMPOSITION.md 7 records the build.
Also in this batch:
- Makefile: make init installs qemu-user -- hosted-check needs it
(Captain Bob: "Add to make init now").
- .claude/memory/rules.md, rule 5: stop and ask when something comes up
(Captain Bob: "you are supposed to stop and ask").
Tests first (logs/20261113-005047/heat-op): test_node did not compile;
test_host_quit 3 of 1337 failed. Suite: 199 programs, 7448554 checks, 0
failures; hosted-check, the three alike. Timing: no measurable cost.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
A1 (the hook at call), the merge of master (the six runner.capacity
comments) and HEAT-OP/ANTICLOCK (#189). POST 538/538 on the four boots;
word_count 320 and dict_hash 0xd148437fb39a1afd on all three
architectures (318 and 0x8c92508f748f99cf before: the two new words);
test-32 39 programs 0 failures; proofs; images; documents, the
dictionary covering all 320 live words. logs/20261113-005100/PROTOCOL.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob, 2026-11-13: "Create the issues then fix them before the long
acceptance test runs"; #16: "Nightly".
#190 -- make -C v4 post79 did not reproduce POST:
- mkpost.py runs v3 with address randomisation off (setarch -R): its
printed addresses (>IN, FIND) are the same each run, so POST79.md is.
- v4/Makefile post79 builds hosted v3 from this tree's source first and
uses it, not the lfs/ copy, which can be older.
- post_cases.c and POST79.md regenerated: 539 cases -- LOAD.rest_of_line,
which v3 gets through since it changed after 2026-10-07. Two runs the
same. POST_FLOOR 538 -> 539 (#173).
- test_host_unit.c takes the table's count, not 538 typed in the test.
#16 -- scripts/nightly.sh v4-checks: make -C v4 sanitize hosted-check,
logged under logs/<TS>/v4-checks/, refused without qemu-user; nightly.yml
runs it after publishing, before the proofs, installing qemu-user if the
image lacks it.
Evidence (logs/20261113-014927/post79, -024314): before, regenerated with
nothing changed: 539 cases and addresses differing run to run; after: the
same twice, hosted POST 539/539; test, test-32 99 programs 0 failures;
v4-checks 79 programs under the sanitizers 0 failures, the hosted builds
alike. The first verification failed at test_host_unit.c:442 (kept); a
stray v4-checks run stopped at once is kept, labelled (-014453).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
#190 (POST regenerated, 539 cases) and #16 (v4-checks nightly). POST
539/539 on the four boots, at the new floor; word_count 320 and dict_hash
0xcac4cf58ed493172 on all three architectures; test-32 39 programs 0
failures; proofs; images; documents, the dictionary covering all 320
live words. logs/20261113-031936/PROTOCOL.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Plan 2026-10-20 (ENGINE step 4), Task A2, for #189. Captain Bob,
2026-11-13: "The ten"; "Accept it".
- compile.f18 (COMPILE,): a word is copied in line only when it is flagged
in-line and compile-only (24) -- the ten that work on the return stack,
>R R> R@ I J LEAVE and the later UNLOOP 2>R 2R> 2R@. Every other word,
in-line ones too, is a call the hook at `call` sees.
- EXECUTE calls the xt: `call xt` (opcode 3, slot 0) and a ; in (EXEC),
run. INTERPRET runs a word by (GO): `call xt` and `jump (IL)` in
(RUNX), entered by a jump, at the old return-stack depth.
- host_map.h: (EXEC) BUF0_W-22, (RUNX) BUF0_W-115, from what was free.
- Its cost, accepted: one return entry at the bottom of a chain (30 deep
from the prompt, not 31) and one per explicit EXECUTE while its word
runs; the hosted boot +2.4% (2.917 s to 2.987 s), instructions +0.4%.
- Tests first (logs/20261113-045122/a2): inline_words_are_called,
residue_stays_inline, execute_passes_the_hook, rstack_kept_across_call,
abandoned_definition_not_called -- 3 of 1096 failed before. The depth
checks moved to the new figures; SEE shows the calls; two layout checks
use in-line test words; 2>R 2R@ 2R> checked in test_host_quit.
- DECOMPOSITION.md EXECUTE, the plan's A2 and rulings, FABRIC-3 XLVIII.22.
Suite: test, test-32 99 programs 0 failures; v4-checks 79 under the
sanitizers 0 failures, the hosted builds alike; hosted POST 539/539.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
A2 (#189): every word but the ten called. POST 539/539 on the four
boots; word_count 320 and dict_hash 0x47570d4ced6ce255 on all three
architectures; test-32 39 programs 0 failures; proofs; images; documents.
Boot times (stage, with its build) against be456d44's run: amd64 123 s ->
118 s, aarch64 208 -> 197, riscv64 96 -> 93 -- no slower under QEMU.
logs/20261113-053019/PROTOCOL.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
#189, ENGINE step 4, Phase 1 (plan 2026-11-13-engine-step4-phase1), Tasks
1-4. Captain Bob, 2026-11-13: Q1 "A new node key", Q2 "Per heartbeat
tick", Q3 "All of v3's", Q4 "In Phase 1" / "Full births now", who runs
next "Hera by compudynamics now" (Task 5, to come).
- stadium.c/.h: stadium_node_key(node) ("NODE" | number),
stadium_boot_init_for(first), stadium_birth_node(key, name);
stadium_boot_init() and stadium_birth_hera() are them for Hera's key, so
v3's path is as it was (only an impossible panic's words changed).
- kernel_main.c: the system is chosen and said first, then the Stadium,
sessions, switch-signal and kernel-Hermes tables for both systems in
v3's order; v4: Hera, node 10, patron zero under her node key, the word,
block and opcode layers. The quota self-test and the heartbeat's line
are functions now, called by both systems.
- sk_v4.c/.h: sk_v4_run() goes -- sk_v4_boot() returns, the kernel's
checks after POST (opcodes, words, conservation, quota, heartbeat), then
sk_v4_prompt(). sk_v4_charge_opcodes() at the heartbeat's service points
(the key wait, after each line) and once after POST.
- stadium_ops.c/.h: a node's 32 opcodes as Stadium patrons (HEAT-OP's
feed, "Both"): retirements charged as the word layer's dispatches, a
quantum each, the reservoir's floor kept, shared by retirements when
short; cooling per tick as words. ~0.5 us a charge on the host: kept.
- qemu_session.py: a v4 acceptance boot fails without "Stadium: N cells",
"Stadium conservation: CONSERVED" and the heartbeat's ticks (#189).
Tests first (logs/20261113-122142): the physics check failed on the v4
boot; test_stadium_node, test_stadium_ops (and its share case, after a
real boot showed ";" taking all the heat) failed before. Now: kernel host
tests all pass; amd64 v4 boot passes, POST 539/539, dict_hash unchanged,
"Stadium opcodes: 30 resident, hottest opcode 8", CONSERVED; a system=v3
boot prints its Stadium lines as on 2026-11-06.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
#189 Phase 1, Tasks 1-4: v4 in the kernel's boot. On the four boots: the
Stadium made ("Stadium: 83886 cells"), POST 539/539, "Stadium opcodes: 30
resident, hottest opcode 8", "Stadium conservation: CONSERVED", the quota
self-test and the heartbeat's ticks -- the new physics check passing;
word_count 320 and dict_hash 0x47570d4ced6ce255 on all three
architectures, unchanged; test-32 39 programs 0 failures; proofs; images;
documents. logs/20261113-125011/PROTOCOL.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
#189, ENGINE step 4 Phase 1, Task 5a. Captain Bob, 2026-11-13: "Full
births now"; "Hera as now, in a fabric".
- sk_v4.c: the kernel's fabric (v4/fabric.h), Hera's node at its one
place, entered in its registry (v4/manage.h) as node 10. The registry's
answers are real: a birth is "no room" -- the fabric has one place
until 5b; CAPSULE-OPEN/-LINE read the kernel's capsule directory,
checked as the boot checks (no capsule of F18 code is there until 5b);
a born node's parity line. She is not told her number: the boot's
console addresses node 0 (told 10, the boot stopped after POST -- kept,
labelled, logs/20261113-132712).
- boot.h/boot.c: v4_boot's `request` hook (the registry's requests) and
`capsules` (the kernel's own, loaded before the seal, so POST runs on
them). hosted.c sets neither: the hosted product's boot is as it was.
- kernel/Makefile: manage.c in the kernel's v4 engine.
- qemu_session.py: a v4 acceptance boot asks NODE-ME . and wants 0.
- v4-dictionary.tsv: the 55 words of centre.4th and hera.4th, written
from their source and manage.h; mkdictionary.py loads the kernel's
capsules into the hosted system before WORDS, so the live list is the
kernel's -- 375 words.
- .claude/memory/rules.md, rule 6: every boot command under timeout 570.
Seen to fail first (logs/20261113-132410): NODE-ME was an unknown word.
After: amd64 NODE-ME . gives 0 ok; POST 539/539; word_count 375. v4
suite: test, test-32 99 programs 0 failures; v4-checks 79 under the
sanitizers 0 failures, the hosted builds alike.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
#189 Phase 1 Task 5a: Hera in the kernel's fabric and registry. On the
four boots: NODE-ME . gives 0 ok, the physics lines as before, POST
539/539; word_count 375 and dict_hash 0x06531687cc075aee on all three
architectures (centre.4th and hera.4th loaded); test-32 39 programs 0
failures; proofs; images; documents, the dictionary covering all 375
live words. Every boot under timeout 570. logs/20261113-142432/PROTOCOL.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
ENGINE step 4 Phase 1, Tasks 5b-5d, and the edge cases (Captain Bob, 2026-11-13: "Build it to proper
spec"; "Check for the edge cases"; "Yes, add it to v4-checks").
- v4/system/mesh.c, v4/include/v4/mesh.h: the nucleus assembled by the ROM and given to Hera as F18
cells; Hera born empty at place 0, the kernel on port 0, the console on port 1; told she is 10, in the
registry, the kernel's words, FORTH-79, sealed, POST, the kernel's capsules. NODE-BORN / NODE-KILL in
the owner's room.
- No scheduler: v4_mesh_tell returns once the node told says how its text ended, and
v4_mesh_step / v4_mesh_busy let the host step the fabric while it waits, so nodes run between
lines as during them. A text left "waiting" is owed its end: the next line ends it "Interrupted"
and is then run (MESH.md 7b.6).
- kernel: sk_v4.c boots through the mesh; its idle steps the fabric while a node can run, and halts
only when none can (#151 kept). stadium_return_quota and stadium_ops_forget on a kill.
- Tests: v4/tests/test_mesh_boot.c (32 checks: the boot, births, kills, rebirth, a live number, a
fault, no such node, Hera not killing herself, no room, a line too long, a node running for ever,
a wait broken by the next line), run by `make -C v4 test` and under ASan+UBSan by `sanitize`
(v4-checks); kernel/tests quota return; scripts/qemu_session.py births() and the edges on bare
metal.
- Docs: ENGINE.md 6 "who runs next" struck (settled: no scheduler); the Phase 1 plan; FABRIC-3
XLVIII.23; .claude/memory/rules.md: alpha.5 closed to new work.
- Evidence: v4-checks 80 programs under the sanitizers, 0 failures; the hosted builds alike; two
stopped v4-checks runs labelled STOPPED; three development amd64 boots labelled.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
All eight stages passed: the four boots (POST 539/539, word_count=375, dict_hash=0xbbf72b1f752a1a51 on
all three architectures, the edge cases on amd64), test-32 (39 programs, 0 failures), the proofs, the
images, the documents. logs/20261113-170011/PROTOCOL.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
ENGINE step 4 Phase 2 (plan docs/v4.0.0/plans/2026-11-13-engine-step4-phase2-word-accounts.md). Rulings,
Captain Bob 2026-11-13: "ID field in the header"; "Per heartbeat tick"; the ACL "All of it in Phase 4".
- The word ID, at xt-4 before the flags (ENGINE.md 3b). The system's words are numbered as they are
assembled, by a new directive, #id, (and #ids, the count), in the C assembler (text.c) and in the
machine ROM's (capsules/v4/rom), which counts in the system's own (NEXT-ID). (HEADER) numbers later
words; FORGET gives back the IDs it frees, always the newest; COLD resets to (BOOT-ID); (SEAL) records
it. All 313 system headers carry the cell.
- The kernel's records (v4/mesh.h): (TELL-DEFINED) and (TELL-FORGOTTEN), requests -19/-20 on port 0,
which the kernel puts in (WORD-DEFINED)/(WORD-FORGOTTEN) when it seals a node, and reads the system's
own words once. A bit per address marks a word's xt (calls into the ROM or headerless code are not
words); at call, the bit test and a count. v4_mesh_charge hands the owner the calls since.
- The Stadium: stadium_words_charge -- N calls as N of stadium_word_dispatch's dispatches, shared by calls
when the reservoir is short; sk_v4.c charges every node's words at the tick beside its opcodes, and
stadium_word_forget before an ID is used again.
- Tests first: test_text.c, test_host_dict.c, test_host_quit.c (numbering, FORGET, COLD),
test_mesh_boot.c (44 checks), kernel/tests/test_stadium_words.c; qemu_session.py: Hera's words
admitted after POST.
- amd64: Stadium words: promotions=141 after POST, CONSERVED; POST 539/539; word_count=377;
dict_hash 0xa854ba15b437d288 (the new cell); the boot 142 s against Phase 1's 153 s.
- v4 test, test-32: all pass; v4-checks: 80 programs under the sanitizers, 0 failures; the hosted
builds alike. docs/reference/v4-dictionary.tsv: the two new words; FABRIC-3 XLVIII.24.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
All eight stages passed: the four boots (POST 539/539, word_count=377, dict_hash=0xa854ba15b437d288 on all
three architectures, Hera's words admitted to the Stadium after POST on each), test-32 (39 programs, 0
failures), the proofs, the images, the documents. logs/20261113-191247/PROTOCOL.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Rulings, Captain Bob 2026-11-13: the per-call loops "Every call, as v3"; the loops "Rewrite for v4"; Phase 7
accepted on "v3's whole POST", which brings Phase 3's words and Phase 4's ACL in as tasks of it; Mama's
20 cases as v4 words with v3's names -- "do not compose words in C99, use F18 and StarForth"; the C
modules (ACL, break_me) rewritten for v4.
docs/v4.0.0/POST-PARITY.md (Task 0): v3's POST by module against v4 on e8e7ec8b -- 888 cases in the 17
modules mkpost.py reads (550 Required, 317 for words v4 already has, 21 for words it lacks), 146 in the
nine it does not, 12 C tests; about 1,050 in all, 539 on v4 today.
Not pushed: no protocol run covers it yet (rules: batch docs with the next protocol).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
v4_mesh_account gains v3's DictPhysics: heat (execution_heat), last_decay_tick, temperature_q8, and the
kernel's own state flags (pinned, frozen); v4_mesh_physics gives each node v3's VM fields -- its tick,
the decay slope (v3's initial 21845, Q48.16), L8's gate on Loop #3. v4_mesh_state_flags derives a word's
flags as v3 does, immediate and hidden read from the node's header when asked. Test first:
test_mesh_boot.c, 48 checks. The kernel gives each of its 16 places its physics. Not pushed (held,
Captain Bob 2026-11-13: "Hold it, go ahead with Task 1").
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
As v3's physics_pre_execute, in its order (Captain Bob, 2026-11-13: "Every call, as v3"): a word's heat
decays by the ticks since its last touch at the node's decay slope (Q48.16, v3's arithmetic), unless
frozen or L8 has Loop #3 off; +1; its temperature smoothed 3:1; its ID into the node's Rolling Window of
Truth (4096, warm at 1024, a diversity check due every 512). The kernel's heartbeat is read at each call.
Pipelining and the hot-words cache ruled "Built, off as v3" (Task 2b). Test first: test_mesh_boot.c, 56
checks. amd64 boots, RESULT: PASS, 146 s against Phase 2's 142 s. Not pushed (held).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Rulings, Captain Bob 2026-11-13: "Build a hot-words cache"; then, v3 being found to have both off
(ENABLE_PIPELINING, ENABLE_HOTWORDS_CACHE: n in Kconfig.physics and kernel/Makefile), "Built, off as v3".
- The kernel, at every call (mesh.c): v3's hit check, the transition, the likeliest next word (kept as each
count rises; v3 scans 1024 a call -- the same word but in a tie v3's Q48.16 rounding makes), speculation
at v3's thresholds into the node's cache; a word past heat 50 promoted at its call (v3 promotes at a
lookup; v4's heat is the kernel's). Counters from the owner's heap, given back when a word goes.
- The node (dict.f18): (LOOKUP) tries (HOT)'s xts first, in F18, calling only C@; the kernel keeps it true
(entries that go, or whose name is defined again, taken out). Empty unless the build has the cache.
- Tests: test_mesh_boot.c as built (57 checks) and with both on (mesh-boot-on, 65). v4 test: all pass.
- Reported, v3's own: when full, its promote takes the next slot without looking for the word there (a few
hot words churn all 32 slots); its prefetch misses are never counted. Both reproduced.
- amd64 boots, RESULT: PASS, but 174 s and 181 s against 2a's 146 s, while on the host the mesh boot is no
slower: being measured. Not pushed (held to the Phase 7 boundary).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Commit b880a6ea (Task 2a) checked out detached and booted on amd64: 175 s, as slow as Task 2b's 174 s and
181 s, against 2a's 146 s an hour before. 2b has no measurable cost; on the host the mesh boot was no
slower either. Not pushed (held to the Phase 7 boundary).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob, 2026-11-13: "By execution, as v3". v3's kernel runs a VM's heartbeat cycle every 256 word
executions (HEARTBEAT_CHECK_FREQUENCY, vm_core.c:912 and :1210), and its tick_count is the count of cycles.
Task 2a had used the kernel's timer heartbeat as the decay clock: corrected. Each node counts its word
calls; every 256th runs its heartbeat cycle at the call; the cycle's loops follow. The kernel no longer
gives the mesh its timer. Test first: test_mesh_boot.c, 59 checks (67 with pipelining and the cache on).
Not pushed (held to the end of Phase 7).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
In v3's order (vm_heartbeat_run_cycle): the tick; background decay -- 64 words a cycle from a cursor down
the dictionary (v3's vm_tick_apply_background_decay), so a word not called cools; the window's adaptive
pass on its diversity (v3's rolling_window_run_adaptive_pass: halve to 256, double to 4096). Test first:
test_mesh_boot.c, 62 checks (70 with pipelining and the cache on). Inference, L8 and the snapshot next.
Not pushed (held to the end of Phase 7).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob, 2026-11-13: "The 256 should probably be either a kconfig knob or a feedback loop for RWOT".
It is already Kconfig.heartbeat's HEARTBEAT_CHECK_FREQUENCY (default 256, v3's); the mesh now reads it.
Set in the config or on make's line, the kernel passes it to every object (checked: 128 reaches the
compile). A feedback loop from the window would be a design change, not v3's: not made. Not pushed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
v3's vm_heartbeat_update_l8 with its table (ssm_jacquard.c, compudynamics.c), rewritten in integers: the
regime from the window's entropy, the prediction CV and the locality of the last 32 calls; the current card
scored (ANOVA from inference, joint convergence under L5/L6), decayed 0.99, the next chosen by UCB1 over 64
cards; its gates L1-L7 applied to the loops, L1 and L4 within the build's ceilings. Every gate off and card 8
at the start, as v3's kernel boot. Applied through card_bits: v3 applies the card's number, one bit off --
filed #196, ruled "v4 correct, v3 as is". v4_mesh_l8_force as v3's ssm_l8_force_config; the exact decay
and window tests hold a card. Test first: test_mesh_boot.c 67 checks, 75 on. amd64 PASS, 156 s.
Not pushed (held to the end of Phase 7).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
v4/system/inference.c: v3's inference_engine.c and its Q48.16 arithmetic ported to the bit, no C library --
the variance, the window width (Levene), the decay slope (log-linear regression), the run with its 5%
early exit. v4/tests/test_inference.c links v3's own and compares them: POST's fixed cases, 20000 random
multiply/log pairs, 400 random trajectories (make -C v4 inference; in make -C v4 test). In each node's
cycle every 1000 ticks (HEARTBEAT_INFERENCE_FREQUENCY), once the window is warm: the trajectory from the
window's words' heat, the width and slope applied unless it exits early, then Loop #7 -- the heartrate
slower when stable, faster when not, within 1/4..4x; Hera's re-arms the kernel's heartbeat
(heartbeat_set_adaptive_period_ns), as v3's Hera. L8 reads its outcome. Test first: test_mesh_boot.c 72
checks, 80 on. amd64 PASS, 163 s. Not pushed (held to the end of Phase 7).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Each cycle publishes v3's HeartbeatSnapshot: the tick, the window's width and fill, the decay slope, and the
dictionary's heat as inference last counted it (words over 50, cooling, the heat in all). One copy, not
v3's two, which are for its heartbeat thread. With it every loop v3 runs before its prompt runs on v4's
word records on every node. Test first: test_mesh_boot.c 73 checks, 81 on. Not pushed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob, 2026-11-14: "we're supposed to be working on f18 opcode as well as our composed f18
dictionary"; ruled every loop on both, "and an easy revert if opcodes are too granular".
- The engine: a hook at retire (exec.h on_retire), as Part A's at call; NULL unless set. test_exec.c.
- The mesh: 32 opcode records per node, an opcode window, opcode transitions under L4; at every retired
instruction decay at the opcodes' own slope, heat, temperature, the window, the transition; in each cycle
all 32 cooled, the opcode window's service, inference on its heat. L8 scores from the words and its gates
govern both; the tick stays paced by word calls. The window is now one type for words and opcodes.
- The revert: OPCODE_PHYSICS (Kconfig.physics, default y; the kernel's flags) -- 0 and the opcodes are
counted and charged to the Stadium as in Phase 1. make -C v4 mesh-boot-noop builds and tests it.
- Tests first: test_mesh_boot.c 77 checks as built, 86 with pipelining on, 74 with the revert.
- Measured on amd64, back to back: 219 s on, 196 s off (+12% under TCG). Both boots' logs labelled.
Not pushed (held to the end of Phase 7).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
capsules/v4/physics.4th, a kernel capsule on Hera (blocks 8200-8201): FREEZE-WORD UNFREEZE-WORD FROZEN?
HEAT! HEAT@ DECAY-RATE@ Q.VARIANCE INFER-DECAY-SLOPE INFER-WINDOW-WIDTH INFER-RUN INFER-WINDOW@ INFER-DECAY@
INFER-VARIANCE@ INFER-FIT@ INFER-EARLY-EXIT@ -- each a KERNEL-WORD, requests -21..-34 (v4/mesh.h), served by
the kernel on any node: "The compudynamics is stadium/fabric" (Captain Bob, 2026-11-14). A word is named as
v3's, the newest visible entry of CONTEXT then FORTH (the image gains CONTEXT's address); the arithmetic is
v4/system/inference.c, v3's to the bit. Q48.16 results are v4's doubles (low cell, then high). HEAT! starts
the word's decay from now. Test first with v3's POST cases in v4's address units: test_mesh_boot.c 86
checks, 95 on, 83 without opcode physics. docs/reference/v4-dictionary.tsv: the 15 words (392 live).
amd64 PASS, word_count=392. Not pushed (held to the end of Phase 7).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob, 2026-11-14: "v3" ... "we can always change it late". v3 counts an execution per token its
(C) interpreter takes and per word its definitions run. Measured on 8 lines, counting every call on v4 gave
~700 to v3's one for typed lines (the F18 interpreter's own calls) but 1:1 in compiled code, so no single
HEARTBEAT_CHECK_FREQUENCY could match v3. Now INTERPRET counts its tokens in (TOKENS) -- one F18 store, the
map gains the cell -- and the kernel adds them at each call; a call counts, and gets physics, only from
dictionary space or as INTERPRET's or EXECUTE's dispatch (the image gains their addresses). The knob stays
at v3's 256. test_inference.c out of the per-width tests (its own rule). Tests first: test_mesh_boot.c 90,
99 on, 87 without opcode physics; v4 test all pass. amd64 PASS, 181 s, dict_hash 0x5c53b2e18343a9b6.
Not pushed (held to the end of Phase 7).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
capsules/v4/physics.4th, blocks 8202-8204: ENTROPY@ ENTROPY! RESET-ENTROPY as kernel requests -35..-37 (a
word's heat by the address ' leaves, a data word's too; an error for what is not a word, as v3's);
TOP-WORDS and WORD-ENTROPY in StarForth, printing in v3's form what the kernel gives them through (HOT#)
(HOT) (XT-HEAT), -38..-40 (the i-th hottest in v3's stable order); (- in StarForth, skipping to the
matching close paren with nesting, as v3's code does. Tests first: test_mesh_boot.c 101 checks, 110 on, 98
without opcode physics. Listing: 403 live words. amd64 PASS, word_count=403. Not pushed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
v4/tools/mkpost.py --all --out FILE: every case of v3's whole POST (its 22 POST modules in test_runner.c's
order, Q48, stress, integration), v3 the reference, without touching system/post_cases.c or its document.
Run on Hera in the kernel's fabric after Phase 7 Task 4: 1020 cases, 849 pass, 171 fail -- by word and kind
in docs/v4.0.0/POST-PARITY.md section 5. Not pushed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob, 2026-11-14: where v3 does not do what its own table says, v4 is held to the table; where a
case's output names v3's internals, to the same form with v4's values; both listed. POST-PARITY.md, the
Phase 7 plan's Task 8. Not pushed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob, 2026-11-14: FORTH-79 where it defines the word, FORTH-83 for words 79 lacks, v3 (its table where
it contradicts it) only for its own extensions; each rewritten case listed with its reason. POST-PARITY.md.
Not pushed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob, 2026-11-14: the standard first, in the order 79, 83, ANS, then v3; the texts in the repository
as references. FORTH-79 and FORTH-83 from the University of Vienna's FORTH archive (complang.tuwien.ac.at),
ANS Forth's final draft (dpANS) from Taygeta, as fetched, with their sources in
docs/reference/standards/README.md. Not pushed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob, 2026-11-14: "Yes, FORTH-83's layout". FORTH83.TXT, "number": "The address of the least
significant 16 bits is two greater than the address of the most significant 16 bits." v4 stored the low
cell first, as v3. dict.f18 (2,) and words.f18 (2@ 2!); test_host_dict.c and test_host_quit.c to the new
layout. POST 539/539; v4 test-64/32 dict and quit pass. Not pushed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
v3's whole POST on v4 at 983 of 1010 (the table's own count with its drops), from 850.
What remains is capacity ("Dictionary full", the 100-900-deep stacks): next, as ruled.
Found and fixed under the rulings (docs/v4.0.0/POST-PARITY.md):
- IS and DEFER@ refuse a word DEFER did not make, as v3: error 22, "Not deferred".
v4's IS stored into any word -- POST's `42 IS DUP` broke DUP for every case after it.
DEFER marks its word with flag 128 (dict.f18).
- SMUDGE and HIDDEN are compile-only, as v3's table: at the prompt v4's hid the newest word.
- Q./ by zero gives 0 and no error (ruled: v3's table); error 11 retired. The ROM's,
the test's reference and its hand-assembled copy changed alike.
- FENCE is v3's: it protects what is defined so far. The variable is (FENCE).
- The line's length is (#SRC), no longer SPAN: an EXPECT in a line cut the rest of
the line off, and 0 EXPECT could not leave SPAN 0 (FORTH-83).
- POST gains V4_POST_HEAD: the values left and how the output begins -- the form of a
printout whose values are the run's own (TOP-WORDS, WORD-ENTROPY).
post_all_rules.py: the rest of the whole table where it is not taken as v3 ran it --
v4's units for the cases that wrote through HERE as a byte address (CMOVE>, BLANK,
ERASE, FILL wrote into the system's code), the standard's results (COMPARE, SEARCH,
BASE output, THRU, >BODY with FIND's address), v3's descriptions where v3 printed
otherwise, and the cases dropped with their reasons (SP@/SP! by D-2, DABS/DMIN
underflow, CMOVE>.bounds, THRU.negative_end).
make -C v4 test: all v4 tests passed (both widths, mesh-boot, inference).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Phase 7, Task 8: PARITY:V4_POST tests=990 pass=990 fail=0 on Hera (mesh-boot); the hosted
product 945 of them, all passed -- the 45 cases of physics.4th's words need the kernel.
- Enlarged, as ruled ("Enlarge v4 to fit"): a node is 32768 words, its data stack 1024
deep (v3's), host and kernel alike; the dictionary's end moves with them.
- POST's runner compares every value a case leaves (v3's leave 101); V4_POST_KERNEL marks
a case of the kernel's words, and a host without a kernel does not run it.
- mkpost.py --all is POST's table (make -C v4 post79), with docs/v4.0.0/POST.md; it reads a
module's #defines (v3's Mama cases are MAMA_IMPLEMENTED, 0: v3's own POST skips all 20);
it cuts no line between a word and the name it takes.
- Rules for the last of the table: the min/max and array-sum programs written as meant,
CREATE's 26-character name (FORTH-79: 31), many_defs.
- POST_FLOOR 990.
- Tests sized for a 32-deep stack and a 16384-word node made deep-stack-ready
(test_host_depth, test_host_quit, test_host_node, hosted-check, depthsweep.py).
make -C v4 test: all v4 tests passed. make -C kernel/tests test: all passed.
depthsweep at 1024: 230 runs, 0 failures. v4-checks running.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Ruled (Captain Bob, 2026-11-14): "Find the TCG cost first". v3's whole POST took the amd64
accept from about 3 m 15 s to 8 m 45 s, and aarch64 past its 570 s (logs/20261114-091651,
STOPPED, committed). Profiled on the host (gprof, mesh-boot): 87% of the time in POST's
lines, about 280,000 opcodes a line -- the interpreter's dictionary lookup.
- POST puts the dictionary back before each of v3's suites, as v3's run_test_suite does
("to remove test-created words"): a mark, (POST), forgotten and made again
(V4_POST_SUITE). Only the dictionary: SCR, BASE and the rest stay as the cases left them.
- (LOOKUP) tries each entry by its count and first character first -- one fetch of the
name's first cell, masked -- and only then its hidden flag and the whole name. The answer
is the same; an entry that cannot be it costs a few opcodes, not some forty.
mesh-boot: 37.7 s -> 15.7 s, POST 990 of 990. make -C v4 test: all v4 tests passed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob's "Find the TCG cost first", continued. aarch64 still ran out of its 570 s at 281972ab
(logs/20261114-133155, STOPPED); amd64 passed in 253 s (logs/20261114-132732).
- (LOOKUP): the count-and-first-character key in register B while the entries are walked, and the
xt kept on the stack until an entry passes it -- about 19 opcodes for an entry that cannot be the
word, from 28. The answer is the same.
- POST: a suite begins with one line, DECIMAL FORTH DEFINITIONS FORGET (POST); its mark is made by
the state line of its first case run (a skipped kernel case passes it on).
mesh-boot: 15.7 s -> 12.2 s (37.7 s before this work), POST 990 of 990. make -C v4 test: all passed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob's CI directive (2026-11-17) and the design he ruled point by point
(docs/v4.0.0/plans/2026-11-17-emergency-diagnostic-build.md).
- v4/system/dump.c, v4/include/v4/dump.h: what a node was doing when it stopped -- the fault's kind,
address and count; P A B; the data stack, T first; the return stack, R first, each entry named by
the word it falls in (the backtrace; ROM and below-every-word said so); the running line. Bounded
dictionary walk; no C library. Ruled: every build, on a stop only (a handled fault keeps its line).
- kernel/src/v4/sk_v4.c: the dump before each of the three halts (not started; POST's blocks not
let go; a node stopped on a fault) -- Hera and any stopped node. kernel/Makefile links dump.o.
- scripts/nightly.sh diagnose [<stage>]: after a failed stage, that configuration built again with
EMERGENCY_CONSOLE_ENABLED=1 -- build only, nothing booted (ruled) -- its artifacts kept in
logs/<TS>/diagnose/ only from a build that passed, with a README saying how to launch it with
--log-level=debug. Always exits 0: evidence, never a result. The protocol runs it after a failed
stage and still exits with that stage's code; acceptance.yml, nightly.yml and release.yml run it
`if: failure()`, before the logs are kept. Any failed stage triggers it (ruled).
- capsules/BLOCK_MAP.md: as the build writes it -- the v4:physics.4th row, missing since 28ffed8f
(ruled: commit the regenerated map).
- Test first: test_dump.c failed with no dump.c (logs/20261117-dump), then 15 checks at both widths
and under ASan+UBSan. The diagnose stage: a real EMERGENCY amd64 build; a failing build exits 0 and
keeps no artifacts (a defect found and fixed this way); the protocol path keeps the stage's exit.
- Also the write-up of the regression at 046ce0be, for ruling.
Not yet: acceptance on a clean tree (the next run); the kernel's stop wiring is compile-tested only
-- no boot reaches a stop on purpose.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Under set -euo pipefail its lookups of PARITY:V4_POST / PARITY:V4_SYSTEM in a boot log that has
neither -- an aarch64 boot out of its 570 s in POST, 2026-11-17 -- ended the script silently (exit 1),
so a failing protocol had no report; so did a missing test-32.txt. The lookups are guarded as
dict-hash and post-count already guard theirs: the report is written, Result: FAIL, "none" for the
missing lines, "not run" for test-32 (ruled by Captain Bob: "Fix it, test first").
- scripts/test_nightly_protocol_report.sh (new): seen to fail first (7 of 7,
logs/20261117-dump/test_protocol_report-before.log), then 7 of 7 pass; dict-hash's and post-count's
tests still pass. NIGHTLY_PROTO, the protocol's scratch, settable for it as NIGHTLY_OUT is.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob, 2026-11-17: "ONE ADAPTIVE HEARTBEAT. EVERY VM USES THAT ONE CLOCK ... THEY ADVANCE ONE TICK!"
Ruled: the beat is the fabric's steps; HEARTBEAT_CHECK_FREQUENCY steps a beat; the Stadium charged on it;
Loop #7 adapts the beat only; the heartbeat first, then mechanism A. Nothing built yet.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob, 2026-11-17: "The adaptive heartbeat is its own thing ... It has its own independent RWOT and slope
provider. Only one output from the entire stadium is used as a load measurement for heart rate pacing. ... ONE
AS IN THE NUMBER 1 heartbeat (the adaptive one) runs the entire system." Ruled one at a time: the kernel's
heartbeat; resident heat its one input; held in POST; the Stadium charged as before today
(docs/v4.0.0/plans/2026-11-17-one-adaptive-heartbeat.md, its first version kept as superseded).
- kernel/src/heartbeat.c, timer.h: heartbeat_pace_from(load) -- its own window of samples, its own slope
provider (v4's inference arithmetic) every HEARTBEAT_PACE_EVERY samples, Loop #7 setting its own period
(not stable 20% faster, stable 25% slower, 1/4..4x the 10 ms base). No source (v3's path): nothing runs.
- kernel/src/vm/stadium.c: stadium_resident_heat_total(), read only -- the one output it paces by.
- v4/system/mesh.c, mesh.h: the owner's clock read at every fabric step; each tick delivered, one by one, to
every node present -- its heartbeat cycle run, active, idle or blocked alike. The tick by execution, the
nodes' own Loop #7 and the heartrate callback are gone.
- kernel/src/v4/sk_v4.c: the clock gives the heartbeat's bottom half its turn; 0 until POST has ended.
- Tests first, each seen to fail (does not build): kernel/tests/test_heartbeat.c (11 checks),
test_stadium_node.c (16), test_mesh_boot.c's lockstep checks. Seven Phase 7 checks that assumed the tick by
execution now drive the test's heartbeat (ruled). POST 990/990 with v3's heat cases as written;
mesh-boot 103, -on 112, -noop 100, san-mesh-boot 103, every v4 program at both widths, every kernel host
test: 0 failures. The full `sanitize` target not run (it outlasts the foreground limit).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
FABRIC-3 XLVIII.25 covers the day's work:
- the regression found at 046ce0be (the fabric step walking every place);
- the CI directive and its rulings;
- the fault dump, the diagnose stage and the protocol-report fix;
- the one adaptive heartbeat;
- the protocol on cc09a88b (FAIL: aarch64 stopped in POST);
- next, mechanism A.
XLVIII.24 is corrected in place: its "no measurable cost" was measured
against an already-slow baseline.
The Phase 7 plan's Task 3 tick by execution is marked superseded.
ENGINE.md section 6 records the tick as settled.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
The direct loop before 046ce0be was architecturally incorrect. The fabric
wiring removed it and exposed the slowdowns, and the instruments (ASan,
UBSan, perf) locate them.
FABRIC-3 XLVIII.24 and .25 and both 11-17 plans are corrected in place,
and every measurement is kept. Mechanism A is the first speed-up found,
not a fix.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
The fabric keeps its occupied places as a list in ascending order, and
the step's nine passes walk that list in the same order as before. A
place removed mid-walk keeps its link, so the walk goes on to the next
node.
Test first: test_fabric_walk.c, 13 checks.
Every existing fabric and mesh test passes with the same checks, and so
do test-32 and ASan/UBSan. The host mesh boot runs POST 990/990. Over
five alternating runs its median falls from 11.695 s to 7.999 s.
Logs are in logs/20261117-mechanism-a.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Every stage passes. On all four boots POST is 990/990, with one
dict_hash (0x627f392ccbaa96f1). aarch64 is inside its 570 s; before
mechanism A it stopped in POST. test-32, the proofs, the images and the
docs pass.
The docs stage runs longer than one foreground command; it ran in the
background at Captain Bob's ruling.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
The kernel boots the fabric with Hera at place 0. Nodes are born and
killed through her. The fabric is lockstep. Each step walks only the
places that hold a node, and every node ticks on the one adaptive
heartbeat.
Ruled by Captain Bob.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Every stage passes. On all four boots POST is 990/990, with one
dict_hash (0x627f392ccbaa96f1). test-32, the proofs, the images and the
docs pass. The docs stage ran in the background at Captain Bob's ruling.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob ruled that MACHINE.md stands. The kernel's system capsules
are now starshipos.4th, centre.4th, hera.4th and physics.4th, after
forth79.4th.
COLD, after POST, leaves FORTH alone in the search order, so the
prompt reaches STARSHIPOS's words by naming the vocabulary. The ruling
is that init.4th will set the search order; the capsule's header now
says so.
Test first in test_mesh_boot.c: its parity line, and `STARSHIPOS S" ok"
DOE`. It failed before the change. 105 checks now pass, POST is
990/990, and ASan/UBSan and the amd64 build are clean.
Logs are in logs/20261118-starshipos.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Every boot passed: POST 990/990 on all four, dict_hash
0x3172c04e863199f4 on the three ISAs. test-32, the proofs and the
images passed.
The docs stage was stopped. The protocol is superseded because a latent
fault was found: COLD drops STARSHIPOS from the vocabulary chain
(VOC-LINK 0). Captain Bob ruled: fix first, then the protocol.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
COLD set VOC-LINK to 0, assuming the sealed system has one vocabulary.
With starshipos.4th sealed in, STARSHIPOS fell out of the chain after
COLD. A word defined in it then forgotten, or COLDed away, hung the
node. ACL-INIT-PRIMITIVES would also have skipped its words.
FORGET's loop is now (PRUNE-ALL). It drops the vocabularies above (Q)
and prunes the lists of those that stay. FORGET calls it as before, and
COLD calls it with the sealed DP. No new cell and no new word.
Test first in test_mesh_boot.c: FORGET and COLD of a word in
STARSHIPOS, and STARSHIPOS's own words staying. The before-run hung and
was stopped at 500 s. After: 109 checks pass, POST is 990/990, all 52
64-bit tests and test-32 pass, ASan/UBSan are clean, and the amd64
build is clean.
Found by Captain Bob's "make sure we're not applying incorrect logic".
Ruled: fix first, then the protocol. Logs are in
logs/20261118-cold-vocabularies.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
All seven other stages passed: POST 990/990 on all four boots, one
dict_hash (0xf4e9fdc94f6bfc77, word_count=404), test-32, the proofs and
the images.
aarch64 went past 590 s twice in the foreground, killed at the BIRTH
step. One background run, at Captain Bob's ruling, passed in 429 s with
the birth's parity recorded.
The docs stage failed because the dictionary gate found no entry for
STARSHIPOS. The entry comes next, then the protocol again.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Every stage passes. On all four boots POST is 990/990, with one
dict_hash (0xf4e9fdc94f6bfc77, word_count=404). test-32, the proofs,
the images and the docs pass. accept-aarch64 (401 s) and docs ran in
the background, alone, at Captain Bob's rulings.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Every stage passes. On all four boots POST is 990/990, with one
dict_hash (0xf4e9fdc94f6bfc77). test-32, the proofs, the images and
the docs pass. accept-aarch64 (397 s) and docs ran in the background,
each alone.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob ruled that the offset is the RAM block count, not a literal
2048.
- v3/include/blk_layout.h defines it, 2048 by default. block_subsystem.h
derives BLK_RAM_BLOCKS (2080) and BLK_DISK_START from it.
- Kconfig.kernel has the symbol. kernel/Makefile passes it to the
kernel's C code and builds mkcapsule and mkartemis with it.
- mkcapsule's two literal 2048s are now the knob.
The default value is unchanged, so the layout and the media are as before.
Test first: scripts/test_capsule_ram_blocks.sh, 8 checks.
- The before-run failed 4: the knob did not exist and mkcapsule ignored
it. Its three Makefile checks also hit a SIGPIPE flaw in the test,
fixed before the after-run.
- After: all 8 pass.
The amd64 build and mkartemis build clean. All 40 capsules lint. The
mesh boot passes 109 checks with POST 990/990. test_host_blocks,
test_host_quit, test_host_unit and test-32 pass.
Logs are in logs/20261118-ram-blocks-knob.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob ruled "Overwrite. That's it. Deterministic and final", within
a capsule and across capsules, "consistent with VOCABULARY FORGET & FENCE
philosophy, not implementation". Of what only prevented collisions:
"All of it goes".
- The cross-capsule collision gate is gone.
- tools/capsule-reserved.txt and its check are gone.
- The interactive resolver and its claim registry are gone.
- BLOCK_MAP.md says "Blocks used", and marks a shared number "the later
write overwrites". It is a convenience, not a gate.
- An unknown option is refused with the usage. A removed mode was
otherwise taken for the capsules directory, with its next argument as
the output file.
Test first: scripts/test_mkcapsule_overwrite.sh, 9 checks.
- The before-run failed 7. A repeat within one capsule already built,
since there was never a check for it.
- After: all 9 pass.
step 1's test still passes, the amd64 build is clean, all 40 capsules
lint, and the mesh boot passes 109 checks with POST 990/990.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob ruled: Block n lands at n-R; the lowest block is
interpreted; chaining is the developer's; together, v4 first.
- v4_blocks_place_capsule copies each Block n into RAM block n - R as 16
lines of 64, space-padded, in any file order. A later block of the
same number overwrites. It refuses a block below R.
- The mesh's load(), its capsule reader (CAPSULE-OPEN/(LINES), so a
child's forth79.4th via BIRTH too), boot.c and test_host_unit's model
each place a capsule, then interpret "<lowest> LOAD".
- mkdictionary.py skips --> when it types capsules at the hosted prompt.
- The v4 capsules are renumbered into R..2R-1, landing at RAM 1900 and
up, away from POST's blocks 0-5. Each is chained with -->, and
centre.4th's gap at 8103 is closed.
Test first:
- test_host_blocks.c gained 6 placement checks; it failed to build
before.
- test_mesh_boot.c checks forth79's lowest block in RAM block 1900; it
failed before.
After: POST is 990/990; all 52 tests at 64 bits, test-32, hosted-check
and ASan/UBSan pass; the amd64 kernel builds; the dictionary has 404
words.
Logs are in logs/20261118-capsule-place.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
This covers capsule blocks steps 1-4 for v4. Every stage passes.
- POST is 990/990 on all four boots.
- One dict_hash on the three ISAs: 0x847bd8ef42f53e14 (new, from
loading capsules through blocks).
- test-32, the proofs, the images and the docs pass.
- accept-aarch64 (435 s) and docs ran in the background, each alone.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Every stage passes. POST is 990/990 on all four boots, with one
dict_hash (0x847bd8ef42f53e14). test-32, the proofs, the images and the
docs pass. accept-aarch64 (416 s) and docs ran in the background, each
alone.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob ruled: the line for new work is drawn with a merge, then a
tag, v4.0.0-alpha.5. This sets LITHOS_VERSION in kernel/Makefile and
updates README, CHANGELOG and CLAUDE.md to match.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Every stage passes. POST is 990/990 on all four boots, with one
dict_hash (0x847bd8ef42f53e14). test-32, the proofs, the images and the
docs pass. accept-aarch64 (411 s) and docs ran in the background, each
alone.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
CI's Acceptance fails on aarch64 at BIRTH on every head since 58d0e887.
The birth takes more than 300 s there and does not reproduce here: about
11 s, unsigned. The heartbeat spiral was ruled out by an instrumented
scratch kernel.
Captain Bob ruled "Diagnose on CI". On a failed check, the session script
asks QEMU's monitor for the registers three times, a second apart, and
prints the program counters, so CI's next failure shows where the kernel
is and whether it moves. A monitor that does not answer does not stop
the run.
Test first: scripts/test_qemu_session_snapshot.sh, 5 checks; 3 failed
before. The register text was checked on real, paused QEMU for aarch64,
riscv64 and x86.
The evidence and the measurements are in logs/20261119-ci-snapshot. The
local aarch64 timing run is in logs/20261118-235613.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
accept-aarch64 failed at BIRTH. The new CPU snapshot showed the kernel
looping, not hung. The other stages passed, including riscv64, the
stick, test-32, the proofs and the images. I ran them before reading
aarch64's result, which was the wrong order.
The repro boot (logs/20261119-012050) gave the code at each PC. Matched
in the loader EFI's map, they name bitmap_get inside
stadium_resident_heat_total, then background_decay and isqrt64. That is
the heartbeat's load, which walked all 335,544 Stadium cells every tick.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Eight words at the PC, by its address, name the function offline. The
image loads where the firmware puts it, and the boot does not print
that address. aarch64's monitor has no $pc, so the address is taken
from the register line; checked on a real, paused QEMU.
Test first: the snapshot test, 6 checks; the new one failed before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
The heartbeat's one input walked all 335,544 Stadium cells every tick,
about 15 ms on aarch64 TCG, longer than a tick. At fast pacing it took
the CPU, and a birth missed the session's 300 s on CI. That was my
11-17 change.
Captain Bob ruled: sample less often, scaled by the cells, measured
first, as a slice each tick; the slope inference later.
- stadium_heat_walk_step() walks STADIUM_HEAT_SLICE_CELLS cells per
tick (Kconfig, 5500, about 0.25 ms on aarch64 TCG).
- A completed walk is one sample.
- The heartbeat's load takes a sample only when a walk is done.
Test first: test_stadium_node (21 checks) and test_heartbeat (13). Both
failed to build before. All kernel host tests pass. Four aarch64 boots
in a row pass 18/18 each, BIRTH included.
The evidence is in logs/20261119-heat-slice.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Every stage passes. accept-aarch64 passes BIRTH and every later check,
in 459 s. POST is 990/990 on all four boots, with one dict_hash
(0x847bd8ef42f53e14). test-32, the proofs, the images and the docs pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
rajames
merged commit fdf9c1946c into master2026-11-19 11:19:50 +00:00
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
What and why
StarForth-v4.0.0intomaster: 122 commits since268e5593. Most of them carry out #189: the kernel'sboot of the v4 fabric, built to spec (Captain Bob: "Build it to proper spec").
046ce0be). Hera is born empty at place 0 and given the nucleus as a capsule. POSTruns in the fabric. Nodes are born and killed through her. The fabric is lockstep: every node that is
not blocked executes one instruction word at each step. It is not a scheduler (ENGINE.md 3d; MESH.md
ruling 7).
call, and the loops (#189 Phases 2 and 7). v4's POST is v3'srun_all_tests, as written: 990 cases.docs/v4.0.0/plans/2026-11-17-one-adaptive-heartbeat.md). Thekernel's heartbeat paces itself. Its only input is the Stadium's resident heat. Every VM ticks on it in
lockstep, active or not. It is held through POST.
edbb79e2). Each step walks only the places that hold a node. The host mesh boot'smedian fell from 11.695 s to 7.999 s over five alternating runs. aarch64 had stopped in POST at 570 s
since 37469f6c; it now boots inside it. The direct loop before
046ce0bewas architecturally incorrect;the fabric wiring removed it and exposed the slowdowns, and mechanism A is the first speed-up the
instruments found (FABRIC-3 XLVIII.25-26).
v4/system/dump.c). When a node stops, in every build, it prints the fault,P/A/B, both stacks, the return stack named by word, and the running line.
scripts/nightly.sh diagnose. After any failed stage, the EMERGENCY build is made and itsartifacts kept; the workflows run it on failure.
protocol-reportnow writes a report when a bootstops before POST.
section 13, capsules/README, .claude/CLAUDE.md.
Acceptance
logs/20261117-235006/PROTOCOL.mdfor the head: the protocol onc30e4a0a, PASS.[zuse@Hera] ok>withPARITY:V4_POST tests=990 pass=990 fail=0and onedict_hash,0x627f392ccbaa96f1.scripts/check_protocol.shaccepts58d0e887.logs/:logs/20261117-mechanism-a/(timing, sanitizers, tests),
logs/20261117-heartbeat/,logs/20261117-dump/,logs/20261117-195259/fabric-cost/.The record
docs/fabric/FABRIC-3.mdhas the rulings and what was built: XLVIII.24-26 and the entriesbefore them.
CHANGELOG.md: not updated in this branch.🤖 Generated with Claude Code
https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
90c5c140-- all five stages pass 1b1fe66dba9415d7ed-- all five stages pass, and the images stage d6846b3487feb740c5, and the stage outputs of 9415d7ed's acceptance 3c1ccabc8edd580e07-- PASS, all eight stages (#135)defd47e5-- PASS, all eight stages54311be1(#168) -- PASS, all eight stageskernel/Makefile gave capsules/v4/sys/map-64.f18 and map-32.f18 one rule with two plain targets -- in GNU make, two rules sharing a recipe. Under -j both ran at once: two make -C v4 sysmap built and ran the same v4/build/mkmap-<w> side by side (Build run 255's RASPI job: "mkmap-32: Permission denied"). Now one grouped target (&:, GNU make 4.3+, as here and in the CI image). Seen to fail first: the two targets with -j8, mkmap removed each time, 12 runs -- 6 failed ("Text file busy", "Permission denied"). After: 0 of 12. Both sets of runs in logs/20261112-152711/race-185/. The map files are unchanged. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm- heat.c call_index(): the comment read as an unfinished thought ("hash-mod or just take low bits?", "ignore for now? Or clamp?"). It now states the rule the code follows: indexed by the target, a target past the table not counted, -1 telling the callers so; the table D-6's stand-in, which ENGINE step 4 Part A1 plans to remove. - stack.h points 1 and 2 said there was no overflow detection "and there never will be" and that a depth exceeded corrupted silently -- D-2, which D-16 revised. Now: guarded before every opcode, a mesh node 10 and 9, the host node 32 and 32 (D-17); C code that serves a node checks the depth itself. - stack.c named an 8-ring and depths 10 and 9 only, and said D-2's rule still held. - DATASHEET-F18.md section 13 item 4 (this gap) struck, resolved in place. Comments only: test_stack and test_heat unchanged at both widths (730,137 and 8,272 checks, 0 failures). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm9700a7b4-- PASS, all eight stagessk_v4_run's three stops -- "not started", "did not let go of POST's blocks", "stopped on a fault" -- ended in for (;;) { }: a core spinning at 100%, the thing #151 removed from the prompt. Each now calls sk_v4_stop() (kernel/src/v4/sk_v4_stop.c): interrupts off, then arch_halt() for ever. Each still says why first. sk_v4_stop.c added to the loader's and the kernel's v4 sources. kernel/tests/test_sk_v4_stop.c, with arch_halt and arch_disable_interrupts faked (the halt jumps back out after three calls; an alarm fails a spin): against the stop as it was -- "spun for 2 s without halting"; now -- 2 checks, 0 failures: halts in a loop, interrupts off before the first halt. Built for amd64, aarch64 and riscv64. #177's "seen on the three architectures": no boot reaches these stops, so it is shown on the host only. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbmd7b2f6bf-- PASS, all eight stages 83898b524dCaptain Bob, 2026-11-12: "correct deficiencies". heartbeat_capture_tick_snapshot wrote cache_hits_delta and bucket_hits_delta as 0 ("TODO: wire ..."), though physics_hotwords_cache.c counts both. Now each is read from the VM's hot-words cache as a delta since the previous tick, the way word_executions_delta is (and after the cache's stats are reset, the count since then); a VM with no cache has 0. word_executions_delta was measured already: doe_log.c's "TODO: 0" for it was stale, and its three column notes are corrected. kernel/tests/test_heartbeat_export.c, 4 checks: a VM in memory with a cache whose counters the test sets, captured tick by tick. Against the code as it was: 3 failures (the hits 0 where the cache counted 5 and 3, then 7 and 7). Now: 0. Built for amd64, aarch64, riscv64 and the hosted v3. Shared v3 code, under the ruling. v3 is not part of acceptance; no DoE has run since 2026-09-16. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbme0dcaa16-- PASS, all eight stagesc76a742f-- PASS, all eight stagesdd6c393a-- PASS, all eight stagesbe456d44-- PASS, all eight stages892e3889-- PASS, all eight stagesb1cac339-- PASS, all eight stagesf10d2bc8-- PASS, all eight stages046ce0be-- PASS (#189, the mesh boot to spec and its edges)e8e7ec8b-- PASS (#189 Phase 2, the word accounts at call)v3's whole POST on v4 at 983 of 1010 (the table's own count with its drops), from 850. What remains is capacity ("Dictionary full", the 100-900-deep stacks): next, as ruled. Found and fixed under the rulings (docs/v4.0.0/POST-PARITY.md): - IS and DEFER@ refuse a word DEFER did not make, as v3: error 22, "Not deferred". v4's IS stored into any word -- POST's `42 IS DUP` broke DUP for every case after it. DEFER marks its word with flag 128 (dict.f18). - SMUDGE and HIDDEN are compile-only, as v3's table: at the prompt v4's hid the newest word. - Q./ by zero gives 0 and no error (ruled: v3's table); error 11 retired. The ROM's, the test's reference and its hand-assembled copy changed alike. - FENCE is v3's: it protects what is defined so far. The variable is (FENCE). - The line's length is (#SRC), no longer SPAN: an EXPECT in a line cut the rest of the line off, and 0 EXPECT could not leave SPAN 0 (FORTH-83). - POST gains V4_POST_HEAD: the values left and how the output begins -- the form of a printout whose values are the run's own (TOP-WORDS, WORD-ENTROPY). post_all_rules.py: the rest of the whole table where it is not taken as v3 ran it -- v4's units for the cases that wrote through HERE as a byte address (CMOVE>, BLANK, ERASE, FILL wrote into the system's code), the standard's results (COMPARE, SEARCH, BASE output, THRU, >BODY with FIND's address), v3's descriptions where v3 printed otherwise, and the cases dropped with their reasons (SP@/SP! by D-2, DABS/DMIN underflow, CMOVE>.bounds, THRU.negative_end). make -C v4 test: all v4 tests passed (both widths, mesh-boot, inference). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbmPhase 7, Task 8: PARITY:V4_POST tests=990 pass=990 fail=0 on Hera (mesh-boot); the hosted product 945 of them, all passed -- the 45 cases of physics.4th's words need the kernel. - Enlarged, as ruled ("Enlarge v4 to fit"): a node is 32768 words, its data stack 1024 deep (v3's), host and kernel alike; the dictionary's end moves with them. - POST's runner compares every value a case leaves (v3's leave 101); V4_POST_KERNEL marks a case of the kernel's words, and a host without a kernel does not run it. - mkpost.py --all is POST's table (make -C v4 post79), with docs/v4.0.0/POST.md; it reads a module's #defines (v3's Mama cases are MAMA_IMPLEMENTED, 0: v3's own POST skips all 20); it cuts no line between a word and the name it takes. - Rules for the last of the table: the min/max and array-sum programs written as meant, CREATE's 26-character name (FORTH-79: 31), many_defs. - POST_FLOOR 990. - Tests sized for a 32-deep stack and a 16384-word node made deep-stack-ready (test_host_depth, test_host_quit, test_host_node, hosted-check, depthsweep.py). make -C v4 test: all v4 tests passed. make -C kernel/tests test: all passed. depthsweep at 1024: 230 runs, 0 failures. v4-checks running. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbmRuled (Captain Bob, 2026-11-14): "Find the TCG cost first". v3's whole POST took the amd64 accept from about 3 m 15 s to 8 m 45 s, and aarch64 past its 570 s (logs/20261114-091651, STOPPED, committed). Profiled on the host (gprof, mesh-boot): 87% of the time in POST's lines, about 280,000 opcodes a line -- the interpreter's dictionary lookup. - POST puts the dictionary back before each of v3's suites, as v3's run_test_suite does ("to remove test-created words"): a mark, (POST), forgotten and made again (V4_POST_SUITE). Only the dictionary: SCR, BASE and the rest stay as the cases left them. - (LOOKUP) tries each entry by its count and first character first -- one fetch of the name's first cell, masked -- and only then its hidden flag and the whole name. The answer is the same; an entry that cannot be it costs a few opcodes, not some forty. mesh-boot: 37.7 s -> 15.7 s, POST 990 of 990. make -C v4 test: all v4 tests passed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm1b65ac63, clean) -- amd64, riscv64, test-32 PASS; aarch64 STOPPED at 570 s in POST; stick-high's stage FAIL on the missing aarch64 dict_hash; two diagnose builds; labelled c890cadc121b65ac63-- FAIL (aarch64 STOPPED in POST at 570 s; stick-high's dict_hash check; proofs, images, docs not run), written by the fixed report; the report's test before and after 65d562d4e1cc09a88b(the one adaptive heartbeat) -- FAIL: aarch64 STOPPED in POST at 570 s (the fabric step's cost); amd64, riscv64, stick POST 990/990, one dict_hash 0x627f392ccbaa96f1; test-32 PASS; labelled046ce0be, the fault dump, the one heartbeat046ce0beis not a regression (Captain Bob)edbb79e2(mechanism A): PASSedbb79e2f93cde222dedbb79e2(990 tests, word_count=403, dict_hash 0x627f392ccbaa96f1) c2364b3646046ce0be) -- the kernel's mesh fits each born node with the mesh ROM 151b52f15ac30e4a0a: PASSb80bd5cd, superseded before its docs stage 8301d09db0295cd518-- FAIL at docs only (STARSHIPOS had no dictionary entry); superseded 66d0af55d0573d386b: PASS913a896e: PASS53b1e98e: PASSe29a00f7: PASS0565a039(release: v4.0.0-alpha.5): PASS36b25805-- FAIL: aarch64 BIRTH stalled here too; the repro boot that caught it b6c87b30457975bab0(the heartbeat's walk in slices): PASS