Every cell laid down that holds an address of the assembly is recorded,
with its branch slot or 7 for the whole cell, as ARTIFACT.md's
relocations need. Test first: test_text 13 checks failed before.
test_host_rom_nucleus runs with the record on: 1825 branches, 664 whole
cells, the ROM still equal word for word. v4 test, test-32 and
ASan+UBSan pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob ruled "As v3: add at call", "Engine: call redirects" and
"Build it".
- The engine's call hook returns where P goes.
- The kernel checks every call from code defined at run time. A TTL is
counted down at the call. At a TTL of 0, or a denied word, the call is
sent to (ACL-CALL), with the target on the data stack, which rechecks
or refuses it by the word's own name, then goes on to it.
- A word being compiled is no longer checked, as in v3.
Two things were added while building, both from v3:
- (ACL-BUSY) turns the check off while the policy rechecks. Without it
the recheck recursed.
- (ACL-CALL) does its own refusal and leaves WORD's buffer alone. A first
version broke POST's CONVERT.digits.
Test first: test_mesh_boot's five new checks failed before; now 120 pass
and POST is 990/990. test_exec tests the redirect. test_host_quit's two
v4-only expectations now follow v3: a denied word compiles.
All 52 tests, test-32 and ASan/UBSan pass. An amd64 session passes 18/18.
Five alternating host boots: +1.4% median, inside the noise.
DECOMPOSITION.md 5.20 is updated, and the FPGA's call opcode is recorded
as owing the same redirect.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm
Captain Bob ruled "ACL at boot", loaded "From init.4th", the include
"Compose[d] in FORTH".
- capsules/v4/init.4th, at block 3980: S" v4:ACL.4th" CAPSULE-INCLUDE,
then STARSHIPOS.
- capsules/v4/ACL.4th is the policy, git mv'ed from v4/capsule/ACL.fth
and reflowed into chained 64x16 blocks, its text unchanged.
- CAPSULE-INCLUDE in centre.4th places a capsule and LOADs its lowest
block. It is composed from CAPSULE-OPEN and CAPSULE-LINE, with no new
C word.
- The mesh loads v4:init.4th after POST and COLD, its one fixed step
there.
Test first in test_mesh_boot: the parity line, DOE without naming
STARSHIPOS, and ACL-BOOT pinned. The before-run failed.
After: 113 checks pass and POST is 990/990. All 52 tests, test-32 and
ASan/UBSan pass. The dictionary has 405 words. One amd64 session passes
18/18 with "ACL: active".
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
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 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
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
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
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
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
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'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
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
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
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, 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: "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
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
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
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
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: "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