Skip to content

t27b: an anytype parameter the body never reads is void, and a compile-time argument passes for it, from a t27 plan (Closes #7876) - #8028

Merged
gHashTag merged 11 commits into
masterfrom
t27b-unread-anytype
Oct 9, 2026
Merged

t27b: an anytype parameter the body never reads is void, and a compile-time argument passes for it, from a t27 plan (Closes #7876)#8028
gHashTag merged 11 commits into
masterfrom
t27b-unread-anytype

Conversation

@gHashTag

@gHashTag gHashTag commented Oct 9, 2026 •
edited
Loading

Copy link
Copy Markdown
Owner

Closes #7876. Part of #6063 (t27b coverage), which is item C10 of #6488. It follows #7857 (#7875), which covers anytype on a fn no test reaches.

Decision

specs/port/browseros/trios/agent-server/apps/server/src/lib/identity.t27 passes the reference 8/8. t27b refused isNotFoundError(err: anytype) bool as type anytype, then refused each call as ExprCall(rejected fn). The fn's body is return false;, and the tests call it with "ENOENT", "other error" and null.

What the reference does. t27c prints the fn with _ = err; // unused by the spec body, and prints each call as written. Zig 0.16 instantiates the fn once per argument type: *const [6:0]u8, *const [11:0]u8 and @TypeOf(null). The body never reads the parameter, so every instance computes the same thing from the other parameters. Nothing here passes by accident: no value goes unread except the one Zig never reads either.

t27b now lowers such a parameter as t27b's void, the struct with no fields (#7267), so one body stands for every instance. An argument for it is taken when evaluating it runs nothing:

  • null;
  • a value t27b's own lowering folds at compile time to an integer or a string (a literal, a constant's name).

The value passed is void's, which nobody reads. Any other argument is refused by name as ExprCall(anytype argument), because evaluating it might run code that t27b would then skip.

A body that names the parameter keeps type anytype. That case is #7878: queen-public-agents.t27's heartbeatTime dispatches on @TypeOf(value) == i64.

Probes on the t27c lab (zig 0.16.0):

call reference t27b
is_nf("a\qb"), is_nf(99999999999999999999999) pass pass
is_nf(bump(1)), is_nf(2.5), is_nf(undefined) pass ExprCall(anytype argument) (conservative)
reads(err: anytype) with return err == 1;, called pass type anytype

Where the decisions live

They are in specs/tri/t27b/any_param_plan.t27, which provides param, arg, what and why. t27c gen-rust produces gen/rust/tri/t27b/any_param_plan.rs, which lower.rs mounts. The glue has three parts:

Conformance spec first

specs/tri/t27b/conformance/unread_anytype.t27 covers:

  • string, constant-name and null arguments;
  • an integer literal beside a run-time parameter;
  • the parameter in the middle of three, with every branch of the other two taken.
result
reference, t27c test-report 4/4, 0 vacuous, 9 runtime asserts; sealed (seal --save, --verify: all MATCH)
master's t27b refused: type anytype
this branch's t27b (aarch64 under qemu, t27b test --check) 4/4; identity.t27 8/8
  • Conformance mutants. tri mutate spec kills 6 of 6.
  • Plan. t27c test-report gives 4/4, 0 vacuous; sealed. tri mutate spec kills 5 of 5.
  • Glue mutants. All 4 are caught by tests/tail.rs:
    • never void;
    • every argument taken;
    • a named parameter treated as unread;
    • null refused.

Budget port (strict gate, see #8236)

The strict budget lets hand-written code under cli/t27b only shrink. This PR's glue adds 24 lines to lower.rs, so the same PR ports more than that out of hand Rust:

  • CmpOp's tables. CmpOp::symbol, negate, swap and holds_f64 in cli/t27b/src/ir.rs were match tables. They are now t27c gen-rust of specs/tri/t27b/cmp_op.t27 (gen/rust/tri/t27b/cmp_op.rs, mounted in ir.rs as co). Each method is one line, and a compile-time assert! pins CmpOp's discriminants to the spec's codes. The spec's 6 tests are the old tables row by row, the two involutions, and IEEE NaN behaviour. Reference 6/6, 0 vacuous; 25 of 25 assert mutants fail under both t27b and the reference. CmpOp::holds stays: it compares i128, which t27b cannot run yet.
  • One Rust test dropped. The tail.rs test of the anytype parameter is gone. Its pass case is conformance/unread_anytype.t27, which the corpus runs; its refusals are any_param_plan.t27's decisions.

Hand lines under cli/t27b against master: +34 -41, net -7.

Lines (git diff --numstat --no-renames origin/master...HEAD)

file kind added deleted
cli/t27b/src/ir.rs hand-written, ported out 10 40
cli/t27b/src/lower.rs hand-written glue 24 1
gen/rust/tri/t27b/cmp_op.rs t27c gen-rust, not hand-edited 91 0
gen/rust/tri/t27b/any_param_plan.rs t27c gen-rust, not hand-edited 36 0
specs/tri/t27b/cmp_op.t27 plan spec, 6 tests 138 0
specs/tri/t27b/any_param_plan.t27 plan spec, 4 tests 84 0
specs/tri/t27b/conformance/unread_anytype.t27 conformance spec, 4 tests 56 0
.trinity/seals/*.json (3) t27c seal --save, schema 2 81 0
docs/reports/t27b_expectations.json ledger 5 2
AGENTS.md t27b remainder line 6 1

Gates and checks (t27c lab, tree merged with master cde548e)

  • check_budget() over git diff --numstat --no-renames cde548ecd...HEAD, built from cde548e's gen/c/policy/own_language.c: exit 0.
  • check_all() with cde548e's tools/policy/foreign-exceptions.txt and the name-status: exit 0.
  • cargo test --release -p t27b (aarch64 under qemu): 14 suites ok, 0 failed.
  • Generated code. t27c gen-rust of both plans with that tree's t27c is byte-identical to the committed files.
  • Seals. All three re-minted with that t27c (schema 2); seal --verify: all hashes MATCH.
  • AGENTS.md, re-measured with wc -l: 14756 plus 1222, tests 8284, against master cde548e's 14763 plus 1222 and 8284.
  • Ledger. lib/identity.t27 goes from blocked to pass, and three new rows all pass. max_not_pass is recounted from the rows.
  • Corpus. t27b corpus specs over master's tree, master's t27b against this branch's: exactly one file changes, lib/identity.t27, from blocked (type anytype) to pass (8 tests, 8 asserts). Of 1731 files, no other changes.

t27b lab: signed receipt (#7686), after the merge

This PR landed as c6237a7. The receipt compares that squash commit with its parent on master, 66a28fc. The base is the lab's own master run (no nonce); the head was requested with a fresh 32-byte challenge.

/work/t27c-master corpus-receipt compare BASE HEAD --challenge-head <mine>, on the t27b lab from /work/t27 (input and output lines left out):

base "66a28fc172558d4da43f342fb83b14088cda1a51" AUTH_MISSING_NONE AUTHOR leaves-bound true
head "c6237a71b71d6344903d6c68c0851d9e7dbc6c20" AUTH_MISSING_NONE FRESH leaves-bound true
  verdict changed specs/port/browseros/trios/agent-server/apps/server/src/lib/identity.t27
  verdict only-head specs/tri/t27b/any_param_plan.t27
  verdict only-head specs/tri/t27b/cmp_op.t27
  verdict only-head specs/tri/t27b/conformance/unread_anytype.t27
  lane improved specs/port/browseros/trios/agent-server/apps/server/src/lib/identity.t27
  lane neutral specs/port/trios/crates/trios-scarab-types/rings/SR-02/src/lib.t27
  lane improved specs/tri/t27b/any_param_plan.t27
  lane improved specs/tri/t27b/cmp_op.t27
  lane improved specs/tri/t27b/conformance/unread_anytype.t27
totals false inputs false verdicts false outputs false: IMPROVED_ONLY

Exit 3, IMPROVED_ONLY. The moves are the ones before the budget port: identity improved, the new specs improved, SR-02 neutral (only its asm moved). The CmpOp port changes no verdict.

Earlier receipt (before the budget port)

Head 61855f0 against master 7a07828 gave IMPROVED_ONLY: identity improved, the two new specs improved, and SR-02 was neutral. SR-02 passes on both sides; only its asm moved, because its fmt(self, f: anytype) fns never name f. Reference disagree was 0 / 0.

Generated with Claude Code

@claude
t27b: an anytype parameter the body never reads is void, and a compil…
9642699 
…e-time argument passes for it, from a t27 plan (Closes #7876)

identity.t27's `isNotFoundError(err: anytype) bool` never names `err`; the
reference instantiates it per argument type ("ENOENT", null) and every
instance returns the same thing. t27b refused it as `type anytype`.

specs/tri/t27b/any_param_plan.t27 (t27c gen-rust, mounted in lower.rs):
such a parameter is t27b's `void`, so one body stands for every instance,
and the argument for it is taken when evaluating it runs nothing: `null`,
or a value t27b folds at compile time to an integer or a string. Any other
argument is refused as `ExprCall(anytype argument)`; a body that names the
parameter keeps `type anytype` (#7878 for the @typeof dispatch).

Conformance: specs/tri/t27b/conformance/unread_anytype.t27, t27c
test-report 4/4, 0 vacuous, 9 runtime asserts; tri mutate spec 6 of 6
killed. Plan 4/4, 0 vacuous, 5 of 5 mutants killed. 4 of 4 glue mutants
caught by tests/tail.rs. Both specs sealed (seal --save, --verify: MATCH).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@gHashTag
gHashTag enabled auto-merge (squash) October 9, 2026 06:24
This was referenced Oct 9, 2026
Merged
Merged
Merged
@claude
Merge master (7a07828) into t27b-unread-anytype (Closes #7876)
61855f0 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@github-actions github-actions Bot mentioned this pull request Oct 9, 2026
Merged
@github-actions

github-actions Bot commented Oct 9, 2026

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-10-09 06:42:05 UTC

Summary

Status Count
Total Open PRs 50
PRs with Failing Checks 37
PRs with All Checks Green 13
READY 0
FAILING 37
PENDING 0
NO CHECKS YET 0

These columns do not partition: 0 + 37 + 0 + 0 = 37, and there are 50 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=3c0ade9e73e4 != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

This was referenced Oct 9, 2026
Merged
Merged
Merged
Merged
Merged
Merged
@claude
Merge master (2dfe978) into t27b-unread-anytype (Closes #7876)
685261b 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

# Conflicts:
#	AGENTS.md
This was referenced Oct 9, 2026
Merged
Merged
@github-actions

github-actions Bot commented Oct 9, 2026

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-10-09 07:04:37 UTC

Summary

Status Count
Total Open PRs 50
PRs with Failing Checks 35
PRs with All Checks Green 15
READY 0
FAILING 35
PENDING 0
NO CHECKS YET 0

These columns do not partition: 0 + 35 + 0 + 0 = 35, and there are 50 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=3c0ade9e73e4 != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

@github-actions github-actions Bot mentioned this pull request Oct 9, 2026
Merged
@github-actions

github-actions Bot commented Oct 9, 2026

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-10-09 07:28:39 UTC

Summary

Status Count
Total Open PRs 50
PRs with Failing Checks 44
PRs with All Checks Green 6
READY 0
FAILING 44
PENDING 0
NO CHECKS YET 0

These columns do not partition: 0 + 44 + 0 + 0 = 44, and there are 50 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=3c0ade9e73e4 != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

This was referenced Oct 9, 2026
Merged
Merged
Merged
Merged
Merged
Merged
@claude
Merge master (09ffc2f) into t27b-unread-anytype (Closes #7876)
c30cce9 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

# Conflicts:
#	AGENTS.md
This was referenced Oct 9, 2026
Merged
Merged
Merged
@github-actions

github-actions Bot commented Oct 9, 2026

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-10-09 16:53:27 UTC

Summary

Status Count
Total Open PRs 50
PRs with Failing Checks 50
PRs with All Checks Green 0
READY 0
FAILING 50
PENDING 0
NO CHECKS YET 0

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=557cd271f4e3 != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

This was referenced Oct 9, 2026
Merged
Merged
Merged
Merged
Merged
Merged
Merged
Merged
Merged
Merged
Merged
Merged
@claude
Merge master cde548e into t27b-unread-anytype; no hand lines added (C…
42f5946 
…loses #7876)

Conflicts were the AGENTS.md remainder line and the ledger cap only; the
cap is recounted from the rows (17). The three seals of this PR are
re-minted with cde548e's t27c (seal schema 2). Hand lines under cli/t27b
against cde548e: +34 -41, net -7.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@github-actions github-actions Bot mentioned this pull request Oct 9, 2026
Merged
@claude
Merge master b9d22d0 into t27b-unread-anytype; no hand lines added (C…
4050228 
…loses #7876)

The only conflict was the ledger; master's rows and this PR's are kept and
max_not_pass is recounted from the rows.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@github-actions

github-actions Bot commented Oct 9, 2026

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-10-09 18:49:41 UTC

Summary

Status Count
Total Open PRs 50
PRs with Failing Checks 46
PRs with All Checks Green 4
READY 0
FAILING 46
PENDING 0
NO CHECKS YET 0

These columns do not partition: 0 + 46 + 0 + 0 = 46, and there are 50 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=557cd271f4e3 != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

@claude
Merge master 3ac8bef into t27b-unread-anytype; no hand lines added (C…
4fdc960 
…loses #7876)

Master brings #8252 (lower.rs's source-text rules move to source_text.t27).
The only conflict was the AGENTS.md remainder line, re-measured with wc -l.
Hand lines under cli/t27b against 3ac8bef: net -7.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@github-actions github-actions Bot mentioned this pull request Oct 9, 2026
Merged
@github-actions

github-actions Bot commented Oct 9, 2026

Copy link
Copy Markdown
Contributor

PR Dashboard

Generated at: 2026-10-09 19:21:12 UTC

Summary

Status Count
Total Open PRs 50
PRs with Failing Checks 47
PRs with All Checks Green 3
READY 0
FAILING 47
PENDING 0
NO CHECKS YET 0

These columns do not partition: 0 + 47 + 0 + 0 = 47, and there are 50 open PRs. A PR is being counted twice or not at all.

Seal Status

  • ⚠️ STALE -- sha256(compiler.rs)=557cd271f4e3 != manifest seal=87e5cbd3ad94.
    The committed NMSE numbers were certified against an older compiler.rs.
    Run scripts/reseal-check.sh locally for the two-step reseal command (advisory; not a merge gate).

@gHashTag
gHashTag merged commit c6237a7 into master Oct 9, 2026
39 of 40 checks passed
gHashTag pushed a commit that referenced this pull request Oct 9, 2026
@claude
Merge master 325b0b1 into t27b-const-div; no hand lines added (Closes #…
2bc1eef 
…7812)

Master brings #8028. Conflicts were the AGENTS.md remainder line (re-measured
with wc -l) and, where both sides moved rows, the ledger (rows kept from both
sides, max_not_pass recounted). Hand lines under cli/t27b against 325b0b1:
net -21.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@gHashTag gHashTag mentioned this pull request Oct 9, 2026
Closed
gHashTag pushed a commit that referenced this pull request Oct 9, 2026
@claude
Merge master 325b0b1 into t27b-int-cast-index-7902; no hand lines add…
d212dcd 
…ed (Closes #7902)

Master brings #8028. The only conflict was the AGENTS.md remainder line,
re-measured with wc -l. The ledger merged without a conflict, but both sides
had lowered max_not_pass by one to the same text; it is recounted from the
rows (18). Hand lines under cli/t27b against 325b0b1: net -5.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gHashTag pushed a commit that referenced this pull request Oct 9, 2026
@claude
Merge master 325b0b1 into t27b-repeat-local-7909; no hand lines added (…
fb1c972 
…Closes #7909)

Master brings #8028. Conflicts were the AGENTS.md remainder line (re-measured
with wc -l) and, where both sides moved rows, the ledger (rows kept from both
sides, max_not_pass recounted). Hand lines under cli/t27b against 325b0b1:
net -7.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@gHashTag gHashTag mentioned this pull request Oct 9, 2026
Merged
gHashTag added a commit that referenced this pull request Oct 9, 2026
@gHashTag @claude
verified(corpus_receipt): a request the lab has not fetched yet waits…
8f2c8da 
… 30 min instead of being dropped (Closes #8320) (#8329)

`corpus-receipt admit` dropped a lane request at once when its sha was
not a branch head or on master's first-parent line as the lab last
fetched it. On 2026-10-09 that dropped four lane heads and the first
request for #8028's squash commit c6237a7, before the lab had fetched
it.

request_verdict() now answers WAIT_ORIGIN (5) for a sha not on origin
that has waited at most REQUEST_ORIGIN_GRACE_S (1800 s), and
request_exit() maps it to 2. lab.py already keeps any request whose
admit exit is neither 0 nor 1, so no hand-written line changes. Waiting
never runs a request: running still needs the sha on origin, so the
bound on who can put code in front of the signing step is unchanged.

Generated bootstrap/gen/rust/verified/corpus_receipt.rs with t27c
gen-rust; seal re-saved, 32 of 32 tests pass. On the t27c lab:
admit gives WAIT_ORIGIN/2 for 0,0,0 and 1800,3,0; NOT_ON_ORIGIN/1 for
1801,0,0; ADMIT/0 for 0,0,1; QUEUE_FULL/1 for 0,4,1; EXPIRED/1 for
21601,0,1. cargo test -p t27c -- receipt: 18 passed.

Co-authored-by: Claude <claude@anthropic.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
gHashTag added a commit that referenced this pull request Oct 9, 2026
@gHashTag @claude @dmitrii-f-t27
t27b: the node scans of lower.rs move to specs/tri/t27b/ast_scan.t27 (C…
e4338f4 
…loses #8347) (#8350)

* t27b: the node scans of lower.rs move to specs/tri/t27b/ast_scan.t27 (Closes #8347)

names_in, array_locals, decls_of, calls_in, scan_addr_taken,
count_assigns, ref_mutable_names, misprinted_ifs and mark_tail_returns
are now t27c gen-rust of specs/tri/t27b/ast_scan.t27
(gen/rust/tri/t27b/ast_scan.rs), mounted in lower.rs as `ax`. Each scan
reads ast_walk.t27's bytes of the node list (`flat`) and writes a mark
per node it collects; `marked` returns the marked nodes in preorder, and
each collector keeps its signature with a one-line body.

A differential of the deleted collectors against the new ones agrees on
1,000,000 random forests: sets, counts, the preorder of calls_in and
decls_of, and the node addresses misprinted_ifs and mark_tail_returns
collect.

docs/reports/t27b_expectations.json gets the new spec's row: t27b passes
it (9 of 9 tests).

Hand Rust under cli/t27b: 111 lines deleted, 20 added, net -91. See
#6198.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* verified: ed25519.t27 states it makes no constant-time claim (Refs #8041) (#8065)

Replaces the one-line constant-time remark with an honest paragraph: no
timing claim is made or checked, sign/public_key touch the secret seed,
and the masked selects are shape, not a guarantee. Records the #8041
Wycheproof and differential-fuzz results. Comment-only change: generated
Rust is byte-identical; seal spec_hash refreshed, 16/16 tests pass.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Claude <claude@anthropic.com>

* verified(corpus_receipt): a request the lab has not fetched yet waits 30 min instead of being dropped (Closes #8320) (#8329)

`corpus-receipt admit` dropped a lane request at once when its sha was
not a branch head or on master's first-parent line as the lab last
fetched it. On 2026-10-09 that dropped four lane heads and the first
request for #8028's squash commit c6237a7, before the lab had fetched
it.

request_verdict() now answers WAIT_ORIGIN (5) for a sha not on origin
that has waited at most REQUEST_ORIGIN_GRACE_S (1800 s), and
request_exit() maps it to 2. lab.py already keeps any request whose
admit exit is neither 0 nor 1, so no hand-written line changes. Waiting
never runs a request: running still needs the sha on origin, so the
bound on who can put code in front of the signing step is unchanged.

Generated bootstrap/gen/rust/verified/corpus_receipt.rs with t27c
gen-rust; seal re-saved, 32 of 32 tests pass. On the t27c lab:
admit gives WAIT_ORIGIN/2 for 0,0,0 and 1800,3,0; NOT_ON_ORIGIN/1 for
1801,0,0; ADMIT/0 for 0,0,1; QUEUE_FULL/1 for 0,4,1; EXPIRED/1 for
21601,0,1. cargo test -p t27c -- receipt: 18 passed.

Co-authored-by: Claude <claude@anthropic.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* skills: t27b-loop, one tick of the unattended t27b improvement loop (Closes #8331) (#8332)

Budget guard, health checks with self-repair (hooksPath, master's red
checks, lab staleness and deploy drift, disk, agents), merge-when-green
rules, one plan step through the t27c lab, and a ledger line. Records
the 2026-10-09 traps that stalled lanes: absolute core.hooksPath,
UNSTABLE refusing --auto, specs landing without ledger rows, the lab
dropping unfetched requests, closing keywords in prose. Part of the
loop epic (see #8326).

Co-authored-by: Claude <claude@anthropic.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* t27c silicon reuses its bitstream; a ported script pays for the glue (Closes #8295) (#8296)

* t27c silicon reuses its bitstream; math_compare's Pell code moves to t27 (Closes #8295)

Wires specs/verified/bitstream_reuse.t27 (#8291) into t27c silicon. The
cache key hashes every input that can change a bit. A complete, intact
entry is loaded instead of running yosys, nextpnr and fasm2frames, and
every passing build is stored. Measured on this Mac with --skip-hardware
(so no board was touched): ternary_link built in 39 s, then the second
run printed 'bitstream REUSED' and finished in 4 s. The control bitstream
and the readback are untouched.

Hand-written code may not grow (owner rule 2026-10-09), so the same diff
ports math_compare.rs's Pell code to specs/math/pell_hybrid.t27: pell_u64,
pell_f64, hybrid_inner_product, hybrid_v2_cosine.
- The spec has 7 tests, 0 vacuous. Its GOLDEN_V2 vectors are held to
  1e-9; the Rust tests held them to 1e-6.
- math_compare.rs now calls the generated code, and all 10 of its existing
  Rust tests pass.
- Foreign lines: +39 -68, so a net of -29.

gen-rust lowers @sqrt of a bare float literal to (5.0).sqrt(), which rustc
rejects. The spec works around it with a typed constant.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* Pay for the bitstream-reuse glue by deleting a ported script, not math_compare.rs (Closes #8295)

own-language refused the previous push: bootstrap/src/math_compare.rs is not
on the owner's exception list, so it may not be edited at all, not even to
remove code. The Pell port is withdrawn from this PR.

The glue's +32 lines are now offset by deleting scripts/gen_w662.py (145
lines). Its .t27 port, specs/port/scripts/gen_w662.t27, already exists
(#8160), and nothing references the .py. Deleting a whole file is allowed
for any hand-written file. Net hand-written change: -114.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* bitstream reuse: the venv in the key, and every 10th hit is audited (Closes #8295)

Two weak spots from the loop ledger (#8314), both from Bazel's remote-cache
experience:
- an undeclared input poisons a cache, and fasm2frames and prjxray run
  under the openXC7 venv, so its pip freeze joins the key;
- nothing ever rechecked a hit. Every AUDIT_EVERY-th (10) hit is now
  rebuilt and compared; a mismatch deletes the entry and says so.

The first audit run found a real anomaly. Two builds of the same inputs
differed in 3 bytes, at offsets 162..165: the .bit header's build date
and time. So the audit compares the configuration stream from the sync
word 0xAA995566 (BIT_SYNC_WORD). Re-measured without hardware: audit run
'byte-identical', hits 10, then a REUSE run in 5 s.

Spec: 7 tests, 0 vacuous; the two new mutants are killed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude <claude@anthropic.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* t27b: / and % of two integer constants fold at compile time, as Zig folds them (Closes #7812) (#8027)

t27c prints `pub const CLK_DIV : u32 = CLK_HZ / (UART_BAUD * 16);` as
written, and Zig evaluates it at compile time. coerce_plan.t27 folds
`+ - * & | ^` on two typed constants and left `/` and `%` to run, so the
quotient was not a constant and specs/port/fpga/vivado/gf16_uart_sim_bench.t27
was refused as ConstDecl.

The decision is in specs/tri/t27b/const_div_plan.t27 (4 tests; `%`'s byte is
coerce_plan.t27's, taken by `use`), generated with `t27c gen-rust` to
gen/rust/tri/t27b/const_div_plan.rs and mounted in cli/t27b/src/lower.rs.
`/` folds when the divisor is not 0 and the quotient fits the type; `%`
folds on an unsigned type under the same conditions. The rest keeps its path:
a signed `%`, a zero divisor and a quotient that does not fit are refused in a
module constant, as the reference refuses them.

The conformance spec is specs/tri/t27b/conformance/const_division.t27:
`t27c test-report` 5/5, 0 vacuous. Glue: lower.rs +5 -2, tests/tail.rs +17,
tests/source.rs +1 -1 (module_var_rejections_are_precise used `B / 2` as an
initializer t27b could not fold; it now uses `B / 0`, which both sides refuse).
Ledger: gf16_uart_sim_bench.t27 now passes; the two new specs are new rows.

Part of #6063.

Co-authored-by: Claude <claude@anthropic.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

* spec(automation): tri-claim-batch v1 -- withdraw every accepted spec in a few wallet transactions (#8349)

Owner, 2026-10-09: "withdraw everything" -- 874 accepted specs on
dmitrii-f-t27, one press each until now. At most 16 mint messages per
request (4 earnings at a time to the signers), the wallet's own message
limit per transaction (4 by default, 255 at most), only earnings neither
revoked nor sent, oldest first, a cancelled prompt stops the run. The
minter is unchanged.

6/6 PASS, 9/9 negative controls caught, validate-vacuity 0 of 6.

Closes #8348

phi^2 + 1/phi^2 = 3 | TRINITY

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Claude <claude@anthropic.com>
Co-authored-by: Dmitrii Fedorov <dmitrii.f@t27.ai>
This was referenced Oct 9, 2026
Open
Open
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

t27b: an anytype parameter the body never reads, called with literals and null (identity.t27)

2 participants

@gHashTag @claude