D-CMH-1: class-qualified methods compile into Quack algebra, with no Rust per method by AdaWorldAPI · Pull Request #1504 · AdaWorldAPI/lance-graph

AdaWorldAPI · 2026-10-12T15:54:43Z

What this proves

A new method of a concept becomes executable by writing a declaration. No Rust function, adapter or interpreter is written for it.

SELECT billable_work_entry::billable_hours(cats) FROM cats
  → resolve_call     concept via ogar_vocab::canonical_concept_id
                     method label → MethodKey; payload = FROM table = bound table
  → MethodTable      prepared once per glove: SchemaSeal, ReadingId, Universe,
                     WideFieldMask required ⊆ available (Exact | Converted claims),
                     Quack programs lowered once
  → MethodBody::Query { filter: [Cond], folds: [Fold], law: Law }   (data)
  → prepared Quack programs → mask-risc                         (frontend 1)
  → R2IL procs: select_u32 + count | sum | group_sum            (frontend 2, parity)
  → Law finisher → Answer::Values | Answer::Truths

Rust supplies two reusable layers, each written once:

Algebra: conditions, Count, Sum, GroupSum.
Laws: certified finishers, today Values and NarsRevision.

A method composes these layers. It cannot supply its own algebra or law.

Each law declares the body shape it requires, and preparation checks that shape. For example, NarsRevision refuses a body that does not select independent evidence.

Experiments A–E

Full results and the capability / insufficiency ledger are in .claude/board/entries/2026-10-12-class-method-polyfill-ledger.md.

A. The seam

The class-qualified call, the same query built directly through Quack's binder, R2IL's procedures and a row loop all agree at n ∈ {1, 64, 65, 4097}.

A prepared method's programs() equals lower(direct query). So a method call executes exactly the program a direct Quack query would, and no separate row evaluator exists.

No SQL frontend reaches Quack today. resolve_call recognises one call form and is the seam such a frontend would hand its parsed call to.

B. Frozen executor

At e984885, hours_by_billable was declared in tests/method.rs only; that commit changes one file.

Quack: correct.
R2IL: typed refusal. Its SUM, GROUP_SUM and KEEP took their mask only from a computed slot. A body with no condition hands them the resident live plane instead.

Fixed in 8d5a36f: one shared mask resolution in R2IL, with no new op. Both frontends now agree. The method needed no Rust; the first condition-free method exposed an executor gap.

C. NARS revision

Built from two grouped sums plus Law::NarsRevision.

Independent positive and conflicting evidence: matches a floating-point oracle and OpenNARS's three reference revisions.
Repeated stamp, replayed delta, same stamp in two hypothesis layers: truth unchanged, given the producer's two marks (replay by delta_id, independence by (statement, stamp)).
Retraction: typed insufficiency, pinned by a test. A negative-weight delta gives the right answer for a stamp that was asserted, and a silently wrong one for a stamp that never was.

The missing reusable algebra is a first-occurrence mask over an ordered key, plus, for retraction, a semijoin of retraction stamps onto asserted stamps.

D. Faults

Each fault is either refused before execution with a typed error or turns its test red. A disable run was made for each guard: 13 so far, all red.

swapped lanes
schema seal
reading
population
NULL read as a value
duplicate stamps
law shape
basis of 257

A compute body has no MethodBody variant, so it cannot be written at all.

E. Cost

Measured on 1,048,576 rows at the repo's x86-64-v3 tier, with the hot paths interleaved per repetition:

path	median	allocations per call
prepare one method (cold)	0.8 µs	13
warm call, Quack	600 µs	2
warm call, R2IL	605 µs	5
direct Quack, lowered once	599 µs	0
scalar row loop	366 µs	0

Once prepared, the method layer adds no measurable cost over direct Quack. Both are about 1.6× slower than a fused scalar loop, because the program writes a mask and then sums it. That is the executor's cost, not the method layer's; fusing filter and fold is the lever.

Changes

glove-parity/src/method.rs: MethodSpec, Glove, MethodTable, PreparedMethod (with programs lowered once), the GroupSum fold, Law::NarsRevision, SchemaSeal / ReadingId / Universe.
glove-parity/src/call.rs: resolve_call.
glove-parity/examples/method_cost.rs: the Experiment E probe.
r2il/src/procs.rs: select_u32 and count.
r2il/src/lib.rs: POP_COUNT, SUM, GROUP_SUM and KEEP read a resident plane as their mask.
No new mask-risc or R2IL op.

Gates: fmt and clippy -D warnings pass on glove-parity and r2il. Tests pass for glove-parity (method 10, polyfill 12), r2il and r2il-mask-abi-probe.

Open

A FirstOfRun predicate over an ordered key, so the dedup marks become algebra.
Retraction via a semijoin onto asserted stamps.
Three-valued conditions through Quack's sql_where.
Mint method keys and the NARS concept in OGAR (feat(callcenter): add OGIT + OWL/DOLCE semantic layer foundation #357).
Fix the compute-DAG check to 256 positions before any compute body lands.
A fused filter + fold terminal.

🤖 Generated with Claude Code

https://claude.ai/code/session_011BMJnDBAfzcuQh5HeT39Rh

Summary by CodeRabbit

New Features
- Added support for preparing and calling class-qualified methods, with query results including counts, sums, grouped sums, and NAL revision truths.
- Method calls can run through either Quack or R2IL, with checks for incompatible methods and mismatched data.
- Added R2IL procedures for selecting rows by unsigned values and counting matches.
Improvements
- R2IL folds can now use resident mask planes directly for counting, summing, grouped sums, and keeping rows.

First slice of D-CMH-1 (class × method × payload → shared algebra). A MethodSpec belongs to the concept (0x0103) and names its fields in the canonical ClassView basis. MethodTable::prepare binds each spec once against one glove's claims and Binder; a call carries only numbers and runs through either Quack (Query → lower → mask-risc) or R2IL (FoldDialect → mask-risc). - required is a WideFieldMask over the canonical basis; a method prepares only when required ⊆ available (fields with an Exact/Converted claim). `duration` (Hypothesized) is refused at preparation. - An action body prepares and is refused by every pure call. - Labels resolve to a MethodKey once; only keys reach a call. - lance-graph-r2il: procs::count_eq (live plane AND U32 = v, popcount). Tests: count_billable agrees across Quack, R2IL and an oracle that reads the DTO's documented indicator values (n = 1, 64, 65, 4097); the bound query equals the one Quack binds from names; a swapped lane breaks parity on both frontends; wrong concept, unknown key, action refused. Not yet zero-friction: the body kind is a closed enum and the R2IL path calls a per-body procedure. The next commit makes the body declarative. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011BMJnDBAfzcuQh5HeT39Rh

coderabbitai · 2026-10-12T15:54:54Z

Note

Currently processing new changes in this PR. This may take a few minutes, please wait...

⚙️ Run configuration

Configuration used: Organization UI
Review profile: CHILL
Plan: Essentials
Run ID: e3528047-b8d6-4724-a3a2-0837f315de2f

📥 Commits

Reviewing files that changed from the base of the PR and between ec3fee7 and 4b52305.

📒 Files selected for processing (13)

.claude/board/entries/2026-10-12-class-method-polyfill-ledger.md
.claude/board/entries/README.md
.github/workflows/glove-parity.yml
crates/lance-graph-glove-parity/Cargo.toml
crates/lance-graph-glove-parity/examples/method_cost.rs
crates/lance-graph-glove-parity/src/call.rs
crates/lance-graph-glove-parity/src/lib.rs
crates/lance-graph-glove-parity/src/method.rs
crates/lance-graph-glove-parity/tests/method.rs
crates/lance-graph-glove-parity/tests/polyfill.rs
crates/lance-graph-r2il/src/lib.rs
crates/lance-graph-r2il/src/procs.rs
crates/lance-graph-r2il/tests/procs.rs

 _____________________
< Git gud or git out. >
 ---------------------
  \
   \   (\__/)
       (•ㅅ•)
       / 　 づ

✨ Finishing Touches

📝 Generate docstrings

Commit to this branch
Create a new PR

Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Warning

Billing warning: we have not been able to collect payment for this subscription for more than 72 hours. Please update the payment method or pay any pending invoices in Billing to avoid service interruption.

_{Comment @coderabbitai help to get the list of available commands.}

A method body is now MethodBody::Query { filter: [Cond], folds: [Fold], law }: conditions (= / !=) over code fields, folds Count and Sum. One generic binder turns any such body into Quack queries (one per fold) and the same selection into R2IL calls. src/ names no method. r2il: count_eq is replaced by the general select_u32 (live plane AND per-condition U32 equality, optionally negated) and count; POP_COUNT now counts a resident plane in place, so an empty condition list counts the live rows. Tests declare every method in the test file. Two new shapes added there only (a negated condition; Count + Sum over one selection) run on both frontends against row-loop oracles, with no change to src/. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011BMJnDBAfzcuQh5HeT39Rh

- call.rs: `SELECT <concept>::<method>(<table>) FROM <table>` resolved once into a typed MethodCall: the concept through OGAR's codebook (canonical_concept_id), the method through the table's labels, the payload checked against the FROM table and the bound table. Exactly one form is recognised; everything else is refused by name. - Binding seals: a spec carries the SchemaSeal of the canonical basis it was written for and the ReadingId it assumes; a call names the Universe of its planes. Any mismatch is refused. A basis wider than 256 positions is refused whole. - Grouped SUM fold, and Law::NarsRevision: revision of independent evidence compiled from two grouped sums and one certified finisher. The body must select the payload's first-occurrence mark, or it is refused (EvidenceNotIndependent): the algebra cannot compute that mark yet. tests/polyfill.rs: the call agrees with the same query written in Quack, with R2IL procedures and with a row loop; refusals for concept, method, payload, table, schema, reading, population, NULL and width; revision agrees with a float oracle and OpenNARS's reference points on both frontends; the repeated-stamp counterexample; the dropped out-of-domain key, pinned as a known gap. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011BMJnDBAfzcuQh5HeT39Rh

What the slice executes, what was measured, and the capability / insufficiency ledger: first-occurrence mask (NARS dedup), dropped out-of-domain group keys, NULL, the SQL frontend, unminted NARS concept and method identity, fixture reading/population ids, the fail-open compute-DAG check, unfused folds. Extension criteria for new algebra and laws. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011BMJnDBAfzcuQh5HeT39Rh

PreparedMethod now holds its Quack programs, lowered when the method is prepared (a lowering error is PrepareError::Lower); a call executes them and never lowers. `programs()` exposes them, and the tests assert they equal the program the same query written directly in Quack lowers to: no separate row evaluator sits behind a method. examples/method_cost.rs measures cold preparation, resolution and lowering apart from warm execution, with the hot paths interleaved per rep, allocations counted, and every path's answer asserted first. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011BMJnDBAfzcuQh5HeT39Rh

`hours_by_billable` (hours grouped by the billable code) is declared in tests/method.rs only, against the executor frozen at e984885. No other file changed (see this commit's stat). Result: correct on Quack against a per-group row loop at n = 1, 65, 4097. R2IL refuses it with a typed error: its GROUP_SUM (like SUM and KEEP) takes its mask only from a computed slot, and a body with no condition hands it the resident live plane. The test pins that refusal; the executor gap is closed in the next commit. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011BMJnDBAfzcuQh5HeT39Rh

SUM, GROUP_SUM and KEEP took their mask only from a computed slot; POP_COUNT already read a resident plane. A fold over a body with no condition (the live plane alone) was therefore refused. One shared resolution now serves all four: shape first (`fold_mask`, so a wrong-kind operand is still reported as wrong-kind), then the epoch for a slot (`resolve_mask`). mask-risc's terminals already took any Operand. Found by Experiment B (a method declared after the executor was frozen): `hours_by_billable` now agrees with its row loop on both frontends. r2il test: every_fold_takes_a_resident_plane_as_its_mask (failed before this change on SUM with WrongOperandKind). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011BMJnDBAfzcuQh5HeT39Rh

Replay identity (delta_id) and evidential identity (stamp) are kept apart: a replayed delta and the same stamp arriving in a second hypothesis layer leave the revised truth unchanged on both frontends, given the producer's marks; without them the raw sums move. Retraction is a pinned typed insufficiency: Law::NarsRevision does not declare it. A negative-weight delta is exact when the stamp was asserted and silently wrong when it was not; the sums cannot tell which. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011BMJnDBAfzcuQh5HeT39Rh

Seam, frozen-executor authoring, NARS replay/layers/retraction, faults, and measured cost, appended to the D-CMH-1 ledger entry. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011BMJnDBAfzcuQh5HeT39Rh

Regenerated the generated .claude/board/entries/README.md with entries_index.py, keeping this branch's curated summaries; --check passes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011BMJnDBAfzcuQh5HeT39Rh

The probe reports `avx512f` / `avx2` as compiled target features, so a timing is labelled by the arm that produced it, not by whichever cargo config happened to apply (lance-graph's default is x86-64-v3; native gives AVX-512 on this host). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011BMJnDBAfzcuQh5HeT39Rh

chatgpt-codex-connector · 2026-10-12T16:35:17Z

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-12T16:37:11.251585Z	`89fd1c6`	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 89fd1c65c9

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Review fixes (Codex on #1504): - a repeated MethodKey is refused (DuplicateKey) instead of resolving to the first entry; a repeated label is refused (DuplicateLabel) and resolves to nothing; - two proven claims mapping different native fields onto one canonical field are refused (AmbiguousClaim) instead of binding by claim order; the same claim repeated is not ambiguous; - Law::NarsRevision with a zero scale (NaN) or zero horizon (confidence exactly 1) is refused at preparation (LawParameter). Each is pinned by a test with a twin that still prepares. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011BMJnDBAfzcuQh5HeT39Rh

…#1504 Operator rulings 2026-10-15: CE64 header bits mean the same everywhere (K1 per-class readings withdrawn); 40..42 = Pearl method, 50..52 = the plasticity ladder (the Maslov pyramid), 59..63 = licence; O25 is a defect with a known fix over temporal.rs coordinates; N15 wires Pearl, Wang NAL S-P and the other families as views on the one wire. Records O22 and R1 as done, O20 and N10 as closed on main, O26 as dead code behind a non-compiling wip feature. Adds the 11 missing arc entries from the merged PR bodies, STATUS rows for D-RESIDUE-FP-0 / D-MATURATION3-0/-1, and moves D-HL-0 / D-NALP-0 to Shipped (#1496). Claude-Session: https://claude.ai/code/session_01EFw2WdKr1oxvaKCJC2ua2R

claude added 3 commits October 12, 2026 16:00

AdaWorldAPI changed the title ~~D-CMH-1: a concept's method bound once, run through Quack and R2IL~~ D-CMH-1: class-qualified methods compile into Quack algebra, with no Rust per method Oct 12, 2026

claude added 7 commits October 12, 2026 16:19

AdaWorldAPI mentioned this pull request Oct 12, 2026

build: native target-cpu by default, debug 0 everywhere #1505

Merged

AdaWorldAPI marked this pull request as ready for review October 12, 2026 16:35

chatgpt-codex-connector Bot reviewed Oct 12, 2026

View reviewed changes

Comment thread crates/lance-graph-glove-parity/src/method.rs Outdated

Comment thread crates/lance-graph-glove-parity/src/method.rs Outdated

Comment thread crates/lance-graph-glove-parity/src/method.rs Outdated

AdaWorldAPI merged commit c853cff into main Oct 12, 2026
9 of 10 checks passed

AdaWorldAPI mentioned this pull request Oct 15, 2026

O22 counterfactual verdict, R1 pin, ledger correction (CE64 as one wire contract) #1560

Merged

Navigation Menu

Repository navigation

D-CMH-1: class-qualified methods compile into Quack algebra, with no Rust per method - #1504

D-CMH-1: class-qualified methods compile into Quack algebra, with no Rust per method#1504
AdaWorldAPI merged 12 commits into
mainfrom
ccr-9f3b8075-74u47l

AdaWorldAPI commented Oct 12, 2026 •

edited by coderabbitai Bot

Loading

Uh oh!

coderabbitai Bot commented Oct 12, 2026 •

edited

Loading

Uh oh!

chatgpt-codex-connector Bot commented Oct 12, 2026 •

edited

Loading

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

AdaWorldAPI commented Oct 12, 2026 • edited by coderabbitai Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What this proves

Experiments A–E

A. The seam

B. Frozen executor

C. NARS revision

D. Faults

E. Cost

Changes

Open

Summary by CodeRabbit

Uh oh!

coderabbitai Bot commented Oct 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

chatgpt-codex-connector Bot commented Oct 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codex Review Summary

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

AdaWorldAPI commented Oct 12, 2026 •

edited by coderabbitai Bot

Loading

coderabbitai Bot commented Oct 12, 2026 •

edited

Loading

chatgpt-codex-connector Bot commented Oct 12, 2026 •

edited

Loading