Skip to content

D-COGNITIVE-POLYFILL P0: MORK / MeTTa / NAL / Atomese over native algebra #1944

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
AdaWorldAPI merged 1 commit into from
Nov 3, 2026
Merged

D-COGNITIVE-POLYFILL P0: MORK / MeTTa / NAL / Atomese over native algebra #1944

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
28 changes: 28 additions & 0 deletions .claude/board/entries/2026-11-03-d-cognitive-polyfill-p0.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,28 @@
# D-COGNITIVE-POLYFILL P0: four frontends, one native algebra (2026-11-03)

**Status:** measured, lab only. Probe `crates/cognitive-polyfill-probe` (own workspace); report `.claude/probes/2026-11-03-d-cognitive-polyfill/REPORT.md`. No production change.

## Measured

- The relational subset common to MORK, MeTTa, Narsese and Atomese is a ground, bounded, conjunctive one. A scoped two-hop EXISTS/COUNT over it runs as ONE mask-risc program over borrowed rows of a DR W64 tile. Results:
- 12/12 tests; 125 cases agree with an independent oracle;
- 0 bytes allocated per hot run;
- 6 source disables red, each with its anchor asserted.
- Routing a selection by global id is required: joining on raw local positions gives the wrong answer (c04).
- The Set fold cannot answer the Bag question (2 vs 3, c08). The existing Bag route is `lance_graph_xlog::bag_count` (TRACKED MISSING-3, EXO-108).
- Existence of a NAL belief is not an edge. Existence projection and truth projection disagree on the same beliefs (c09).

## Where each language leaves the subset

- MeTTa:
- D-MTP-1 covers functional relations only and refuses the fixture with `SecondFact`.
- `BindingsSet` is a multiset (hyperon `matcher.rs:883`).
- MORK: the space is a Set (`set_val`, `space.rs:1561`). D-MPS-8 builds its own `JoinDomain`, beside DR.
- Narsese: cannot spell a binary relation without the product compound (R10).
- Atomese: no Classic AtomSpace source within reach.

## Open (owner)

- **W-1:** one cold symbol → DR id step for MeTTa/MORK. Today xlog takes pre-localized `u32`, MORK builds its own `JoinDomain`, and NAL uses `u16` terms.
- **W-2:** `Guards::check` needs a `FrameStore`, but `MapTable` keeps its store private.
- EXO-46, EXO-108, R10, the NAL → edge projection policy.
3 changes: 2 additions & 1 deletion .claude/board/entries/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,11 +25,12 @@ index row, (3) no duplicate entry id. Checks 1 and 2 are deliberately
opposite directions; the stranding this convention prevents shows up in
exactly one of them, never both.

562 entries, 2026-08-06 .. 2026-11-03.
563 entries, 2026-08-06 .. 2026-11-03.

| date | entry id | finding | file |
|---|---|---|---|
| 2026-11-03 | `D-XLOG-META-WIRE-P0` | | [2026-11-03-d-xlog-meta-wire-p0.md](2026-11-03-d-xlog-meta-wire-p0.md) |
| 2026-11-03 | `D-COGNITIVE-POLYFILL` | | [2026-11-03-d-cognitive-polyfill-p0.md](2026-11-03-d-cognitive-polyfill-p0.md) |
| 2026-11-03 | `D-BAG-FOLD` | | [2026-11-03-d-bag-fold-witnessed-existential.md](2026-11-03-d-bag-fold-witnessed-existential.md) |
| 2026-11-03 | `D-BAG-FOLD-V4` | | [2026-11-03-d-bag-fold-v4-attested-lens-gated-exec.md](2026-11-03-d-bag-fold-v4-attested-lens-gated-exec.md) |
| 2026-11-03 | `D-BAG-FOLD-V3` | | [2026-11-03-d-bag-fold-v3-production-witnesses.md](2026-11-03-d-bag-fold-v3-production-witnesses.md) |
Expand Down
207 changes: 207 additions & 0 deletions .claude/probes/2026-11-03-d-cognitive-polyfill/REPORT.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,207 @@
# D-COGNITIVE-POLYFILL P0 — four frontends, one native algebra

**Date:** 2026-11-03. **Status:** LAB PROBE + SOURCE AUDIT. Nothing here changes
a production crate, ships a parser or admits a path. The probe is
`crates/cognitive-polyfill-probe` (own workspace, excluded from the main one).

**Question:** how much of MORK, MeTTa, Narsese and Atomese can lower through the
existing DR / Quack / RelAlg / R2IL / mask-risc / ndarray stack, and where does
each one first need something that stack does not have?

**Short answer:**
- Only the **ground, bounded, conjunctive relational subset** is common to all
four. It already lowers today:
- `lance-graph-xlog::support` / `bag_count` answer the chain shape
`q(Y) :- r(X,Y), s(Y,Z)`;
- this probe answers the scoped EXISTS/COUNT form in one mask-risc program.
- What each language adds beyond that subset is different semantics, not a
different engine:
- MeTTa: binding multisets and evaluation;
- Narsese: compound terms and truth laws;
- Atomese: typed nested links and Values;
- MORK: rewriting into a Set space.
- The first missing piece is the same for MeTTa and MORK: the cold step that
turns symbols into DR global ids and back. It has no home.

---

## A. Reuse map (verified in source)

| Native capability | Where | Serves |
|---|---|---|
| DR identity: `LocalMap` global↔local, `MapTable` frames, `locate` | `lance-graph-dt64/src/dr.rs:556`, `:805`, `:955` | all four (term ids) |
| Chain support + Bag count over a DR Y map | `lance-graph-xlog/src/lib.rs:255` (`support`), `:280` (`bag_count`) | MeTTa/MORK/Atomese 2-atom chains, XLOG |
| Co-addressed presence AND (leapfrog's first level), bag-count fold | `mork-pathmap-probe/src/join.rs:1-45`, `:512` (`joint`) | MORK-shaped conjunctive bodies (lab, not admitted) |
| Two-hop functional join via Quack `Semijoin` + datafrog oracle | `lance-graph-metta/examples/join_probe.rs` (D-MTP-1) | MeTTa, **functional relations only** |
| α-acyclic vs cyclic body test (GYO) | `lance-graph-relalg/src/join_body.rs:21` | routing any conjunctive body |
| Sealed-order apertures, Question / TruthAtLeast / Choice / Revision as prepared programs | `lance-graph-nal/src/lib.rs`, `src/nal.rs:159-387` | Narsese atomic statements |
| ONNX ALIAS/INLINE over native functions | `lance-graph-onnx/src/lib.rs:1-25` | ONNX numerics only; no symbolic semantics |
| Tract harvest → DR³ + operator registry (lab) | `tract-harvest-probe` (plan status: LAB PROBE IMPLEMENTED) | ONNX graphs; TFLite is PROPOSAL only (`d-tflp-…-capstone-v1/v2`) |

Re-run on this checkout (`origin/main` `ebbdb27`):
- `lance-graph-nal`: 19 passed;
- `lance-graph-metta`: 13 passed;
- `mork-pathmap-probe`: 47 passed (MORK/PathMap as oracles).

## B. Semantic compatibility matrix

Codes: CON = existing and connected · UNW = existing but unwired · LAB = lab only ·
NAT = native composition possible · NEW = a new semantic contract is required ·
OWN = owner decision · UP = upstream only · REF = refuse · NV = not verified.

| Capability | MORK | MeTTa | Narsese | Atomese | Native substrate |
|---|---|---|---|---|---|
| Identity / symbol binding | NEW: symbol → DR id step (`Tag` byte encoding, MORK `expr/src/lib.rs:91`) | NEW: same (D-MTP-2) | LAB: `u16` term addresses in `guid()`, `nal/src/lib.rs:157`; not DR ids | NEW | DR `LocalMap`: CON |
| Atomic ground facts | UP | UP (`SpaceMut::add`, hyperon-space `lib.rs:237`) | LAB (`Beliefs::seal`) | NV (no AtomSpace source in reach) | CON |
| Typed relations | UP | UP (untyped by default) | copula enum only | NEW | ClassView: CON; per-relation class check: LAB (this probe) |
| WHERE / selection | NAT | NAT | LAB (aperture = sealed key prefix) | NAT | Quack `Keep`: CON; cross-frame routing: LAB (this probe, #1943) |
| Shared-variable join (2 atoms) | LAB (D-MPS-8) | LAB (D-MTP-1, functional only) | REF: needs `(*,A,B)` compound | NAT | xlog `support`: CON |
| Repeated-variable equality `R($x,$x)` | UP | UP (`match_atoms`, matcher.rs:1050) | n/a | UP | NAT (diagonal), not built |
| Set semantics | UP (space is `PathMap<()>`, `set_val`, space.rs:1561) | — | — | NV | `Count`/`Any`: CON |
| Bag / multiple bindings | — | UP: `BindingsSet` is a `SmallVec` multiset (matcher.rs:883), GroundingSpace `ALLOW_DUPLICATION`; PeTTa backtracks (bag) | — | NV | xlog `bag_count` + TRACKED `checked_binary_i64` (EXO-108 / MISSING-3) |
| Ordered nested terms | UP | UP | REF in crate (R10) | NEW | REF |
| Hyperedges targeting edges | UP | UP | — | NEW | NEW |
| Structural unification | UP | UP (loops filtered after the match, no occurs check, matcher.rs:1084) | — | UP | REF (P2) |
| Substitution / construction | UP | UP | — | UP | NEW (it creates structure) |
| Nondeterministic evaluation | UP | UP | — | — | REF (no second interpreter) |
| Rewrite rules | UP (`exec`, space.rs:1902) | UP | — | — | PARK |
| Recursive closure | UP (`metta_calculus`) | UP | — | UP | OWN (EXO-73); relalg `μ` not built (`relalg/src/lib.rs:28`) |
| Frequency / confidence | — | — | LAB (`Beliefs` f, c lanes) | NV (TruthValue) | the I32 lanes exist |
| Evidence provenance / revision | — | — | LAB (W3: 4 evidence classes, `Revision`) | — | grouped fold: CON |
| Temporal inference | — | — | NEW | — | NV |
| Type / guard admission | — | UP (type checker) | global, no class check (by design) | NEW | dt64 `Guards`: CON; see W-2 |
| Mutations / frames | UP | UP | `seal` only | UP | `MapTable::advance`: CON |
| Native terminal result | — | — | `Count`, `Max`, revision | — | `Any`/`Count`/`Keep`/`MaskedSum`: CON |

**What is not common** (this keeps the matrix from reading as "all four are
equivalent"):
- **Narsese cannot spell the common test at all.** A binary relation needs
the product compound `<(*, A, B) --> R>`, which the NAL crate refuses as
ledger R10 (`nal/src/lib.rs:91`). `<A-->B>, <B-->C> ⊢ <A-->C>` is deduction:
it carries a truth value, so it is not an existence test.
- **Classic Atomese is not reachable in source here.** No OpenCog AtomSpace
code exists in the four upstream repos. The only Atomese-shaped text is
MeTTa-encoded: `hyperon-experimental/python/tests/scripts/a2_opencoggy.metta:8-12`.

## C. P0 falsifier — results

**Fixture:**
- facts `R(A,B) R(B,C) R(C,D) R(A,E)`, with one DR W64 tile `[A,B,C,D,E]`;
- MeTTa `(, (R A $y) (R $y C))`, MORK `(exec P (, (R A $y) (R $y C)) …)` and
the Atomese projection all hand-lower to one `Pattern`;
- native arm: `Any`/`Count`/`Keep` over `out[a] ∧ inn[c] ∧ W`, one mask-risc
program over borrowed rows.

**Oracle:** an independent nested loop over the borrowed fact slice.

**Result: 12/12 tests green** (`cargo test --manifest-path
crates/cognitive-polyfill-probe/Cargo.toml`). clippy `-D warnings` and fmt are
clean.

| # | check | result |
|---|---|---|
| 1 | 125 cases (5×5 pairs × 5 WHERE subsets), `Any` and `Count` vs oracle | agree; both answers present |
| 2 | delete `R(B,C)` | A→?→C flips to false |
| 3 | delete `R(C,D)` | A→?→C and A→?→D unchanged; control B→?→D flips |
| 4 | B at local 2 in S1 and local 0 in S2; relation tile has B at 1 | both route to bit 1, answer true; **raw local join gives false** (bit 2 is C there); an unheld id reported as `dropped` |
| 5 | DR round trip `global(local(g)) == g` | holds; an unmapped id is refused |
| 6 | foreign frame on ingest, prepare, route | `Stale` each; control routes |
| 7 | wrong relation; scope via dt64 `Guards` | `WrongRelation`; `ScopeChanged` |
| 8 | `R(A,B)`×2, `R(B,C)`, `R(A,E)`, `R(E,C)` | Set 2, Bag 3: the Set fold cannot answer Bag |
| 9 | NAL beliefs `<A→B> f=1`, `<B→C> f=0 c=0.9` | existence projection says A→?→C true, truth projection false; a confidence-only revision moves no edge |
| 10 | free endpoint, `R($x,$x)`, nested / Narsese product, two relations, one atom | each refused with its own reason |
| 11 | program without `inn[c]` | A→?→D: false becomes true |
| 12 | 1000 hot runs | 0 bytes allocated (per-thread counter) |
| 13 | no copied upstream index | structural: planes are `&relation.out[i..i+1]` / `&relation.inn[..]`; `inn` is a derived index built at ingest (cost noted) |
| 14 | no second evaluator | structural: `execute_into` is the only evaluator called |

**Source disable runs.** Each anchor was asserted present before patching, and
each patch was reverted:

| disable | result |
|---|---|
| join operand | 6 tests red |
| selection routing | c04 red |
| repeated-variable refusal | c10 red |
| Bag multiplicity | c08 red |
| WHERE operand | c01, c04 red |
| frame check in `prepare` | c06 red |

**First failure per language (where the common subset ends):**
- **MeTTa:**
- D-MTP-1's `Relation2::ingest` refuses this fixture with
`SecondFact { subject: A }` (`join_probe.rs:61`): the D-MTP-1 path covers
functional relations only.
- The general route is `xlog::support`, which takes `&[(u32,u32)]` tuples
and already-local ids.
- Missing: the Space adapter that maps `Space::query`'s `BindingsSet`
multiset onto a Set/Bag choice (D-MTP-2).
- **MORK:** the symbol → DR id step. D-MPS-8's `JoinTable` builds its own
value-sorted `JoinDomain`; that is a second identity space beside DR, which
D-RMAP-1 already flags as unadmitted.
- **Narsese:** the product compound term (R10). Atomic inheritance questions
already run natively.
- **Atomese:** no source to verify against; only a typed-projection proposal
is possible.

## D. Findings that need an owner

- **W-1 (wire): one term-resolution step for symbol frontends.** MeTTa and MORK
symbols need a cold, once-per-frame map `symbol → DR global → local`, with
the inverse for binding emission.
- Today this is done three different ways: xlog takes pre-localized
`u32`; D-MPS-8 builds its own `JoinDomain`; NAL uses `u16` terms.
- Owner: the DR contract (EXO-46) and the GPT-07 proposal (OGAR owns the
codebook).
- **W-2 (small gap): `Guards::check` needs a `FrameStore`, but `MapTable`'s
store is private** (`dr.rs:806`). A frame sealed from a `MapTable` therefore
cannot be guard-checked for Scope against its own store. The probe checks
scope on a separate store (c07).
- **W-3: routing a selection across frames is per frame, not per query.** This
is the D-GLOBAL-WHERE finding (#1943); c04 re-confirms it from the
frontend side.
- **Set vs Bag must be declared per frontend, not inferred:**
- MORK's space is a Set;
- Hyperon's default space and PeTTa's `match` are Bags;
- NAL evidence is neither: it is weighted, with its own duplicate rule.
- EXO-108 owns the Bag product.
- **NAL → relation projection is an owner decision** (c09). Existence of a
belief is not an edge; any threshold is policy.

## E. Minimal roadmap (ranked)

1. **WIRE, low risk.** Bind the MeTTa/MORK 2-atom chain to the existing
`lance_graph_xlog::support` / `bag_count`, after a cold symbol→DR step
(W-1). Owner: DR + D-MTP-2.
- Falsifier: c01–c08 rerun against xlog instead of this probe's program,
plus Hyperon's golden corpus.
2. **PROBE.** The D-MTP-2 `LanceSpace`, read-only and non-recursive: does
`Space::query` receive whole conjunctions (`complex_query`,
hyperon-space `lib.rs:340`) or single conjuncts? Report the multiset vs
Set choice.
3. **PROBE (P2).** Bounded structural matching (`f(X, g(X))`): can a cold
matcher lower its surviving constraints to DR relations? No native
unification without an owner decision.
4. **OWNER DECISION.** Narsese compound terms (R10), and the NAL → edge
projection policy.
5. **PARK.**
- recursion (EXO-73);
- rewriting / `exec`;
- nondeterministic evaluation;
- Classic AtomSpace;
- TFLite (PROPOSAL only).

**Verdicts:**
- **REUSE NOW:** DR `LocalMap`/`MapTable`; xlog `support`/`bag_count`; Quack
`Any`/`Count`/`Keep`; relalg `is_alpha_acyclic`; NAL Question/Choice/Revision;
ONNX aliases.
- **WIRE:** symbol → DR term resolution (W-1); MeTTa/MORK chain → xlog.
- **PROBE:** `LanceSpace` multiset boundary; bounded structural matching.
- **OWNER DECISION:** EXO-46 selection sharing; EXO-108 Bag product; R10
compounds; NAL projection; W-2.
- **PARK:** recursion, rewrite, evaluation, AtomSpace, TFLite runtime.
- **SKIP:**
- a second term store;
- a copied AtomSpace;
- a new interpreter;
- a mask-risc opcode for AND/OR/COUNT, which already exist.
3 changes: 3 additions & 0 deletions Cargo.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -138,6 +138,9 @@ exclude = [
# D-MTP-1: its own workspace root (datafrog oracle is a git dev-dependency).
# `cargo test --manifest-path crates/lance-graph-metta/Cargo.toml`.
"crates/lance-graph-metta",
# D-COGNITIVE-POLYFILL P0 lab probe; own workspace root.
# `cargo test --manifest-path crates/cognitive-polyfill-probe/Cargo.toml`.
"crates/cognitive-polyfill-probe",
"crates/symbiont",
# OGAR (Open Graph of Active Record) activation crate — re-exports OGAR's full
# AR surface (ogar-vocab Class/codebook + ogar-class-view impl ClassView +
Expand Down
20 changes: 20 additions & 0 deletions crates/cognitive-polyfill-probe/Cargo.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,20 @@
[package]
name = "cognitive-polyfill-probe"
version = "0.0.0"
edition = "2021"
publish = false
description = "D-COGNITIVE-POLYFILL P0 lab falsifier: the relational subset common to MORK, MeTTa, Narsese and Atomese (a scoped two-hop existential over ground facts) run through existing DR addressing (lance-graph-dt64) and one mask-risc program. Lab only; nothing here is wired into Quack, R2IL or production."

# Its OWN workspace root, like lance-graph-nal / lance-graph-metta. Verify via
# cargo test --manifest-path crates/cognitive-polyfill-probe/Cargo.toml
[workspace]

[dependencies]
# DR identity: LocalMap (global <-> local), MapTable frames, Guards/Scope.
lance-graph-dt64 = { path = "../lance-graph-dt64" }
# The only bulk engine.
lance-graph-mask-risc = { path = "../lance-graph-mask-risc" }

# One ndarray package identity: point mask-risc's git coordinate at the sibling.
[patch."https://github.com/AdaWorldAPI/ndarray"]
ndarray = { path = "../../../ndarray" }
Loading
Loading