CE64 ISA: strict decoder, register methods, field contracts

[AdaWorldAPI]

Reasoning chooses the operation. CausalEdge64 defines what that operation means.

Gadamer decides whether evidence is admitted; the register only computes.

Methods and signatures

method	Compose	kind
a.deduction(b, compose) -> Self	yes	total
a.induction(b, compose) -> Self	yes	total
a.abduction(b, compose) -> Self	yes	total
a.synthesis(b, compose) -> Self	yes	total
a.revision(b) -> Self	no	total, truth only: writes F and C, keeps every other field of a
a.counterfactual(b, compose) -> Result<Self, IsaFault>	—	partial, always refuses (no implementation yet)
a.intervention(b, compose) -> Result<Self, IsaFault>	—	partial, always refuses (no implementation yet)
a.execute(op, b, compose) -> Self	yes	the chain step that forward runs
a.forward(b, s, p, o) -> Result<Self, IsaFault>	yes	decode b's code, then dispatch to execute

Compose is the S/P/O payload algebra only. It never selects an opcode. Using the same operands with two different algebras changes S/P/O and leaves every other field identical (tested).

Dispatch over all 16 inference codes

raw code	result
+1	Deduction
+2	Induction
-1	Abduction
+4	Revision step
+5	Synthesis
0, -8, -2, ±3, -4, -5, ±6, ±7	IsaFault::Unsupported { mantissa }

Before this PR, ±6 and ±7 ran the Synthesis mean and came back stamped ±6 and 7. The other unsupported codes ran some other instruction and came back stamped with that instruction's code.

Field-liveness matrix

Each row is declared in isa::contracts and checked in ce64_isa_contract.rs by perturbing one operand field at a time.

instruction	computes	passes through	constant	Compose	fails when
forward / execute	S/P/O, F, C, Pearl (AND of both operands)	inference, Direction, Plasticity, all from B	W = 0, Epi5 = 0	yes	B carries an unsupported code
revision	F, C	everything else, from A	—	no	never
learn	S/P/O, F, C, Plasticity	Pearl, Direction, inference, W, Epi5, from A	—	no	never
syllogize	S/P/O, F, C, Pearl, inference	—	Direction 0, Plasticity 0b111, W 0, Epi5 0	no	returns None without a figure

forward and syllogize already wrote W and Epi5 as zero before this PR. They are now declared constants rather than an undocumented loss; the behaviour itself is unchanged.

Golden vectors (raw u64)

• Legacy capture, 126 vectors recorded from the code before this PR:
  • 71 reproduce bit for bit;
  • the 55 whose weight carries an unsupported code now refuse with exactly that code;
  • nothing else changed.
• Normative, 23 vectors:
  • run through the named methods;
  • checked against an independent f64 reference, within ±1 on F and C;
  • checked against the declared field contract.

Max-confidence revision defect

revision(c=255, c=255) returns f = c = 0. The path:

1. c = 1.0 ≥ 0.999, so evidence_weight = f32::MAX;
2. MAX + MAX = +inf;
3. inf / inf = NaN;
4. the float→u8 cast saturates NaN to 0.

A single c=255 operand gives a finite result but dominates: its frequency wins outright. The capped reference instead gives a weighted mean.

Both cases are pinned as known-bad and kept out of the normative set. The smallest fix is to cap c before evidence_weight, as the reference does; that is a separate PR.

Gadamer stays above the register

tests/gadamer_isa_revision.rs:

• ten echo encounters or closed cycles leave confidence unchanged, because the gate never calls revision;
• ten new independent roots raise it;
• revision alone would inflate an echo, which is why the gate sits above it.

isa.rs names no reasoning-policy types (enforced by a source guard).

Tests

• causal-edge: 81 lib tests plus 12 contract, 6 golden and 8 op-contract tests, all passing. Clippy passes with -D warnings.
• planner: chain_confidence, gadamer_isa_revision and moore_nars16_probe (15) pass, as do the full planner, cognitive-shader-driver and p64-bridge suites.

Every disable run below asserts that its patch applied, and each turned the listed tests red:

invariant disabled	tests red
Counterfactual decoded as Synthesis	6
forward ignores the decoded opcode	4
learn clears W	4
revision clears Epi5	1
revision formula diverges	4
known-bad case promoted to normative	2
a policy type named in isa.rs	1
counterfactual computes a result	1
Gadamer gate bypassed	2

Open

• forward's revision step (+4) composes the payload, while revision(b) does not. Whether a payload-composing revision step should exist at all is a decision this PR does not make.
• forward writes the executed code into bits 46..49. The transition-loop reading ("do not whisper the operation back into the result edge") says it should not. This predates the PR and is unchanged.
• learn mixes arithmetic with policy: archetype reassignment and the 0.9 / 0.7 freeze thresholds.
• syllogize both chooses its rule from the figure and computes it.
• The planner's truth-function copies outside the register (TruthValue, nars_infer, NarsTables) still differ from the canonical ones.

API change: forward, forward_chain, NarsEngine::forward_edge and replay_step now return Result. ReplayError gains an Isa variant.

Docs: docs/architecture/ce64-semantic-upper-half.md now separates physical encoding, declared reading and ISA operation. Board: .claude/board/entries/2026-10-08-ce64-isa-register-contract.md.

🤖 Generated with Claude Code

https://claude.ai/code/session_01EFw2WdKr1oxvaKCJC2ua2R

---

Generated by Claude Code

[coderabbitai]

Warning

Review limit reached

Your organization has reached its usage spending cap. Adjust your spending cap in the billing tab.

Next included review available in 5 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used the included review currently available. Your 61 included PR review attempts over the past 7 days set your current allowance at 1 review per hour.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: CHILL
• Plan: Essentials
• Run ID: c9184858-ad81-4cfc-9471-8ec9b3800a81

📥 Commits

Reviewing files that changed from the base of the PR and between 92c4798 and bad7b15.

📒 Files selected for processing (19)

• .claude/board/entries/2026-10-08-ce64-isa-register-contract.md
• .claude/board/entries/README.md
• crates/causal-edge/src/edge.rs
• crates/causal-edge/src/isa.rs
• crates/causal-edge/src/lib.rs
• crates/causal-edge/src/network.rs
• crates/causal-edge/src/syllogism.rs
• crates/causal-edge/tests/ce64_isa_contract.rs
• crates/causal-edge/tests/ce64_isa_golden.rs
• crates/causal-edge/tests/ce64_legacy_capture.txt
• crates/causal-edge/tests/ce64_op_contract.rs
• crates/lance-graph-planner/examples/dcr_w0_replay_budget.rs
• crates/lance-graph-planner/src/cache/nars_engine.rs
• crates/lance-graph-planner/src/cache/stage26_v3_parity.rs
• crates/lance-graph-planner/src/chain_replay.rs
• crates/lance-graph-planner/tests/chain_confidence.rs
• crates/lance-graph-planner/tests/gadamer_isa_revision.rs
• crates/lance-graph-planner/tests/moore_nars16_probe.rs
• docs/architecture/ce64-semantic-upper-half.md

• Autopilot · 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.}

[chatgpt-codex-connector]

Codex Review Summary

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

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-08T08:48:01.866480Z	9fcfb8c	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: 9fcfb8cb07

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".