[AdaWorldAPI]

This is a read-only experiment under D-CODEGRAPH-BEHAVIOR-RULING-1. It adds no word, classid or binding, and it prepares and executes nothing. The area owner is EXO-155 (D-CGMM-0). Coordination is on the board P-EXO-18, channel EXO-232.

It touches neither impact_report nor the Codegraph crate, so it does not depend on #2001 or #2002 and conflicts with neither.

Question

A function changes. Which registered behaviours may change with it, and on what evidence?

What this adds

• .claude/plans/codegraph-behaviour-v1.md contains:
  • the ownership crosswalk: nine proposed stages mapped onto EXO-155, EXO-121, EXO-241, EXO-234, EXO-235, EXO-237, EXO-238 and EXO-95, with no new board slices;
  • the dependency order;
  • a reconciliation proposal for D-CODEGRAPH-RECON-2 Gate A + D-CRAI-REPORT-1: coverage-aware impact and a read-only review report #2001 and D-CODEGRAPH-RECON-2: coverage-aware impact report, report-rs #33 from compiler facts #2002 (§4).
• Two tests in crates/lance-graph-r2il/tests/integration/polyfill.rs:
  • They read the ordered algebra-word uses of two existing test bindings through BindStore::method_ref and admit.
  • They cross-check words_used and users_of.
  • They export the rows when CG_BEH_USES_OUT is set.
• .claude/probes/2026-11-11-cg-beh-1/ holds beh.py (words, dispatch, impact), disable_runs.py, the outputs and REPORT.md.
• A board entry, and the regenerated entries index.

Result

The trace has three hops and two kinds of evidence:

hop	read from	evidence
code → opcode	the match f arms of FoldDialect::call	source structure
opcode → word	POLYFILL[p], cross-checked against binding::EXECUTABLE	source structure
word → method, in order	method_ref + admit in a test run	DT64 registration

Measured at 317c4ffc0:

• The lens 0xFFFF_F000 has 17 words, and all 17 opcodes have a dispatch arm.
• A change at lib.rs:980 is inside the INT_S_LESS arm (lines 976 to 1018). That is word 5, used at ordinal 0 by methods 1 and 2 of class 0x7F0B_0001.
• At symbol level the one function realizes all 17 words. Three are used in this store. The other 14 print "no registered user in this uses file", never "unaffected".

Negative controls

control	result
dispatch mapping of one opcode removed	its word is UNKNOWN, its uses rows are withdrawn, every other row is unchanged
source changed after the witness was taken	STALE, exit 4, no impact row
a function whose name matches a word but is not dispatched	no edge
a symbol that dispatches nothing	"no behaviour edge (not: no impact)"
uses file missing	ARTIFACT_MISSING, exit 3

• beh.py --self-test runs 13 tests.
• disable_runs.py runs 12 cases, and all go red.
• The Rust ordering assertion goes red when the effects are read in reverse.

What it does not show

• realizes is an arm reading from source. It is not a compiler call fact and not a behavioural witness for the word's declared meaning.
• The native lowering path (binding_native::prepare_native) and everything below it are not mapped.
• No production crate registers a binding with DO effects. Foundry's can_cook_method has effects: Vec::new(), so the uses come from one test store.
• ON and WHEN are not read; only DO is joined.

For the #2001 and #2002 sessions

Plan §4 compares the two impact_report shapes field by field and proposes one. One point needs a decision either way: UnresolvedDispatch means trait dispatch in #2001 and dynamic calls in #2002. Nothing was pushed to either branch.

Checks run

• cargo test -p lance-graph-r2il --test integration: 57 passed.
• cargo clippy -p lance-graph-r2il --all-targets -- -D warnings and cargo fmt: clean.
• python3 -I beh.py --self-test and python3 -I disable_runs.py: pass.
• codegraph_index.py --check: current. The supersession index regenerates unchanged.

---

Generated by Claude Code

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

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

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

Share

• X
• Mastodon
• Reddit
• LinkedIn