fix(engine): refuse drops that would make a negating rule fail open, and reject ground negation at registration by jsam · Pull Request #253 · inputlayer/inputlayer

jsam · 2026-10-04T14:29:25Z

Intent

Keep fixing bugs and QCing the engine. (Context: the seven-role review of the SDK, messaging and value proposition verified these defects on main.) The defect in scope is the review finding that "the model proposes; the rules decide" is not yet a security boundary: an Editor key can turn a policy off by dropping a relation that a rule negates. Reproduced: with the docs rule shape check_needed(Order, Shipment) <- shipment(Order, Shipment), tool_policy(Tool, "auto"), Tool = "carrier_check", !kill_switch(Tool) and the kill switch on, ?check_needed returns 0 rows; after .rel drop kill_switch the engine replies "Relation 'kill_switch' dropped." and ?check_needed returns [["ORD-1","S-1"]] with no error and no warning - the rule fails open silently. The review asks the engine to refuse (or warn and invalidate the rule on) .rel drop of a negated relation, and to run the negation validity check at rule registration: a ground negation such as !kill_switch("carrier_check") is today accepted at registration and only rejected at query time with "Negated predicate ... shares no variables".

What Changed

.rel drop is now refused with a conflict error (new StorageError::RelationNegated) when a persistent rule negates the relation, either directly or through a rule that reads it. Before this, dropping the relation removed it and the negating rule silently failed open. A new RuleCatalog::rules_negating follows dependents to find the affected rules, and the error message names them.
The same protection now covers rule removal (StorageError::RuleNegated, conflict). .rule drop, -name, .rule drop prefix, .rule clear and .rule remove of a rule's last clause are refused when they would empty a rule that another rule negates. The check runs on the program's end state, so a program can drop a rule together with the rules that negate it, in any order. The error is attributed to the last statement in the program that edited the emptied rule.
validate_rule now rejects, at registration, a negated atom that shares no variables, such as !kill_switch("carrier_check") or !kill_switch(_). Previously these registered and then failed every query. The commands/syntax docs, the Python SDK comment and the GUI "Universal Quantification" example (which relied on !all_eng(_)) are updated to match. Unit tests and tests/statement_status_tests.rs cover the new refusals.

Risk Assessment

✅ Low: The fix round makes the change the user chose: rule_edited_by is now recorded only when change.edits_rules(), so a schema statement can no longer be blamed for emptying a negated rule. The new test fails under the old code, which blamed index 1, and passes under the fix, which blames index 0.

Testing

I built the server and client and ran an isolated server on a temp data directory. Using an Editor login over the WebSocket API, I drove five scenarios: the intent's .rel drop kill_switch repro, ground-negation registration, every rule-removal path (including transitive negation), end-state program checks with atomic rollback and statement blame, and persistence of the guard across a restart through the shipped CLI client. All passed. The change's focused integration and unit tests also pass. I saved CLI and WebSocket transcripts as evidence and removed the temp server, data and driver.

Live validation: ✅ go - 5 of 5 scenarios driven live against the product

Scenario	Result	Live	Evidence
Editor runs `.rel drop kill_switch` while check_needed negates it: refused with a conflict naming check_needed, the relation stays, and ?check_needed still returns 0 rows	✅ pass	live	s1-rel-drop-negated.txt
Registering a rule with ground negation !kill_switch("carrier_check") or !kill_switch(_) is rejected at registration ('shares no variables'); the assignment form is accepted	✅ pass	live	s2-ground-negation-registration.txt
Adversarial: `.rule drop`, `-name`, `.rule clear`, `.rule remove` of the last clause, `.rule drop prefix`, and `.rel drop` of a relation negated through a rule all refuse while the negating rule exist…	✅ pass	live	s3-rule-removal-paths.txt
Multi-statement program: a refused removal rolls back the whole program and blames the rule-editing statement, not a later schema statement; dropping together with negators or redefining in one progra…	✅ pass	live	s4-program-end-state.txt
After a server restart on the same data, the shipped CLI client's `.rel drop kill_switch` is still refused and the rule stays closed	✅ pass	live	s5-after-restart-cli.txt

Evidence: Intent repro: .rel drop of negated kill_switch refused as Editor; rule stays closed

[auth] authenticated role=editor

>>> +shipment[("ORD-1", "S-1")]
<<< {"type":"result","id":"r0","columns":["message"],"rows":[["Inserted 1 fact(s) into 'shipment'."]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":4,"row_provenance":["unknown"],"errors":[]}

>>> +tool_policy[("carrier_check", "auto")]
<<< {"type":"persistent_update","knowledge_graph":"default","relation":"shipment","operation":"insert","count":1,"timestamp_ms":1791135302298,"seq":1}
<<< {"type":"result","id":"r1","columns":["message"],"rows":[["Inserted 1 fact(s) into 'tool_policy'."]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":3,"row_provenance":["unknown"],"errors":[]}

>>> +kill_switch[("carrier_check")]
<<< {"type":"persistent_update","knowledge_graph":"default","relation":"tool_policy","operation":"insert","count":1,"timestamp_ms":1791135302303,"seq":2}
<<< {"type":"persistent_update","knowledge_graph":"default","relation":"kill_switch","operation":"insert","count":1,"timestamp_ms":1791135302308,"seq":3}
<<< {"type":"result","id":"r2","columns":["message"],"rows":[["Inserted 1 fact(s) into 'kill_switch'."]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":3,"row_provenance":["unknown"],"errors":[]}

>>> +check_needed(Order, Shipment) <- shipment(Order, Shipment), tool_policy(Tool, "auto"), Tool = "carrier_check", !kill_switch(Tool)
<<< {"type":"rule_change","knowledge_graph":"default","rule_name":"check_needed","operation":"registered","timestamp_ms":1791135302316,"seq":4}
<<< {"type":"result","id":"r3","columns":["message"],"rows":[["Rule 'check_needed' registered."]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":7,"row_provenance":["unknown"],"errors":[]}

>>> ?check_needed(O, S)
<<< {"type":"result","id":"r4","columns":[],"rows":[],"row_count":0,"total_count":0,"truncated":false,"execution_time_ms":4,"timing_breakdown":{"total_us":4727,"parse_us":202,"sip_us":166,"magic_sets_us":87,"ir_build_us":77,"optimize_us":235,"shared_views_us":0},"errors":[]}

>>> .rel drop kill_switch
<<< ERROR [conflict] Error: Cannot drop relation 'kill_switch': rule(s) check_needed negate it or a rule that reads it, so dropping it would make them derive the rows it blocks. Drop or redefine those rules first (.rule drop <name>), or delete its facts instead.

>>> .rel
<<< {"type":"result","id":"r6","columns":["message"],"rows":[["Relations:"],["  kill_switch (arity: 1, columns: [col0: any], tuples: 1)"],["  shipment (arity: 2, columns: [col0: any, col1: any], tuples: 1)"],["  tool_policy (arity: 2, columns: [col0: any, col1: any], tuples: 1)"]],"row_count":4,"total_count":4,"truncated":false,"execution_time_ms":0,"row_provenance":["unknown","unknown","unknown","unknown"],"errors":[]}

>>> ?check_needed(O, S)
<<< {"type":"result","id":"r7","columns":[],"rows":[],"row_count":0,"total_count":0,"truncated":false,"execution_time_ms":3,"timing_breakdown":{"total_us":3631,"parse_us":177,"sip_us":148,"magic_sets_us":69,"ir_build_us":48,"optimize_us":142,"shared_views_us":0},"errors":[]}

>>> -kill_switch("carrier_check")
<<< {"type":"result","id":"r8","columns":["message"],"rows":[["Deleted 1 facts from 'kill_switch'."]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":1,"row_provenance":["unknown"],"errors":[]}

>>> ?check_needed(O, S)
<<< {"type":"persistent_update","knowledge_graph":"default","relation":"kill_switch","operation":"delete","count":1,"timestamp_ms":1791135302330,"seq":5}
<<< {"type":"result","id":"r9","columns":["O","S"],"rows":[["ORD-1","S-1"]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":4,"row_provenance":["unknown"],"timing_breakdown":{"total_us":3971,"parse_us":179,"sip_us":169,"magic_sets_us":81,"ir_build_us":49,"optimize_us":158,"shared_views_us":1},"errors":[]}

>>> +kill_switch[("carrier_check")]
<<< {"type":"persistent_update","knowledge_graph":"default","relation":"kill_switch","operation":"insert","count":1,"timestamp_ms":1791135302337,"seq":6}
<<< {"type":"result","id":"r10","columns":["message"],"rows":[["Inserted 1 fact(s) into 'kill_switch'."]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":1,"row_provenance":["unknown"],"errors":[]}

>>> .rule drop check_needed
<<< {"type":"rule_change","knowledge_graph":"default","rule_name":"check_needed","operation":"dropped","timestamp_ms":1791135302344,"seq":7}
<<< {"type":"result","id":"r11","columns":["message"],"rows":[["Rule 'check_needed' dropped."]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":5,"row_provenance":["unknown"],"errors":[]}

>>> .rel drop kill_switch
<<< {"type":"schema_change","knowledge_graph":"default","entity":"kill_switch","operation":"dropped","timestamp_ms":1791135302353,"seq":8}
<<< {"type":"result","id":"r12","columns":["message"],"rows":[["Relation 'kill_switch' dropped."]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":8,"row_provenance":["unknown"],"errors":[]}

>>> .rel
<<< {"type":"result","id":"r13","columns":["message"],"rows":[["Relations:"],["  shipment (arity: 2, columns: [col0: any, col1: any], tuples: 1)"],["  tool_policy (arity: 2, columns: [col0: any, col1: any], tuples: 1)"]],"row_count":3,"total_count":3,"truncated":false,"execution_time_ms":0,"row_provenance":["unknown","unknown","unknown"],"errors":[]}

Evidence: Ground/wildcard negation rejected at registration

[auth] authenticated role=editor

>>> +kill_switch[("carrier_check")]
<<< {"type":"result","id":"r0","columns":["message"],"rows":[["Inserted 1 fact(s) into 'kill_switch'."]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":5,"row_provenance":["unknown"],"errors":[]}

>>> +g1(Order, Shipment) <- shipment(Order, Shipment), tool_policy(Tool, "auto"), !kill_switch("carrier_check")
<<< ERROR [validation] Failed to register rule: Unsafe negation in rule 'g1': Negated atom !kill_switch(...) shares no variables with positive body atoms. Negation requires at least one shared variable: bind the constant to a variable first, e.g. `T = "value", !kill_switch(T)`.

>>> +g2(Order, Shipment) <- shipment(Order, Shipment), !kill_switch(_)
<<< ERROR [validation] Failed to register rule: Unsafe negation in rule 'g2': Negated atom !kill_switch(...) shares no variables with positive body atoms. Negation requires at least one shared variable: bind the constant to a variable first, e.g. `T = "value", !kill_switch(T)`.

>>> .rule
<<< {"type":"result","id":"r3","columns":["message"],"rows":[["No rules defined."]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":0,"row_provenance":["unknown"],"errors":[]}

>>> +g3(Order, Shipment) <- shipment(Order, Shipment), T = "carrier_check", !kill_switch(T)
<<< {"type":"result","id":"r4","columns":["message"],"rows":[["Rule 'g3' registered."]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":7,"row_provenance":["unknown"],"errors":[]}

>>> ?g3(O, S)
<<< {"type":"rule_change","knowledge_graph":"default","rule_name":"g3","operation":"registered","timestamp_ms":1791135308614,"seq":10}
<<< {"type":"result","id":"r5","columns":[],"rows":[],"row_count":0,"total_count":0,"truncated":false,"execution_time_ms":3,"timing_breakdown":{"total_us":3086,"parse_us":176,"sip_us":123,"magic_sets_us":56,"ir_build_us":83,"optimize_us":160,"shared_views_us":0},"errors":[]}

>>> .rule drop g3
<<< {"type":"rule_change","knowledge_graph":"default","rule_name":"g3","operation":"dropped","timestamp_ms":1791135308625,"seq":11}
<<< {"type":"result","id":"r6","columns":["message"],"rows":[["Rule 'g3' dropped."]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":5,"row_provenance":["unknown"],"errors":[]}

Evidence: Rule drop / -rule / clear / remove last clause / drop prefix / transitive rel drop refused

[auth] authenticated role=editor

>>> +blocklist[("T9")]
<<< {"type":"result","id":"r0","columns":["message"],"rows":[["Inserted 1 fact(s) into 'blocklist'."]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":4,"row_provenance":["unknown"],"errors":[]}

>>> +order[("O1", "T1"), ("O2", "T9")]
<<< {"type":"result","id":"r1","columns":["message"],"rows":[["Inserted 2 fact(s) into 'order'."]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":3,"row_provenance":["unknown"],"errors":[]}

>>> +p_blocked(T) <- blocklist(T)
<<< {"type":"result","id":"r2","columns":["message"],"rows":[["Rule 'p_blocked' registered."]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":6,"row_provenance":["unknown"],"errors":[]}

>>> +p_allowed(O) <- order(O, T), !p_blocked(T)
<<< {"type":"result","id":"r3","columns":["message"],"rows":[["Rule 'p_allowed' registered."]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":7,"row_provenance":["unknown"],"errors":[]}

>>> ?p_allowed(O)
<<< {"type":"result","id":"r4","columns":["O"],"rows":[["O1"]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":3,"row_provenance":["unknown"],"timing_breakdown":{"total_us":3662,"parse_us":193,"sip_us":139,"magic_sets_us":71,"ir_build_us":55,"optimize_us":107,"shared_views_us":0},"errors":[]}

>>> .rule drop p_blocked
<<< ERROR [conflict] Cannot remove rule 'p_blocked': rule(s) p_allowed negate it or a rule that reads it, so removing it would make them derive the rows it blocks. Drop or redefine those rules first, or remove them together.

>>> -p_blocked
<<< ERROR [conflict] Cannot remove rule 'p_blocked': rule(s) p_allowed negate it or a rule that reads it, so removing it would make them derive the rows it blocks. Drop or redefine those rules first, or remove them together.

>>> .rule clear p_blocked
<<< ERROR [conflict] Cannot remove rule 'p_blocked': rule(s) p_allowed negate it or a rule that reads it, so removing it would make them derive the rows it blocks. Drop or redefine those rules first, or remove them together.

>>> .rule remove p_blocked 1
<<< ERROR [conflict] Cannot remove rule 'p_blocked': rule(s) p_allowed negate it or a rule that reads it, so removing it would make them derive the rows it blocks. Drop or redefine those rules first, or remove them together.

>>> .rule drop prefix p_b
<<< ERROR [conflict] Cannot remove rule 'p_blocked': rule(s) p_allowed negate it or a rule that reads it, so removing it would make them derive the rows it blocks. Drop or redefine those rules first, or remove them together.

>>> .rel drop blocklist
<<< ERROR [conflict] Error: Cannot drop relation 'blocklist': rule(s) p_allowed negate it or a rule that reads it, so dropping it would make them derive the rows it blocks. Drop or redefine those rules first (.rule drop <name>), or delete its facts instead.

>>> ?p_allowed(O)
<<< {"type":"result","id":"r11","columns":["O"],"rows":[["O1"]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":3,"row_provenance":["unknown"],"timing_breakdown":{"total_us":3160,"parse_us":180,"sip_us":172,"magic_sets_us":76,"ir_build_us":42,"optimize_us":77,"shared_views_us":0},"errors":[]}

>>> .rule
<<< {"type":"result","id":"r12","columns":["message"],"rows":[["Rules:"],["  p_allowed (1 clause(s))"],["  p_blocked (1 clause(s))"]],"row_count":3,"total_count":3,"truncated":false,"execution_time_ms":0,"row_provenance":["unknown","unknown","unknown"],"errors":[]}

>>> .rule drop prefix p_
<<< {"type":"result","id":"r13","columns":["message"],"rows":[["Dropped 2 rule(s) with prefix 'p_': p_allowed, p_blocked"]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":13,"row_provenance":["unknown"],"errors":[]}

>>> .rule
<<< {"type":"result","id":"r14","columns":["message"],"rows":[["No rules defined."]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":0,"row_provenance":["unknown"],"errors":[]}

Evidence: Program end-state check, rollback and statement blame

[auth] authenticated role=editor

>>> +blocked(T) <- blocklist(T)
... +allowed(O) <- order(O, T), !blocked(T)
<<< {"type":"result","id":"r0","columns":["message"],"rows":[["Rule 'blocked' registered."],["Rule 'allowed' registered."]],"row_count":2,"total_count":2,"truncated":false,"execution_time_ms":8,"row_provenance":["unknown","unknown"],"errors":[]}

>>> ?allowed(O)
<<< {"type":"result","id":"r1","columns":["O"],"rows":[["O1"]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":4,"row_provenance":["unknown"],"timing_breakdown":{"total_us":3766,"parse_us":206,"sip_us":140,"magic_sets_us":105,"ir_build_us":62,"optimize_us":113,"shared_views_us":0},"errors":[]}

>>> +other(X) <- order(X, _)
... .rule drop blocked
<<< {"type":"result","id":"r2","columns":["message"],"rows":[["Cannot remove rule 'blocked': rule(s) allowed negate it or a rule that reads it, so removing it would make them derive the rows it blocks. Drop or redefine those rules first, or remove them together. (rolled back: none of the program's 2 write statements 0-1 was applied)"]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":0,"row_provenance":["unknown"],"errors":[{"index":1,"code":"conflict","message":"Cannot remove rule 'blocked': rule(s) allowed negate it or a rule that reads it, so removing it would make them derive the rows it blocks. Drop or redefine those rules first, or remove them together. (rolled back: none of the program's 2 write statements 0-1 was applied)"}]}

>>> .rule
<<< {"type":"result","id":"r3","columns":["message"],"rows":[["Rules:"],["  allowed (1 clause(s))"],["  blocked (1 clause(s))"]],"row_count":3,"total_count":3,"truncated":false,"execution_time_ms":0,"row_provenance":["unknown","unknown","unknown"],"errors":[]}

>>> .rule drop blocked
... +blocked(t: string)
<<< {"type":"result","id":"r4","columns":["message"],"rows":[["Cannot remove rule 'blocked': rule(s) allowed negate it or a rule that reads it, so removing it would make them derive the rows it blocks. Drop or redefine those rules first, or remove them together. (rolled back: none of the program's 2 write statements 0-1 was applied)"]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":0,"row_provenance":["unknown"],"errors":[{"index":0,"code":"conflict","message":"Cannot remove rule 'blocked': rule(s) allowed negate it or a rule that reads it, so removing it would make them derive the rows it blocks. Drop or redefine those rules first, or remove them together. (rolled back: none of the program's 2 write statements 0-1 was applied)"}]}

>>> .rule drop blocked
... +blocked(T) <- order(_, T)
<<< {"type":"result","id":"r5","columns":["message"],"rows":[["Rule 'blocked' dropped."],["Rule 'blocked' registered."]],"row_count":2,"total_count":2,"truncated":false,"execution_time_ms":7,"row_provenance":["unknown","unknown"],"errors":[]}

>>> ?allowed(O)
<<< {"type":"result","id":"r6","columns":[],"rows":[],"row_count":0,"total_count":0,"truncated":false,"execution_time_ms":3,"timing_breakdown":{"total_us":3180,"parse_us":206,"sip_us":153,"magic_sets_us":75,"ir_build_us":52,"optimize_us":105,"shared_views_us":0},"errors":[]}

>>> .rule drop blocked
... .rule drop allowed
<<< {"type":"result","id":"r7","columns":["message"],"rows":[["Rule 'blocked' dropped."],["Rule 'allowed' dropped."]],"row_count":2,"total_count":2,"truncated":false,"execution_time_ms":5,"row_provenance":["unknown","unknown"],"errors":[]}

>>> .rule
<<< {"type":"result","id":"r8","columns":["message"],"rows":[["No rules defined."]],"row_count":1,"total_count":1,"truncated":false,"execution_time_ms":0,"row_provenance":["unknown"],"errors":[]}

Evidence: After restart, via inputlayer-client CLI

Connecting to server at http://127.0.0.1:18734...
Connected!

Server status: healthy
Authenticated as: admin
Current knowledge graph: default

> .rule
Rules:
  check_needed (1 clause(s))
> ?check_needed(O, S)
No results.
> .rel drop kill_switch
Error: Cannot drop relation 'kill_switch': rule(s) check_needed negate it or a rule that reads it, so dropping it would make them derive the rows it blocks. Drop or redefine those rules first (.rule drop <name>), or delete its facts instead.
> .rel
Relations:
  blocklist (arity: 1, columns: [col0: any], tuples: 1)
  kill_switch (arity: 1, columns: [col0: any], tuples: 1)
  order (arity: 2, columns: [col0: any, col1: any], tuples: 2)
  shipment (arity: 2, columns: [col0: any, col1: any], tuples: 1)
  tool_policy (arity: 2, columns: [col0: any, col1: any], tuples: 1)
> ?check_needed(O, S)
No results.

Pipeline

Updates from git push no-mistakes

✅ **intent** - passed

✅ No issues found.

✅ **Rebase** - passed

✅ No issues found.

🔧 **Review** - 1 issue found → auto-fixed ✅

ℹ️ src/storage_engine/program_commit.rs:145 - The fix the user already chose (record rule_edited_by only when change.edits_rules(), and add a test that a later schema statement is not blamed) is not in this commit. Lines 144-152 still record change.name() for every catalog outcome, schema changes included (CreateSchema, DefineSchema, RemoveSchema, DefineSessionSchema). Example: the program .rule drop blocked\n+blocked(t: string) runs while allowed negates blocked. The schema statement at index 1 overwrites rule_edited_by["blocked"], so the RuleNegated conflict at line 165 is reported on statement 1 instead of the .rule drop at statement 0. The map also repeats the touched-rule bookkeeping in StagedCatalog::apply (catalog_change.rs:216-222). Fix as decided: wrap the recording in if change.edits_rules() and add a statement_status test covering a schema statement placed after the rule drop. Optionally, have StagedCatalog keep the last statement index for each touched rule so the parallel map can go.

🔧 Fix applied.
✅ Re-checked - no issues remain.

✅ **Test** - passed

✅ No issues found.

Live validation: ✅ go - 5 of 5 scenarios driven live against the product

Scenario	Result	Live	Evidence
Editor runs `.rel drop kill_switch` while check_needed negates it: refused with a conflict naming check_needed, the relation stays, and ?check_needed still returns 0 rows	✅ pass	live	s1-rel-drop-negated.txt
Registering a rule with ground negation !kill_switch("carrier_check") or !kill_switch(_) is rejected at registration ('shares no variables'); the assignment form is accepted	✅ pass	live	s2-ground-negation-registration.txt
Adversarial: `.rule drop`, `-name`, `.rule clear`, `.rule remove` of the last clause, `.rule drop prefix`, and `.rel drop` of a relation negated through a rule all refuse while the negating rule exist…	✅ pass	live	s3-rule-removal-paths.txt
Multi-statement program: a refused removal rolls back the whole program and blames the rule-editing statement, not a later schema statement; dropping together with negators or redefining in one progra…	✅ pass	live	s4-program-end-state.txt
After a server restart on the same data, the shipped CLI client's `.rel drop kill_switch` is still refused and the rule stays closed	✅ pass	live	s5-after-restart-cli.txt

cargo build --bin inputlayer-server --bin inputlayer-client; started the server on a temp --data-dir and port 18734 with a bootstrap admin key
As admin: .user create ed edpw12345 editor and .kg acl grant default ed editor; all later scenarios ran as Editor ed through a Node WebSocket driver (login frame + execute frames)
Scenario 1 (the intent's repro): set up shipment, tool_policy and kill_switch facts and the check_needed rule, then ?check_needed → 0 rows, .rel drop kill_switch → refused, .rel still lists it, ?check_needed still 0 rows; a fact delete still opens it; after .rule drop check_needed the relation drops
Scenario 2: +g1(...) <- ..., !kill_switch("carrier_check") and !kill_switch(_) rejected at registration; T = "carrier_check", !kill_switch(T) accepted
Scenario 3: .rule drop p_blocked, -p_blocked, .rule clear p_blocked, .rule remove p_blocked 1, .rule drop prefix p_b, and .rel drop blocklist (negated transitively through p_blocked) all refused; ?p_allowed unchanged; .rule drop prefix p_ (drops the negator too) succeeds
Scenario 4: multi-statement programs — an unrelated edit plus .rule drop blocked is rolled back and statement 1 is blamed; .rule drop blocked plus +blocked(t: string) blames statement 0; drop plus redefine of blocked succeeds; .rule drop blocked then .rule drop allowed succeeds
Scenario 5: restarted the server on the same data dir, then ran inputlayer-client --script with .rule, ?check_needed, .rel drop kill_switch, .rel, ?check_needed
cargo test --test statement_status_tests negated and cargo test --lib -- negat (the change's focused tests)

✅ **Document** - passed

✅ No issues found.

✅ **Lint** - passed

✅ No issues found.

✅ **Push** - passed

✅ No issues found.

…haring at registration Dropping a relation that a persistent rule negates made the rule read it as empty and derive every row it blocked: `.rel drop kill_switch` turned a kill-switch policy off with no error. The drop is now refused with a conflict error naming the rules that negate the relation, directly or through rules that read it. A negated atom sharing no variable with the body (`!kill_switch("x")`, `!kill_switch(_)`) registered fine and then failed every query of the rule; registration now rejects it with the same explanation. Docs and the GUI's universal-quantification example (which used the rejected shape) are updated.

…ted rule

…time refusal

The rule-removal guard ran per statement, so a program that dropped a negated rule before the rule negating it (`.rule drop blocked` then `.rule drop allowed`) was refused and rolled back, although the error says to remove them together; `.ontology remove` drops a pack's rules in one program in pack order and hit the same refusal. Dropping and redefining a rule in one program was refused too. The check now runs once after the program's catalog changes are staged: a rule left without clauses while a remaining rule negates it is refused, blaming the last statement that edited it.

…a negated rule

…e code. Both attempts of the run died on the runner with ENOSPC (no space left on device) while building. The test build used full debuginfo on top of a restored target/ cache, and the blacksmith-4vcpu runner ran out of disk. Locally, all 83 test binaries pass. As you instructed, I changed only the Tests (fast) job in .github/workflows/ci.yml: - Added job-level env: CARGO_PROFILE_DEV_DEBUG=line-tables-only, CARGO_PROFILE_TEST_DEBUG=line-tables-only and CARGO_INCREMENTAL=0. Line numbers stay in RUST_BACKTRACE output and target/ is much smaller. - Bumped the cache key to `${{ runner.os }}-cargo-test-fast-v2-${{ hashFiles('**/Cargo.toml') }}`, so the old full-debuginfo target/ is not restored alongside the new artifacts. - Kept the existing comment explaining why there are no restore-keys, and added one line saying to bump the key version when build settings change. No engine code or tests were changed. I left the unrelated hnsw recall flake alone. Checks I ran: - Parsed the workflow YAML and confirmed the job's env and new cache key. - Ran `cargo test` on a scratch crate with cargo 1.99.0 using the same env; cargo accepted `line-tables-only` and the tests passed. - actionlint is not installed, so the workflow was not linted. I have not pushed this change, and Tests (fast) has not run with it. Pushing and confirming the CI result is left to the outer executor

jsam added 5 commits October 4, 2026 14:46

no-mistakes(review): Refuse rule drops/clears that would empty a nega…

2055744

…ted rule

no-mistakes(document): Correct SDK negation comment for registration-…

bdcd332

…time refusal

no-mistakes(review): Blame only rule-editing statements for emptying …

6d66874

…a negated rule

jsam force-pushed the fm/fix-reldrop-negated branch from a9c53b3 to 6d66874 Compare October 4, 2026 17:38

jsam merged commit bc381b2 into main Oct 4, 2026
21 checks passed

jsam deleted the fm/fix-reldrop-negated branch October 4, 2026 18:25

jsam mentioned this pull request Oct 4, 2026

Engine release tracker: checkpoints 1-2 live status #199

Closed

Navigation Menu

Repository navigation

fix(engine): refuse drops that would make a negating rule fail open, and reject ground negation at registration - #253

fix(engine): refuse drops that would make a negating rule fail open, and reject ground negation at registration#253
jsam merged 6 commits into
mainfrom
fm/fix-reldrop-negated

jsam commented Oct 4, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

jsam commented Oct 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Intent

What Changed

Risk Assessment

Testing

Pipeline

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

jsam commented Oct 4, 2026 •

edited

Loading