[chennyiiis]

• Description: what and why

Follow-up to review feedback on #1882 (finding 1, mechanism-added-not-removed). The reviewer is right and the title of #1882 was not: unifying eleven pre-benchmark aborts left _record_round_stop and two inline rc-gated tails owning the same concern, so "record the result and decide whether the grid stops" had three implementations instead of one.

All sixteen endings now raise _VariantAborted, and the halt is data on it:

_Halt	meaning	who raises it
CONTINUE	the grid always walks past this	the two post-benchmark tails when rc == 0
ON_FAILURE	keep_going_on_failure decides	the eleven pre-benchmark aborts, a stopped round, the tails when rc != 0
STOP	the batch is over; trailing carries the rows for the variants it will not reach	a round stop that ends the batch, and the AgentX preflight abort

The rc asymmetry that TestKeepGoingAsymmetry documents - an rc == 0 failure always continues, an rc != 0 failure does not - was buried in two if rc != 0 and not keep_going_on_failure tails. It is now the difference between two enum members.

_record_round_stop becomes _abort_on_round_stop and only builds the result: it no longer appends to results, no longer extends the grid with not-run rows, and no longer answers a question its caller then re-asked.

Why not the requested _run_variant extraction. I measured both ways it can be written, and both have a hard cost:

• As a nested async def, which is the only way it avoids plumbing: the gate's function_line_findings de-folds nested definitions - I probed it, a 103-line nested function reports as its own outer.inner 103 entry - so the extraction creates a new ~900-line function-lines finding. CI refuses any baseline entry added relative to the base, so that needs a second baseline-raise. And ruff's mccabe folds nested definitions into the enclosing function, so run_grid's cyclomatic complexity would not drop at all.
• As a module-level def: roughly thirty parameters, which is the signature rule the extraction was meant to serve.

So the extraction trades one oversized unit for another, needs a waiver, and moves no complexity number. This PR instead makes "one abort path" true at a cost of 111 changed lines and no re-indent. If you still want the extraction after the above, I will open it separately - but I would rather it be the split of run_grid into named phases than a single function holding the same body.

Finding 2 (exception-control-flow) is not addressed here and I have not argued it away: the three try: ... except Exception blocks in the loop body are still there, and an abort added inside one of their bodies in future would be caught and re-recorded under the wrong class. Today every raise sits in a handler or at loop level - I re-ran the AST check and all sixteen are safe - but that is a property of the current code, not a guarantee. Closing it properly means removing the exception, which is the extraction above.

• Linked issue(s): close/fix refs

Review feedback on #1882.

• Tests: added/updated? commands run?

TestVariantAbort is kept, with the reviewer's two tidies: the setenv/base-yaml setup moves into _run, and the c0 marker assertion that TestKeepGoingAsymmetry already pins is dropped, leaving the assertion on the aborted variant's own marker.

Run locally: pytest over test_grid_runner.py, test_grid_runner_behavior_lock.py, test_grid_runner_kill_and_branches_unit.py, test_conc_sweep.py - 322 passed; plus test_baseline_warmup_double_run.py, test_agentx_preflight.py and src/hyperloom/orchestrator/actions/executors/tests - 910 passed, the 2 failures being a missing optional TraceLens import in the local environment that reproduces on main.

• Size/complexity triggers crossed: split, or why the shape is right:

unit	cyclomatic	cognitive	function-lines
run_grid	61 to 55	196 to 177	1040 to 1026

_grid_runner.py goes from 2247 lines to 2246. No new unit above the ceiling: _Halt and the extended _VariantAborted carry no branches. run_grid is still far over and this PR does not claim otherwise - what it claims is that the sixteen endings now share one path.

• If this simplifies or refactors: which mechanism was removed, which contract was preserved, and the tests that show it:

Removed: two of the three implementations of "record the result and decide whether the grid stops", the boolean _record_round_stop returned for its callers to re-test, and the two inline rc tails. Preserved: the same rows in the same order (the progress note reads results[idx], so moving the trailing rows after the report changes nothing), the same abort markers, and the same decision at every ending - pinned by TestKeepGoingAsymmetry, TestVariantAbort and the kill/branches suite.

• Observable effect:

None - refactor with nothing observable.

• Breaking changes: no

• PR addresses single concern: yes

• Root cause is upstream: no

[github-actions]

Code metrics gate: PASSED

Judged: the 3 files this change touches. Files it does not touch never fail it; anything found there is listed under Outside this change for information.

Dimension	Limit	New	Worse	Baseline out of date	Baseline entries	vs base
Cyclomatic complexity	fails > 20	0	0	0	55	+0
Cognitive complexity	fails > 30	0	0	0	203	+0
Function length (lines)	fails > 80	0	0	0	346	+0
Nested block depth	fails > 5	0	0	0	20	+0
Parameter count	fails > 5	0	0	0	94	-1
Module length (lines)	warns > 800, fails > 1200	0	0	0	49	+0
Duplicated lines per file	fails > 0	0	0	0	96	+0
Dead code (vulture)	fails > 0	0	0	0	2	+0

Check	Findings	Effect
Comments	0
English only	0
Production code imports test code	0
Repeated literals	0
A1 Private name of another package	0
A8 Flag argument	0
A9 Tuple return	0
A10 Raw vocabulary comparison	0
A11 Hidden global write	0
A12 Pure forwarder	0
A12 Single-implementation abstraction	0

Already over at the base (backlog, not this change's): 2

Unit	Dimension	Base	Now
src/hyperloom/orchestrator/actions/executors/_grid_runner.py:768 run_grid	Cognitive complexity	196	180
src/hyperloom/orchestrator/actions/executors/_grid_runner.py:768 run_grid	Cyclomatic complexity	61	56

108 findings in files this change does not touch

Outside this change (informational): 108

Unit	Dimension	Baseline	Now
src/hyperloom/orchestrator/state/shared_state.py resolve_graded_comparison	Cognitive complexity	absent	34
src/hyperloom_kb/collect/mapping.py _experience	Cognitive complexity	absent	31
src/hyperloom_kb/schema.py FieldDeclaration.__post_init__	Cognitive complexity	absent	32
src/kernelforge/cli/forge_loop_command.py _prepare_measurement_driver	Cognitive complexity	absent	36
src/kernelforge/cli/forge_loop_command.py _read_kb_warm_start	Cognitive complexity	absent	40
src/kernelforge/cli/forge_loop_command.py forge_loop	Cognitive complexity	absent	31
src/kernelforge/loop/runner.py IterationLoop._collect_session_candidate	Cognitive complexity	absent	44
src/kernelforge/loop/runner.py IterationLoop._rule_on_attempt	Cognitive complexity	absent	38
src/hyperloom/inference_optimizer/cli/__init__.py _reexport_session_environment	Function length (lines)	absent	111
src/hyperloom/inference_optimizer/cli/__init__.py _resolve_agent_lineup	Function length (lines)	absent	89
src/hyperloom/inference_optimizer/cli/__init__.py _resolve_run_topology	Function length (lines)	absent	92
src/hyperloom/inference_optimizer/cli/__init__.py _resume_session	Function length (lines)	absent	170
src/hyperloom/inference_optimizer/cli/__init__.py _seed_fresh_session	Function length (lines)	absent	164
src/hyperloom/inference_optimizer/multi_node/cli.py _restart_multinode	Function length (lines)	absent	107
src/hyperloom/inference_optimizer/multi_node/scripts/launch_multinode.py _build_parser	Function length (lines)	absent	84
src/hyperloom/inference_optimizer/multi_node/scripts/launch_multinode.py _spawn_rank_actors	Function length (lines)	absent	84
src/hyperloom_kb/sync.py GlobalSync.pull	Function length (lines)	absent	84
src/kernelforge/cli/forge_loop_command.py _RunReporter.build_result	Function length (lines)	absent	92
src/kernelforge/cli/forge_loop_command.py _prepare_measurement_driver	Function length (lines)	absent	136
src/kernelforge/cli/forge_loop_command.py _read_kb_warm_start	Function length (lines)	absent	149
src/kernelforge/loop/runner.py IterationLoop._close_iteration	Function length (lines)	absent	117
src/kernelforge/loop/runner.py IterationLoop._collect_session_candidate	Function length (lines)	absent	185
src/kernelforge/loop/runner.py IterationLoop._implementer_history	Function length (lines)	absent	112
src/kernelforge/loop/runner.py IterationLoop._rule_on_attempt	Function length (lines)	absent	102
src/kernelforge/loop/runner.py IterationLoop._source_attempt	Function length (lines)	absent	175

...and 83 more; the job log lists every row.

Verdict computed by the gate scripts in /home/runner/work/_temp/code-metrics-gate, not by this tree's copy.

_{Thresholds and sources: [tool.hyperloom.code_metrics] in pyproject.toml. Tools: complexipy 8.0.1, jscpd 5.4.1, ruff 0.16.2, vulture 2.16.}

[chaojhou]

PR #1896 -- refactor: every variant ends on the one abort path

What it does: In run_grid, the decision "record this variant's result and decide whether the grid stops" had three implementations: the _VariantAborted handler, _record_round_stop's boolean that three call sites re-tested, and two inline rc != 0 and not keep_going_on_failure tails. This PR moves every ending onto the one _VariantAborted handler. The stop decision now travels with the exception as a _Halt enum (CONTINUE / ON_FAILURE / STOP), and the not-run rows for a batch-ending stop travel as trailing. The halt mapping is the same as on base at all five rewritten endings, and nothing an operator can see changes.

Blocking issues: 1

1. [V5] The PR conflicts with main in scripts/code_metrics_baseline.txt [verified]
   Problem: meta.txt reports mergeable: CONFLICTING. git merge-tree --write-tree origin/main 9fc336e shows the only conflicted file is scripts/code_metrics_baseline.txt; _grid_runner.py and the test file merge cleanly.
   Impact: The PR cannot merge until the baseline file is reconciled, and the baseline numbers this PR lowers were measured against a tree main has since moved past.
   Action: Rebase onto main and regenerate the run_grid and module-length entries in scripts/code_metrics_baseline.txt from the rebased tree (python scripts/code_metrics.py --update-baseline).

SKIPPED: V4 -- "# PARTIAL: #1869 changes 336 files, only its first 100 were compared", and the same for #1675, #1276, #1154 and #1032. Overlap with those PRs was not checked.
SKIPPED: V2 -- pytest and ruff are not installed on the review host. Behaviour checks rely on the CI test shards and code-metrics, which are green at head 9fc336e.

Checked: _grid_runner.py (_Halt, _VariantAborted, _abort_on_round_stop, the three round-stop sites, the no-ws and invalid-report tails, the except handler, _variant_progress_note); an AST check that no abort raise sits inside a broad try body; test_grid_runner_behavior_lock.py (TestKeepGoingAsymmetry, TestVariantAbort); test_grid_runner.py (cancelled warmup tests); test_grid_runner_kill_and_branches_unit.py (AgentX preflight); main-side drift via git diff base..origin/main | Ran: git show/grep/diff, a Python AST walk; pytest SKIPPED | Base: 190604f | Head: 9fc336e

[chennyiiis]

Rebased on main and fixed what moved underneath it.

main added a parameter-count limit (fails > 5) after this branch was written, and _abort_on_round_stop took nine arguments. The seven describing the variant's run are one thing — which variant, in which slot, what its process returned, when it started, and the servers it left behind — so they now travel as _RoundAttempt. All three call sites build the record; none passes the nine by name.

baseline-raise covers the one remaining entry: _grid_runner.py module length 2089 → 2102. Those 13 lines are the record's definition and docstring — the sentence about server_survivors moved there from the function that used to take it as an argument.

Verification: pytest across the 19 test files that reference run_grid or the grid runner — 1818 passed. The one failure, test_trace_analyze_handler_dry_run_returns_structured_result, is a missing TraceLens install locally and fails the same way on the base commit.

[chennyiiis]

Rebased onto main; the one blocking issue is resolved, and a second thing moved underneath this branch while it waited.

The baseline conflict. scripts/code_metrics_baseline.txt reconciled against the rebased tree. run_grid cyclomatic reads 55 on the rebased tree, matching the entry and the 61 → 55 this PR claims.

A rule that did not exist when this was written. main added a parameter-count limit (fails > 5) and _abort_on_round_stop took nine arguments. The seven describing the variant's run — which variant, in which slot, what its process returned, when it started, and the servers it left behind — are one thing, so they travel as _RoundAttempt. All three call sites build the record; an AST check confirms none passes the nine by name.

	Value
run_grid cyclomatic	55 (baseline 55)
_abort_on_round_stop parameters	9 → 3
_grid_runner.py module length	2089 → 2102
CI	33 pass / 0 fail

The module length is the one entry --update-baseline cannot regenerate, since it only lowers: the 13 lines are _RoundAttempt's definition and docstring — the sentence about server_survivors moved there from the function that used to take it as an argument. That is what baseline-raise covers.

Verification: pytest across the 19 test files that reference run_grid or the grid runner, including test_grid_runner_behavior_lock.py — 1818 passed. The one failure is the local TraceLens gap and fails the same way on the base commit.

@chaojhou ready for another look.