refactor(orchestrator): pass phase, kernel, knowledge and loop inputs as typed objects instead of parallel keywords by lishuoshuo-amd · Pull Request #1977 · AMD-AGI/Hyperloom

lishuoshuo-amd · 2026-10-19T05:06:31Z

Description: what and why

Brings the rest of src/hyperloom/orchestrator (outside actions/, state/, trace_analysis/, which refactor(orchestrator): pass actions, state and trace-analysis inputs as typed objects instead of parallel keywords #1972 covered) under the parameter-count limit (function-args > 5). Many of these functions were helpers extracted from large phase, kernel and knowledge functions that took the caller's locals, or took the fields of one record, request or launch as separate keywords. Each now takes the object the values belong to:
- loop: CoordinatorSettings, AdvisoryServices, RunOptions, HeldLeases, RunnerLinks.
- phases: CloseArtifact, ProposalStep, AuthoredAttempt, ProgressStamp, AuthoringRetry, PhaseFlags, HistoryEntry, PhaseClock, LifecycleMark.
- kernel: KernelPatchRequest, PodPatchApply, ResultsLocation, ConcCurveOptions, MergeLabels, ConflictQuery, LandingContext and private rung / arm records for the conc sweep.
- knowledge: FrameworkOutcome, FactRecord, RemoteRecipeAudit, RecipeFilters; write_if_better takes the KBSelectionProfile its caller already holds.
- enablement, bringup, specialists, prompts, scoring, rehearsal (GenerationScan lives in its own enablement/generation_scan.py so build.py does not import revalidation.py): LaneEnd, RoundSettlement, KeepEvidence, SetupExecution, GenerationScan, ProbeTarget, BootStreams, ReconcileStores / ReconcileLinks / ReconcileLimits, ClassifiedFailure, RebenchBenchmark, RebenchServer, RunOutcome, RunHarvest, SpecialistSpawn, ReapWatch, SessionContext, PromptCycle, PromptAssets, ScorerCallKeys.
- The rehearsal stand-in ScriptedSpecialist.run takes the same SpecialistSpawn as the dispatcher it replaces, and a test pins the two signatures together.
- Existing types are reused where they are the same concept: CriticIteration in critic_agent, PhaseBudget in the predictor (its definition moves from predictor/pump.py to predictor/payload.py, the module that takes it, so the two do not import each other), KBSelectionProfile in the remote recipe store, ServingTarget for the setting script.
- Parameters no caller passes, or that the body never read, are deleted, and canonical_id_from_components (a pass-through to recipe_canonical_id with one internal caller) is removed (evidence below).
No **kwargs, untyped dict, compatibility shim or new fallback branch is added; every call site is updated. Coordinator collaborator properties, ACTION_CATALOGUE and KERNEL_REQUEST_HANDLERS entries keep their names and call shapes.

Scope: 81 functions over the limit outside the ten orchestrator files other open PRs are restructuring (none of which is touched); 70 are done. 11 keep their signatures: check_server_argv, integrate_controller_patches, _geak_revalidation_decision, PatchKB.stage_provenance, LocalRecipeStore.put_recipe, record_lifecycle_event and snapshot_source_layer because a caller lives in one of those ten files; ScriptedLaunchBackend.run because it stands in for run_with_session_kill, which those files call by keyword; KBStoreClient.put_knowledge / put_file / put_dir because _vendor/kb_store_client.py is a byte-identical upstream copy.

Gate count of functions over the parameter limit (function-args entries in scripts/code_metrics_baseline.txt): 165 on main -> 95 (-70); the cyclomatic-complexity entries and the broad-except count are unchanged.
Linked issue(s): none.
Tests: added/updated? commands run?
- Tests changed only where a signature changed; assertions are kept. Tests that bind run_task_registered, _finalize and _reap_loop directly are updated to the new signatures.
- Base-vs-head output on the same inputs is identical (wall-clock floats and tmp-path hashes normalised) for: 260 orchestration-prompt variants and the conflict-resolver prompt; machine_state decisions and rows; kb_writeback, setup-ledger and remote-recipe audit rows; the setting script; the predictor payload; recipe-KB search and audit events; apply_kernel_patch results and manifests; the attempt summary and scorer traces.
- Every changed production module imports under python:3.10-slim and python:3.11-slim.
- Locally, per test file with PYTHONPATH=src (never the full suite): every test file under orchestrator/{bringup,enablement,loop,phases,rehearsal,kernel,actions/executors}, 259 inference_optimizer test files that import a changed module, and 3 kernelforge test files. All pass except 4 TraceLens-dependent tests that fail identically on the base here (TraceLens is not installed locally).
- A whole-tree sweep for calls to every changed signature (including inspect.signature(...).bind contract tests and monkeypatch fakes) is clean.
- ruff check / ruff format --check on the changed files; python scripts/code_metrics.py --base-ref origin/main: parameter count -70, duplication and design checks 0; the remaining findings are the length items below.
Size/complexity triggers crossed: split, or why the shape is right:
- Diff budget: 3870 production Python lines changed (+2005 / -1865 in 51 production .py files). One concern (parameter count of the remaining orchestrator modules and their call sites). Each commit touches at most 5 production files and is grouped by module.
- baseline-raise: 21 length findings, each in a function or module already over its limit at the base. The lines this PR adds come from wrapping keyword arguments into a record constructor at existing call sites (an opening and closing line per record) and, for modules, from the record types defined in the module that owns them. Lines added relative to the base: functions cli/__init__.py::_run_optimize +6, integrate_patch.py::_enablement_keep_records +2 and _maybe_write_framework_kb_record +3, enablement/build.py::_enqueue_build_launch_probe +2, enablement/lane.py::maybe_rearm_enablement +4, conc_sweep.py::_sweep_one_arm_single_server +10, controller_patch_integration.py::integrate_controller_patches +2, recipe_journal.py::_record_fact_per_variant +2, framework.py::_maybe_rearm_authored_lane +1, _submit_framework_agent_candidate_for_review +1, maybe_reauthor_from_critic_feedback +2; modules cli/__init__.py +10, integrate_patch.py +7, conc_sweep.py +21, recipe_journal.py +13, loop/coordinator.py +19, loop/dispatcher.py +7, phases/framework.py +23, phases/machine_state.py +4, specialists/runner.py +5, specialists/subprocess_.py +23. The alternatives are formatting tricks, which this PR does not use.
If this simplifies or refactors: which mechanism was removed, which contract was preserved, and the tests that show it:
Removed: the parallel keyword lists and the canonical_id_from_components pass-through. Preserved: persisted state, KB, journal and ledger rows, prompts, and the name-resolved wiring, shown by the base-vs-head comparisons above.

Deleted parameters (checked across src, scripts and tests): never passed: Coordinator(sub_agent_runner=, bus_class=), run_conc_sweep(num_prompts_factor=), invoke_runtime_cli(stderr_truncate=), MockRowScanBackend(dedup_key=); never read: write_framework_record(session_dir=), LocalRecipeStore.search(prefer=) (the body did del prefer).
Observable effect: what an operator will now see, and what the old behaviour cost them (if none, why the change is unobservable):
None: no output file, state.json field, KB record, prompt, CLI flag or log line changes.
Breaking changes: yes, for Python callers only: the refactored functions take their typed objects, and canonical_id_from_components is no longer exported from hyperloom.orchestrator.knowledge.recipe_kb (use recipe_canonical_id). Every in-tree caller is updated.
PR addresses single concern: yes (parameter count of the remaining orchestrator modules).
Root cause is upstream (Magpie/TraceLens/GEAK/IntelliKit/AgentKernelArena), ticket filed: n/a.

github-actions · 2026-10-19T05:07:44Z

Code metrics gate: PASSED with waived findings (`baseline-raise`)

Judged: the 134 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	Worse	Baseline entries	vs base
Cyclomatic complexity	fails > 20	0	69	+0
Cognitive complexity	fails > 30	0	275	+0
Function length (lines)	fails > 80	11	414	-2
Nested block depth	fails > 5	0	26	+0
Parameter count	fails > 5	0	95	-70
Module length (lines)	warns > 800, fails > 1200	10	51	+0
Duplicated lines per file	fails > 0	0	97	-1
Dead code (vulture)	fails > 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

Worse than the baseline (waived by the `baseline-raise` label): 21

Unit	Dimension	Baseline	Now
`src/hyperloom/inference_optimizer/cli/__init__.py:2483` `_run_optimize`	Function length (lines)	273	279
`src/hyperloom/orchestrator/actions/executors/integrate_patch.py:3249` `IntegratePatchExecutor._enablement_keep_records`	Function length (lines)	211	213
`src/hyperloom/orchestrator/actions/executors/integrate_patch.py:4240` `IntegratePatchExecutor._maybe_write_framework_kb_record`	Function length (lines)	114	117
`src/hyperloom/orchestrator/enablement/build.py:386` `EnablementBuild._enqueue_build_launch_probe`	Function length (lines)	84	86
`src/hyperloom/orchestrator/enablement/lane.py:516` `EnablementLane.maybe_rearm_enablement`	Function length (lines)	104	108
`src/hyperloom/orchestrator/kernel/conc_sweep.py:716` `_sweep_one_arm_single_server`	Function length (lines)	328	338
`src/hyperloom/orchestrator/kernel/controller_patch_integration.py:243` `integrate_controller_patches`	Function length (lines)	257	259
`src/hyperloom/orchestrator/knowledge/recipe_journal.py:745` `RecipeJournalCollaborator._record_fact_per_variant`	Function length (lines)	114	116
`src/hyperloom/orchestrator/phases/framework.py:1250` `FrameworkPhase._maybe_rearm_authored_lane`	Function length (lines)	103	104
`src/hyperloom/orchestrator/phases/framework.py:1962` `FrameworkPhase._submit_framework_agent_candidate_for_review`	Function length (lines)	97	98
`src/hyperloom/orchestrator/phases/framework.py:2206` `FrameworkPhase.maybe_reauthor_from_critic_feedback`	Function length (lines)	131	133
`src/hyperloom/inference_optimizer/cli/__init__.py:1`	Module length (lines)	2569	2815
`src/hyperloom/orchestrator/actions/executors/integrate_patch.py:1`	Module length (lines)	4741	4748
`src/hyperloom/orchestrator/kernel/conc_sweep.py:1`	Module length (lines)	1429	1450
`src/hyperloom/orchestrator/knowledge/recipe_journal.py:1`	Module length (lines)	1785	1800
`src/hyperloom/orchestrator/loop/coordinator.py:1`	Module length (lines)	1423	1442
`src/hyperloom/orchestrator/loop/dispatcher.py:1`	Module length (lines)	1503	1557
`src/hyperloom/orchestrator/phases/framework.py:1`	Module length (lines)	3139	3167
`src/hyperloom/orchestrator/phases/machine_state.py:1`	Module length (lines)	2189	2198
`src/hyperloom/orchestrator/specialists/runner.py:1`	Module length (lines)	1520	1525
`src/hyperloom/orchestrator/specialists/subprocess_.py:1`	Module length (lines)	1747	1770

How to fix

New or worse: split the unit (extract a function, a class, a module) until it is back under the limit, or no worse than its recorded value. The baseline cannot absorb a new violation: CI refuses any baseline entry added or raised relative to the base branch.
Baseline out of date: something improved. Run python scripts/code_metrics.py --update-baseline and commit scripts/code_metrics_baseline.txt; the command only lowers or removes entries.
Override: the baseline-raise PR label waives new, worse and baseline-growth findings and the failing design checks (they stay listed here). Use it only with the reason in the PR description; it waives nothing else.
Tools for a local run: pip install ruff==0.16.2 complexipy==8.0.1 vulture==2.16 and npm install -g jscpd@5.4.1 (or point CODE_METRICS_JSCPD at another jscpd command, such as npx --yes jscpd@<version>).

Module length warning: over 800 lines, within the 1200-line failure limit (not failing): 4

Module	Lines	Limit
`src/hyperloom/orchestrator/enablement/lane.py`	892	warns > 800
`src/hyperloom/orchestrator/kernel/attempt_summary.py`	1008	warns > 800
`src/hyperloom/orchestrator/phases/close.py`	1010	warns > 800
`src/hyperloom/orchestrator/specialists/dispatch.py`	1055	warns > 800

Already over at the base (backlog, not this change's): 11

Unit	Dimension	Base	Now
`src/hyperloom/inference_optimizer/cli/__init__.py:1814` `_reexport_session_environment`	Function length (lines)	111	111
`src/hyperloom/inference_optimizer/cli/__init__.py:2376` `_resolve_agent_lineup`	Function length (lines)	89	89
`src/hyperloom/inference_optimizer/cli/__init__.py:2271` `_resolve_run_topology`	Function length (lines)	92	92
`src/hyperloom/inference_optimizer/cli/__init__.py:1927` `_resume_session`	Function length (lines)	170	170
`src/hyperloom/inference_optimizer/cli/__init__.py:2099` `_seed_fresh_session`	Function length (lines)	170	170
`src/hyperloom/orchestrator/enablement/lane.py:91` `EnablementLane._maybe_enqueue_enablement_specialist`	Function length (lines)	94	94
`src/hyperloom/orchestrator/kernel/conc_sweep.py:1187` `run_conc_sweep`	Function length (lines)	251	249
`src/hyperloom/orchestrator/knowledge/recipe_journal.py:143` `_record_config_attempts`	Function length (lines)	142	142
`src/hyperloom/orchestrator/prompts/prompt_builder.py:616` `_section_decision_framework`	Function length (lines)	104	104
`src/hyperloom/orchestrator/specialists/runner.py:559` `SpecialistRunner._prepare`	Function length (lines)	199	199
`src/hyperloom/orchestrator/prompts/prompt_builder.py:1`	Module length (lines)	1217	1212

70 findings in files this change does not touch

Outside this change (informational): 70

Unit	Dimension	Baseline	Now
`src/hyperloom/orchestrator/enablement/recipe/projections.py` `project_launch_evidence`	Cognitive complexity	absent	36
`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/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/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	91
`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	184
`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._run_supervisor_review`	Function length (lines)	absent	88
`src/kernelforge/loop/runner.py` `IterationLoop._source_attempt`	Function length (lines)	absent	173
`src/hyperloom_kb/http_service.py`	Module length (lines)	absent	1525
`src/kernelforge/cli/forge_loop_command.py`	Module length (lines)	absent	2063
`src/hyperloom/orchestrator/actions/executors/_geak_sweep.py` `sweep_via_geak`	Cognitive complexity	67	70
`src/hyperloom/orchestrator/actions/executors/explore.py` `ExploreExecutor.__call__`	Cognitive complexity	369	374

...and 45 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.}

…and close the lane from LaneEnd and RoundSettlement

… GenerationScan

…d fan it out as one PodPatchApply

…ocation

…Server, and drop the unused num_prompts_factor parameter

… land a patch with one LandingContext

…utcome and drop the unused session_dir

…emote publish audit from one RemoteRecipeAudit

…Request, and drop the pass-through canonical_id_from_components

…the caller already holds

…gs and AdvisoryServices and drop the unused sub_agent_runner and bus_class

…r holds as one HeldLeases

…one RunnerLinks

… authoring retries from one record each

…tory rows, transitions and lifecycle marks from one record each

…ready holds

…xt, PromptCycle and PromptAssets

… drop the unused stderr_truncate and dedup_key parameters

…allKeys

…re and run a rebench from RebenchBenchmark and RebenchServer

…rvest and trace a turn from its RunnerContext

…d reap it from one ReapWatch

…r signature refactor

…at takes it

…does not import revalidation

…the dispatcher takes

lishuoshuo-amd · 2026-10-19T06:35:03Z

PR #1977 -- refactor(orchestrator): pass phase, kernel, knowledge and loop inputs as typed objects instead of parallel keywords

What it does: Replaces the long keyword lists of 70 src/hyperloom/orchestrator helpers (loop, phases, kernel, knowledge, enablement, bringup, specialists, prompts, scoring) with frozen typed objects built at the call site, so the function-args gate count drops 165 -> 95; it also deletes six parameters nobody passed or read and the canonical_id_from_components pass-through. At head f2cb120 the persisted rows, prompts and name-resolved Coordinator wiring are unchanged, and the rehearsal stand-in ScriptedSpecialist.run now takes the same SpecialistSpawn as the dispatcher it replaces.

Blocking issues: none

deferred: none

Step 7 independent reader: a second reader re-ran the ScriptedSpecialist pin test against the pre-fix stand-in (fails) and the fixed head (passes), and re-swept the tree for other stand-ins mirroring a changed signature.
SKIPPED: V4 -- openprs.txt "# PARTIAL: #1032 changes 1075 files, only its first 100 were compared", overlap there unchecked
SKIPPED: V4 -- openprs.txt "# PARTIAL: #1154 changes 1232 files, only its first 100 were compared", overlap there unchecked
SKIPPED: V4 -- openprs.txt "# PARTIAL: #1276 changes 1311 files, only its first 100 were compared", overlap there unchecked
SKIPPED: V4 -- openprs.txt "# PARTIAL: #1869 changes 441 files, only its first 100 were compared", overlap there unchecked
SKIPPED: V4 -- openprs.txt "# PARTIAL: #1962 changes 205 files, only its first 100 were compared", overlap there unchecked

Checked: new commit f2cb120 (rehearsal/specialist.py:101, test_scripted_backends.py, test_bringup_round_scenario.py); earlier finding re-verified fixed (signatures of ScriptedSpecialist.run and SpecialistSubprocessDispatcher.run both (self, request: SpecialistSpawn), done-file lands in the worktree when one is given); second tree-wide sweep for stand-ins and fakes mirroring a changed signature (every def in src matched by name and parameters against the 72 changed signatures, non-test Scripted/Mock/Fake/Stub/NoOp/Null classes, monkeypatch.setattr targets and dotted-string patch paths, **_ignored/**_kwargs catch-alls in non-test code, CLI Coordinator fakes): no other mirror; tree-wide AST check of all changed defs and constructors against every call in src/scripts/examples/docs (0 stale); Coordinator/dispatcher/machine_state/write_if_better/build_request/canonical_id_from_components checks from the first pass unchanged; whole updated description re-read against the diff (51 production files, +2005/-1865 = 3870, 165 -> 95, per-module deltas, 35 commits, new ScriptedSpecialist bullet); import-edge scan base vs head (no new cycle; the new commit adds no edge) | Ran: ruff F821/F822/F823/F401/F841/F811/F632 over the 134 changed paths (clean); import of all 51 changed production modules under python:3.10-slim and python:3.11-slim (0 failures); PYTHONPATH=src pytest per file: rehearsal/test_scripted_backends (21 passed) and test_bringup_round_scenario (6 passed); the new pin test against the 0302478 stand-in in a /tmp copy (fails, as intended); first-pass files test_machine_state, test_phase_state_machine, test_dispatcher_execution_shutdown, test_coordinator_sync_helpers_coverage_unit, test_conc_sweep, test_specialist_subprocess, test_local_recipe_store, test_predictor_payload, test_patch_conflict_merge, test_apply_kernel_patch_multinode, bringup/test_reconcile (all pass, code unchanged); /tmp/rv/repro_scripted.py and a direct ScriptedSpecialist.run(SpecialistSpawn) call (no TypeError) | Base: a40eec4 | Head: f2cb120

LGTM

chaojhou

PR #1977 -- refactor(orchestrator): pass phase, kernel, knowledge and loop inputs as typed objects instead of parallel keywords

What it does: brings the rest of src/hyperloom/orchestrator (loop, phases, kernel, knowledge, enablement, bringup, specialists, prompts, scoring, rehearsal) under the parameter-count limit. 70 functions now take the record their values belong to (CoordinatorSettings / RunOptions, PhaseFlags, KernelPatchRequest, FactRecord, SpecialistSpawn and others), existing types are reused where they are the same concept (CriticIteration, PhaseBudget, KBSelectionProfile, ServingTarget), six parameters nobody passed or read and the canonical_id_from_components pass-through are deleted, and the rehearsal ScriptedSpecialist.run is pinned to the dispatcher's signature. Prompts, persisted state, KB and journal rows and CLI argv are unchanged.

Blocking issues: none

Non-blocking:

Records that only exist to pass the gate: PhaseFlags is built from the same three getters in three places in machine.py and once in close.py, and a phase_flags() query beside that state would make it one decision. rehearsal/launch.py _LaunchRequest repeats RecordedLaunch's fields, LaneEnd mostly repeats LaneTerminal, and ScorerCallKeys carries three of LLMCallSite's join keys and is copied into one field by field.
Parameters kept only for tests: GenerationScan.attempts = 2 and RecipeFilters are set by no production caller, the same ground on which this PR deletes six other parameters. _SweepRun.num_prompts_factor now always holds DEFAULT_NUM_PROMPTS_FACTOR.
The new records that callers in other packages build (CoordinatorSettings / AdvisoryServices / RunOptions, SpecialistSpawn, and RunnerLinks / BootStreams / RecipeFilters) are not in their modules' __all__, while FrameworkOutcome, ConcCurveOptions and the rebench records are.
Positional constructions of same-typed neighbours: HistoryEntry(from_phase, to_phase, reason, ...) at machine_state.py:1983 and the ResultsLocation unpack in attempt_summary.py:756; keywords would keep a later field reorder from swapping them silently. NamedTuple and frozen dataclass records are mixed within one package.
Docstrings that still name a parameter now inside a record: _record_enablement_round (stop_reason, attempt), build_execution_row (outcome), _reap_loop (progress_cb), apply_kernel_patch (snapshot_dir / patch_path / target_file); SpecialistSpawn.worktree_base describes the diff base, which is worktree_base_commit.
The breaking-changes line could name Coordinator(sub_agent_runner=, bus_class=) and the re-exported MockRowScanBackend(dedup_key=) beside canonical_id_from_components.

Checked: every hunk of the 51 production files in six module groups for field-to-keyword mapping, defaults, eager-vs-lazy reads, frozen records holding caller-mutated objects, positional order, and call sites including monkeypatch fakes and getattr dispatch; callers of the six deleted parameters and canonical_id_from_components across src, scripts, examples, docs and SKILL.md; the PhaseBudget and GenerationScan moves (no cycle; import-linter passes); description figures (function-args 165 -> 95 and the per-function and per-module line deltas match the base and head files) | Ran: base vs head on 192+ orchestration prompt variants (identical hash), Coordinator.run in three stop scenarios, 275 machine_state decision and history cases, apply_kernel_patch (dry-run, apply, shell rebuild, missing patch), conc-sweep plans, KB writeback / RecipeKB put-search-get / journal and audit rows, ladder classify and setup-ledger rows (all identical with clocks normalised); an AST sweep of every call to a changed signature against inspect.signature (0 bad keywords); git merge-tree against origin/main and the open #1957-#1965 (all clean); CI green at head. Not run: pytest (not installed here) | Base: a40eec4 | Head: f2cb120

LGTM

…ignature, before and after main's #1977 Main's #1977 (a452748) takes the scope, metric, value and objective as one KBSelectionProfile. PR #1869 tests the merge with main, where the write-back and CLOSE tests failed with a TypeError. Drop the old keywords once main is merged into multibranch-dev past a452748. Co-authored-by: Cursor <cursoragent@cursor.com>

…ain's #1977 expects Co-Authored-By: Claude <noreply@anthropic.com>

…ne KBSelectionProfile multibranch-dev now has main past #1977 (a0181e8), so the old-keyword path dc29173 kept for the merge is gone. Co-authored-by: Cursor <cursoragent@cursor.com>

lishuoshuo-amd requested a review from a team as a code owner October 19, 2026 05:06

lishuoshuo-amd added size-exception Waives the PR diff budget; added by a maintainer with a reason in the PR description baseline-raise Waives new, worse and baseline-growth code-metrics findings; reason required in the PR description labels Oct 19, 2026

github-advanced-security Bot found potential problems Oct 19, 2026

View reviewed changes

Comment thread src/hyperloom/orchestrator/predictor/payload.py Fixed

Comment thread src/hyperloom/orchestrator/enablement/build.py Fixed

Base automatically changed from refactor/args-optimizer to main October 19, 2026 05:15

lishuoshuo-amd added 22 commits October 19, 2026 13:51

refactor(bringup): run a probe from one ProbeTarget

46be401

refactor(bringup): classify a boot from one BootStreams

744c3d7

refactor(bringup): build a Reconciler from stores, links and limits

81e9150

refactor(enablement): write the setting script for one ServingTarget …

bb9cf9a

…and close the lane from LaneEnd and RoundSettlement

refactor(enablement): probe a KEEP environment from one KeepEvidence

ee5f64d

refactor(enablement): build a ledger row from one SetupExecution

3d51948

refactor(enablement): open a task row past spent generations from one…

5c27a70

… GenerationScan

refactor(kernel): apply a kernel patch from one KernelPatchRequest an…

bb92299

…d fan it out as one PodPatchApply

refactor(kernel): render an attempted-kernel row against one ResultsL…

3dfb86d

…ocation

refactor(kernel): draw a conc-sweep curve from one ConcCurveOptions

9d8de66

refactor(kernel): record and run sweep rungs from _RungNote and _Rung…

37e9be8

…Server, and drop the unused num_prompts_factor parameter

refactor(kernel): resolve a patch conflict from one ConflictQuery and…

f7a0109

… land a patch with one LandingContext

refactor(knowledge): write a framework-PR outcome from one FrameworkO…

1aa93bc

…utcome and drop the unused session_dir

refactor(knowledge): record a recipe fact from one FactRecord and a r…

c8327ef

…emote publish audit from one RemoteRecipeAudit

refactor(recipe-kb): search with RecipeFilters, read-audit from _Read…

34f26c6

…Request, and drop the pass-through canonical_id_from_components

refactor(remote-recipe): write a session from the KBSelectionProfile …

d7c6a34

…the caller already holds

refactor(coordinator): construct a Coordinator from CoordinatorSettin…

c31fb5c

…gs and AdvisoryServices and drop the unused sub_agent_runner and bus_class

refactor(coordinator): run the tick loop from one RunOptions

1e67bff

refactor(dispatcher): run a registered task with the leases the calle…

d078975

…r holds as one HeldLeases

refactor(sub-agent-runner): hold the live shared state and policy as …

a8a2441

…one RunnerLinks

refactor(close): record a close step's artifact as one CloseArtifact

cf17fd6

refactor(framework): record steps, source attempts, progress rows and…

4e3c36d

… authoring retries from one record each

lishuoshuo-amd added 12 commits October 19, 2026 13:51

refactor(machine-state): decide phases from PhaseFlags and record his…

738f07c

…tory rows, transitions and lifecycle marks from one record each

refactor(predictor): build a request from the PhaseBudget the pump al…

eaca092

…ready holds

refactor(prompts): compose the Orchestration prompt from SessionConte…

83d27fd

…xt, PromptCycle and PromptAssets

refactor(rehearsal): record a scripted launch from one _LaunchRequest

ad97bc5

refactor(roles): record review evidence from the CriticIteration, and…

611f69e

… drop the unused stderr_truncate and dedup_key parameters

refactor(scoring): attribute a scoring call's traces from one ScorerC…

82e9cf6

…allKeys

refactor(specialists): classify a retry exhaustion by ClassifiedFailu…

b522516

…re and run a rebench from RebenchBenchmark and RebenchServer

refactor(specialist-runner): finalize a run from RunOutcome and RunHa…

6ab5dfa

…rvest and trace a turn from its RunnerContext

refactor(specialists): spawn a subprocess from one SpecialistSpawn an…

13eb37d

…d reap it from one ReapWatch

chore(metrics): lower the code-metrics baseline after the orchestrato…

88c758d

…r signature refactor

refactor(predictor): define PhaseBudget beside the request builder th…

a0a594b

…at takes it

refactor(enablement): keep GenerationScan in its own module so build …

0302478

…does not import revalidation

lishuoshuo-amd force-pushed the refactor/args-orch-rest branch from 96da298 to 0302478 Compare October 19, 2026 05:54

refactor(rehearsal): let ScriptedSpecialist take the SpecialistSpawn …

f2cb120

…the dispatcher takes

chaojhou approved these changes Oct 19, 2026

View reviewed changes

chaojhou merged commit a452748 into main Oct 19, 2026
40 of 44 checks passed

chaojhou deleted the refactor/args-orch-rest branch October 19, 2026 07:01

zoroyihan7 added a commit that referenced this pull request Oct 19, 2026

fix(specialists): the agent runner hands the reaper a ReapWatch, as m…

352fc5b

…ain's #1977 expects Co-Authored-By: Claude <noreply@anthropic.com>

chaojhou mentioned this pull request Oct 22, 2026

Add optional fail-closed KTH qualification gate #1672

Open

Navigation Menu

Repository navigation

refactor(orchestrator): pass phase, kernel, knowledge and loop inputs as typed objects instead of parallel keywords - #1977

refactor(orchestrator): pass phase, kernel, knowledge and loop inputs as typed objects instead of parallel keywords#1977
chaojhou merged 35 commits into
mainfrom
refactor/args-orch-rest

lishuoshuo-amd commented Oct 19, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Oct 19, 2026 •

edited

Loading

Outside this change (informational): 70

Uh oh!

Uh oh!

Uh oh!

lishuoshuo-amd commented Oct 19, 2026

Uh oh!

chaojhou left a comment

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

lishuoshuo-amd commented Oct 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Oct 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Code metrics gate: PASSED with waived findings (baseline-raise)

Worse than the baseline (waived by the baseline-raise label): 21

How to fix

Module length warning: over 800 lines, within the 1200-line failure limit (not failing): 4

Already over at the base (backlog, not this change's): 11

Outside this change (informational): 70

Uh oh!

Uh oh!

Uh oh!

lishuoshuo-amd commented Oct 19, 2026

PR #1977 -- refactor(orchestrator): pass phase, kernel, knowledge and loop inputs as typed objects instead of parallel keywords

Uh oh!

chaojhou left a comment

Choose a reason for hiding this comment

PR #1977 -- refactor(orchestrator): pass phase, kernel, knowledge and loop inputs as typed objects instead of parallel keywords

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

lishuoshuo-amd commented Oct 19, 2026 •

edited

Loading

github-actions Bot commented Oct 19, 2026 •

edited

Loading

Code metrics gate: PASSED with waived findings (`baseline-raise`)

Worse than the baseline (waived by the `baseline-raise` label): 21