freemo
a41fc02f11
CI / benchmark-publish (pull_request) Has been skipped
CI / lint (pull_request) Successful in 14s
CI / build (pull_request) Successful in 17s
CI / quality (pull_request) Successful in 18s
CI / security (pull_request) Successful in 35s
CI / typecheck (pull_request) Successful in 41s
CI / unit_tests (pull_request) Successful in 2m30s
CI / integration_tests (pull_request) Successful in 3m7s
CI / docker (pull_request) Successful in 54s
CI / coverage (pull_request) Successful in 5m42s
CI / lint (push) Successful in 12s
CI / build (push) Successful in 16s
CI / quality (push) Successful in 18s
CI / security (push) Successful in 33s
CI / typecheck (push) Successful in 36s
CI / benchmark-regression (push) Has been skipped
CI / unit_tests (push) Successful in 2m45s
CI / integration_tests (push) Successful in 3m11s
CI / docker (push) Successful in 46s
CI / coverage (push) Successful in 5m52s
CI / benchmark-publish (push) Successful in 17m26s
CI / benchmark-regression (pull_request) Successful in 31m25s
Implement StrategyCoordinator and FusionEngine as named facades over the existing ACMS pipeline components, providing clean public APIs for parallel strategy execution with proportional budget allocation and fragment fusion with dedup/conflict resolution/knapsack packing. Key changes: - Add StrategyCoordinator with parallel execution and confidence-based budget allocation - Add FusionEngine with URI+hash dedup, max-depth conflict resolution, greedy knapsack packing - Add budget overage guard with lowest-relevance fragment dropping - Add per-strategy max caps enforcement - Wire into existing ContextAssemblyPipeline - Add Behave BDD tests, Robot integration tests, ASV benchmarks - Add docs/reference/acms_fusion.md ISSUES CLOSED: #192
169 lines
7.9 KiB
Gherkin
169 lines
7.9 KiB
Gherkin
@acms @acms_fusion
|
|
Feature: ACMS Strategy Coordinator and Fusion Engine
|
|
As a CleverAgents developer
|
|
I want named facades for strategy coordination and fragment fusion
|
|
So that the ACMS pipeline has clean public APIs for parallel strategy
|
|
execution with proportional budget allocation and fragment fusion
|
|
with dedup, conflict resolution, and knapsack packing
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# StrategyCoordinator — Parallel execution
|
|
# ---------------------------------------------------------------------------
|
|
|
|
@fusion @coordinator
|
|
Scenario: StrategyCoordinator coordinates a single strategy
|
|
Given the fusion modules are available
|
|
And fusion test strategies with confidences 0.8 and 0.6
|
|
And fusion test fragments with token counts 50 and 50
|
|
When I coordinate with a budget of 200 tokens
|
|
Then the coordination result should contain fragments
|
|
And the coordination result should list strategies used
|
|
|
|
@fusion @coordinator
|
|
Scenario: StrategyCoordinator handles empty strategies list
|
|
Given the fusion modules are available
|
|
And fusion test fragments with token counts 50 and 50
|
|
When I coordinate with no strategies and a budget of 200 tokens
|
|
Then the coordination result should contain 0 fragments
|
|
|
|
@fusion @coordinator
|
|
Scenario: StrategyCoordinator reports circuit-broken strategies
|
|
Given the fusion modules are available
|
|
And a coordinator with a pre-broken circuit for "broken_strat"
|
|
And fusion test fragments with token counts 50 and 50
|
|
When I coordinate with the circuit-broken strategy
|
|
Then the coordination result should report circuit-broken strategies
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# StrategyCoordinator — Budget allocation proportional to confidence
|
|
# ---------------------------------------------------------------------------
|
|
|
|
@fusion @coordinator @budget
|
|
Scenario: StrategyCoordinator allocates budget proportionally
|
|
Given the fusion modules are available
|
|
And fusion test strategies with confidences 0.8 and 0.2
|
|
And fusion test fragments with token counts 50 and 50
|
|
When I coordinate with a budget of 1000 tokens
|
|
Then the allocation for the first strategy should be approximately 800
|
|
And the allocation for the second strategy should be approximately 200
|
|
And the total allocation should equal 1000
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# StrategyCoordinator — Per-strategy max caps
|
|
# ---------------------------------------------------------------------------
|
|
|
|
@fusion @coordinator @caps
|
|
Scenario: StrategyCoordinator enforces per-strategy max caps
|
|
Given the fusion modules are available
|
|
And a coordinator with per-strategy max cap of 300
|
|
And fusion test strategies with confidences 0.9 and 0.1
|
|
And fusion test fragments with token counts 50 and 50
|
|
When I coordinate with a budget of 1000 tokens using the capped coordinator
|
|
Then no strategy allocation should exceed 300 tokens
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# FusionEngine — Dedup by URI + hash
|
|
# ---------------------------------------------------------------------------
|
|
|
|
@fusion @engine @dedup
|
|
Scenario: FusionEngine deduplicates by UKO URI and content hash
|
|
Given the fusion modules are available
|
|
And fusion fragments with duplicates by URI and content
|
|
When I fuse with a budget of 500 tokens
|
|
Then the fusion result dedup count should be greater than 0
|
|
And the fusion result should contain fewer fragments than input
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# FusionEngine — Detail conflict resolution
|
|
# ---------------------------------------------------------------------------
|
|
|
|
@fusion @engine @depth
|
|
Scenario: FusionEngine resolves detail depth conflicts to max depth
|
|
Given the fusion modules are available
|
|
And fusion fragments with same UKO node at depths 2 and 5
|
|
When I fuse with a budget of 500 tokens
|
|
Then the fusion result should retain the depth 5 fragment
|
|
And the fusion result depth resolved count should be greater than 0
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# FusionEngine — Greedy knapsack packing with tie-breakers
|
|
# ---------------------------------------------------------------------------
|
|
|
|
@fusion @engine @packing
|
|
Scenario: FusionEngine packs fragments within budget using greedy knapsack
|
|
Given the fusion modules are available
|
|
And fusion fragments totaling 600 tokens
|
|
When I fuse with a budget of 400 tokens
|
|
Then the fusion result total tokens should be at most 400
|
|
And the fusion result should contain fewer fragments than input
|
|
|
|
@fusion @engine @packing
|
|
Scenario: FusionEngine uses deterministic tie-breakers
|
|
Given the fusion modules are available
|
|
And fusion fragments with equal relevance but different depths
|
|
When I fuse with a budget of 500 tokens
|
|
Then the fusion result fragments should be ordered by depth descending then token count ascending
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# FusionEngine — Budget overage guard
|
|
# ---------------------------------------------------------------------------
|
|
|
|
@fusion @engine @overage
|
|
Scenario: FusionEngine budget overage guard drops lowest-relevance fragments
|
|
Given the fusion modules are available
|
|
And a fusion engine with overage guard enabled
|
|
And fusion fragments that cause budget overage
|
|
When I fuse with the overage-prone fragments and a budget of 100 tokens
|
|
Then the fusion result total tokens should be at most 100
|
|
And the fusion result dropped by overage guard should be greater than 0
|
|
|
|
@fusion @engine @overage
|
|
Scenario: FusionEngine overage guard does not drop when within budget
|
|
Given the fusion modules are available
|
|
And a fusion engine with overage guard enabled
|
|
And fusion fragments totaling exactly 100 tokens
|
|
When I fuse with the exact fragments and a budget of 100 tokens
|
|
Then the fusion result dropped by overage guard should be 0
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# FusionEngine — Empty input
|
|
# ---------------------------------------------------------------------------
|
|
|
|
@fusion @engine
|
|
Scenario: FusionEngine handles empty input gracefully
|
|
Given the fusion modules are available
|
|
When I fuse with no fragments and a budget of 500 tokens
|
|
Then the fusion result should contain 0 fragments
|
|
And the fusion result input count should be 0
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Integration: Coordinator + FusionEngine
|
|
# ---------------------------------------------------------------------------
|
|
|
|
@fusion @integration
|
|
Scenario: StrategyCoordinator output feeds into FusionEngine
|
|
Given the fusion modules are available
|
|
And fusion test strategies with confidences 0.8 and 0.6
|
|
And fusion test fragments with token counts 50 and 50
|
|
When I coordinate then fuse with a budget of 200 tokens
|
|
Then the final fused result should contain fragments within budget
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Configuration
|
|
# ---------------------------------------------------------------------------
|
|
|
|
@fusion @config
|
|
Scenario: CoordinatorConfig defaults are sensible
|
|
Given the fusion modules are available
|
|
When I create a default CoordinatorConfig
|
|
Then the config min_useful_budget should be 64
|
|
And the config executor_timeout should be 30.0
|
|
And the config per_strategy_max_cap should be None
|
|
|
|
@fusion @config
|
|
Scenario: FusionConfig defaults are sensible
|
|
Given the fusion modules are available
|
|
When I create a default FusionConfig
|
|
Then the fusion config overage_guard_enabled should be True
|
|
And the fusion config min_fragment_tokens should be 10
|