GitHub View workflow

OpsRoute v0.1.0

Published experiment

Operational routing under strict contracts.

Two policy families test whether a successor can choose the correct decision, tool, arguments, approval state, policy code, and reason code—without repairing invalid output.

Refund policy routing

Duplicate payments, approval thresholds, fraud review, authorization, evidence completeness, settlement state, and refund windows.

Subscription cancellation and retention

Cancellation confirmation, locked contracts, balances, pauses, retention eligibility, requester authorization, and explicit intent.

Six systems · two surfaces

Capability and resilience are not the same result.

Confirmatory evidence contains 64 clean records. Adversarial evidence contains 32 untouched stress cases.

Separate frozen surfaces. No blended score.

Semantic exactness Strict validity

Qwen base

source_base_supporting

Semantic0.0%
Strict51.6%

Qwen adapted

source_adapted_full

Semantic54.7%
Strict87.5%

OLMo untouched

target_untouched

Semantic0.0%
Strict0.0%

OLMo full retrain

target_full_retrain

Semantic79.7%
Strict100.0%

OLMo limited retrain

target_limited_retrain_10pct

Semantic65.6%
Strict90.6%

OLMo anchored transfer

target_hybrid_anchored_distillation_10

Semantic85.9%
Strict100.0%
Accessible confirmatory metric table
SystemSemantic exactnessStrict validity
Qwen base0.0%51.6%
Qwen adapted54.7%87.5%
OLMo untouched0.0%0.0%
OLMo full retrain79.7%100.0%
OLMo limited retrain65.6%90.6%
OLMo anchored transfer85.9%100.0%

Migration recommendations

The best successor depends on the constraint.

Anchored transfer leads the clean confirmatory surface. Full retraining leads adversarial semantic performance.

Read analyst rationale
Constraint profile

Minimum Direct Labels

OLMo anchored transfer

Eligible target systems: 3. Qwen remains a reference.

Constraint profile

Maximum Confirmed Capability

OLMo anchored transfer

Eligible target systems: 3. Qwen remains a reference.

Constraint profile

Maximum Adversarial Resilience

OLMo full retrain

Eligible target systems: 3. Qwen remains a reference.

Constraint profile

Minimum Complexity

OLMo full retrain

Eligible target systems: 3. Qwen remains a reference.

Constraint profile

No Source Teacher

OLMo full retrain

Eligible target systems: 2. Qwen remains a reference.

Constraint profile

Original Labels Unavailable

No viable trained migration

Pure synthetic transfer never produced a balanced trainable target, so no completed zero-direct-label migration exists in this published case.

Eligible target systems: 0. Qwen remains a reference.