PRODUCTION · Technical alpha
Sign in

Report

Diagnostic report

report_synthetic_nonproduction_001 job_synthetic_nonproduction_001 synthetic/aurelius-tiny @ synthetic-revision-1 2026-01-01T00:05:00Z 2 pairs read

Dataset readout

Chosen−rejected measurements, not outcome predictions.

synthetic_helpfulness
increase 0.125

Measurement interval: unrated.

synthetic_refusal
decrease -0.075

Measurement interval: unrated.

What moved

Concept conclusions
Measure Direction Size Agreement Status
synthetic_helpfulness increase
0.125
unrated Readable
synthetic_refusal decrease
0.075
unrated Readable

Concept evidence

synthetic_helpfulness
Status Readable
Direction increase
Aggregate delta 0.125
Normalized magnitude
0.125
Distribution n=2; min=0.050; median=0.125; max=0.200
Confidence unrated
Calibration unrated
Agreement unrated · sae
Each method's read
sae res_43845903148031cc42fed9f663d064e7531778ed6627380cd12b2becd26e7130 increase · 0.125

Reads are shown separately — we never average disagreeing methods into one number.

Exemplar rows
Dataset row 0 delta 0.200
synthetic_refusal
Status Readable
Direction decrease
Aggregate delta -0.075
Normalized magnitude
0.075
Distribution n=2; min=-0.100; median=-0.075; max=-0.050
Confidence unrated
Calibration unrated
Agreement unrated · linear_probe
Each method's read
linear_probe res_bb459af1b9a6a0635aa0fc410ced104ff3ac98c64139227770a134fbc1d2f7c3 decrease · -0.075

Reads are shown separately — we never average disagreeing methods into one number.

Exemplar rows
Dataset row 1 delta -0.100

Not covered

Unreadable concepts 0
Excluded dataset rows 0
Runtime failures 0
Pinned to
Model
synthetic/aurelius-tiny@synthetic-revision-1
Dataset
dataset_synthetic_preferences_v1
Dataset digest
sha256:132d92e586d6f4bd366b22abfc86e4cafc40a39fb7ecdb3c94110c63588c61f3
Catalog version
synthetic-catalog-0.1.0
Report ID
report_synthetic_nonproduction_001
Manifest
manifest_synthetic_nonproduction_001
Aggregation policy
synthetic-unrated-policy-v0
Report schema
0.1.0
Price version
synthetic-unpriced-v0
Worker image
sha256:cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc
Primitive provenance
ConceptResult ProvenanceMethod family Artifact digestBinding
synthetic_helpfulness res_43845903148031cc42fed9f663d064e7531778ed6627380cd12b2becd26e7130 operator_curated sae sha256:fd031eaba4676cdfd522898c8f3d3ffce4e2712c93841530a51afe71030cdef3 synthetic_fixture_runtime@0.1.0 / synthetic.blocks.0.resid_post
synthetic_refusal res_bb459af1b9a6a0635aa0fc410ced104ff3ac98c64139227770a134fbc1d2f7c3 operator_curated linear_probe sha256:ba618eb5c0b46e04affaa94aed3e5fb2c3aa9ce41ac740b6557cb188abda3538 synthetic_fixture_runtime@0.1.0 / synthetic.blocks.1.resid_post
Canonical frozen manifest JSON
{"aggregation_policy_version":"synthetic-unrated-policy-v0","catalog_version":"synthetic-catalog-0.1.0","concept_set":["synthetic_helpfulness","synthetic_refusal"],"created_at":"2026-01-01T00:00:00Z","dataset":{"canonical_pair_count":2,"content_hash":"sha256:132d92e586d6f4bd366b22abfc86e4cafc40a39fb7ecdb3c94110c63588c61f3","dataset_id":"dataset_synthetic_preferences_v1","validation_report_ref":"validation_synthetic_nonproduction_001"},"fixture":{"notice":"Frozen synthetic manifest; no provider, price, calibration, or aggregation decision.","production_use":false,"synthetic":true},"manifest_ref":"manifest_synthetic_nonproduction_001","model":{"model_ref":"synthetic/aurelius-tiny","revision":"synthetic-revision-1"},"price_version":"synthetic-unpriced-v0","quote_expires_at":"2026-01-02T00:00:00Z","report_schema_version":"0.1.0","resolved_primitives":[{"concept_id":"synthetic_helpfulness","primitives":[{"artifact_digest":"sha256:fd031eaba4676cdfd522898c8f3d3ffce4e2712c93841530a51afe71030cdef3","artifact_ref":"fixtures/artifacts/synthetic-sae-helpfulness-v1.json","binding":{"architecture":{"inference":"synthetic_jumprelu","training":"synthetic_sae"},"dtype":"float32","hook_id":"synthetic.blocks.0.resid_post","model_weight_processing_kwargs":{"center_writing_weights":false},"normalize_activations":"none","runtime":"synthetic_fixture_runtime@0.1.0"},"method_family":"sae","primitive_key":{"artifact_digest":"sha256:fd031eaba4676cdfd522898c8f3d3ffce4e2712c93841530a51afe71030cdef3","binding_digest":"sha256:c72d73c86235b3c3ad2c80b6d64ba50fe6d56a462449d74a64af8e5604aef5df","curator_namespace":"aurelius.synthetic-fixtures","curator_primitive_id":"synthetic-sae-helpfulness-v1","key_type":"operator_curated","revision":"1"},"provenance":"operator_curated","result_ref":"res_43845903148031cc42fed9f663d064e7531778ed6627380cd12b2becd26e7130","scores":{"consumption_contract":null,"intrinsic":null}}]},{"concept_id":"synthetic_refusal","primitives":[{"artifact_digest":"sha256:ba618eb5c0b46e04affaa94aed3e5fb2c3aa9ce41ac740b6557cb188abda3538","artifact_ref":"fixtures/artifacts/synthetic-linear-probe-refusal-v1.json","binding":{"architecture":{"inference":"synthetic_linear_probe","training":"synthetic_linear_probe"},"dtype":"float32","hook_id":"synthetic.blocks.1.resid_post","model_weight_processing_kwargs":{"center_writing_weights":false},"normalize_activations":"none","runtime":"synthetic_fixture_runtime@0.1.0"},"method_family":"linear_probe","primitive_key":{"artifact_digest":"sha256:ba618eb5c0b46e04affaa94aed3e5fb2c3aa9ce41ac740b6557cb188abda3538","binding_digest":"sha256:4aa5c556c7f87468c8babaa7a593a3fa27589f828ba83680b8aa61798afd2358","curator_namespace":"aurelius.synthetic-fixtures","curator_primitive_id":"synthetic-linear-probe-refusal-v1","key_type":"operator_curated","revision":"1"},"provenance":"operator_curated","result_ref":"res_bb459af1b9a6a0635aa0fc410ced104ff3ac98c64139227770a134fbc1d2f7c3","scores":{"consumption_contract":null,"intrinsic":null}}]}],"schema_version":"0.1.0","worker_image_digest":"sha256:cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc"}

How to read this

These numbers are raw activation differences between your chosen and rejected responses, measured inside the model. They show what your dataset emphasizes — they don't predict what training on it will do.

Synthetic contract values only; no model-behavior or training-outcome claim.

All fixture concepts are unrated; this package defines no tier meaning or confidence estimator.

Aggregation policy: synthetic-unrated-policy-v0

Advisories

No advisories are recorded in this report.

Raw delta matrix artifact

Raw chosen−rejected deltas for every accepted dataset row and manifest-selected primitive. No report aggregation or clustering is applied to this artifact.

fixture://synthetic-nonproduction/raw-delta-matrix.json
Canonical report JSON
{"advisories":[],"concepts":[{"aggregate_delta":"0.125","calibration_tier":"unrated","concept_id":"synthetic_helpfulness","confidence":{"lower":null,"status":"unrated","upper":null},"corroboration":{"method_families":["sae"],"status":"unrated"},"direction":"increase","distribution":{"count":2,"maximum":"0.200","median":"0.125","minimum":"0.050"},"exemplar_pairs":[{"dataset_row":0,"delta":"0.200"}],"normalized_magnitude":"0.125","per_family_reads":[{"aggregate_delta":"0.125","direction":"increase","method_family":"sae","primitive_refs":["res_43845903148031cc42fed9f663d064e7531778ed6627380cd12b2becd26e7130"]}],"status":"read"},{"aggregate_delta":"-0.075","calibration_tier":"unrated","concept_id":"synthetic_refusal","confidence":{"lower":null,"status":"unrated","upper":null},"corroboration":{"method_families":["linear_probe"],"status":"unrated"},"direction":"decrease","distribution":{"count":2,"maximum":"-0.050","median":"-0.075","minimum":"-0.100"},"exemplar_pairs":[{"dataset_row":1,"delta":"-0.100"}],"normalized_magnitude":"0.075","per_family_reads":[{"aggregate_delta":"-0.075","direction":"decrease","method_family":"linear_probe","primitive_refs":["res_bb459af1b9a6a0635aa0fc410ced104ff3ac98c64139227770a134fbc1d2f7c3"]}],"status":"read"}],"coverage":{"excluded_rows":[],"runtime_failures":[],"unreadable_concepts":[]},"fixture":{"notice":"Golden schema fixture only; do not present as a real diagnostic report.","production_use":false,"synthetic":true},"generated_at":"2026-01-01T00:05:00Z","job_id":"job_synthetic_nonproduction_001","manifest":{"aggregation_policy_version":"synthetic-unrated-policy-v0","catalog_version":"synthetic-catalog-0.1.0","concept_set":["synthetic_helpfulness","synthetic_refusal"],"created_at":"2026-01-01T00:00:00Z","dataset":{"canonical_pair_count":2,"content_hash":"sha256:132d92e586d6f4bd366b22abfc86e4cafc40a39fb7ecdb3c94110c63588c61f3","dataset_id":"dataset_synthetic_preferences_v1","validation_report_ref":"validation_synthetic_nonproduction_001"},"fixture":{"notice":"Frozen synthetic manifest; no provider, price, calibration, or aggregation decision.","production_use":false,"synthetic":true},"manifest_ref":"manifest_synthetic_nonproduction_001","model":{"model_ref":"synthetic/aurelius-tiny","revision":"synthetic-revision-1"},"price_version":"synthetic-unpriced-v0","quote_expires_at":"2026-01-02T00:00:00Z","report_schema_version":"0.1.0","resolved_primitives":[{"concept_id":"synthetic_helpfulness","primitives":[{"artifact_digest":"sha256:fd031eaba4676cdfd522898c8f3d3ffce4e2712c93841530a51afe71030cdef3","artifact_ref":"fixtures/artifacts/synthetic-sae-helpfulness-v1.json","binding":{"architecture":{"inference":"synthetic_jumprelu","training":"synthetic_sae"},"dtype":"float32","hook_id":"synthetic.blocks.0.resid_post","model_weight_processing_kwargs":{"center_writing_weights":false},"normalize_activations":"none","runtime":"synthetic_fixture_runtime@0.1.0"},"method_family":"sae","primitive_key":{"artifact_digest":"sha256:fd031eaba4676cdfd522898c8f3d3ffce4e2712c93841530a51afe71030cdef3","binding_digest":"sha256:c72d73c86235b3c3ad2c80b6d64ba50fe6d56a462449d74a64af8e5604aef5df","curator_namespace":"aurelius.synthetic-fixtures","curator_primitive_id":"synthetic-sae-helpfulness-v1","key_type":"operator_curated","revision":"1"},"provenance":"operator_curated","result_ref":"res_43845903148031cc42fed9f663d064e7531778ed6627380cd12b2becd26e7130","scores":{"consumption_contract":null,"intrinsic":null}}]},{"concept_id":"synthetic_refusal","primitives":[{"artifact_digest":"sha256:ba618eb5c0b46e04affaa94aed3e5fb2c3aa9ce41ac740b6557cb188abda3538","artifact_ref":"fixtures/artifacts/synthetic-linear-probe-refusal-v1.json","binding":{"architecture":{"inference":"synthetic_linear_probe","training":"synthetic_linear_probe"},"dtype":"float32","hook_id":"synthetic.blocks.1.resid_post","model_weight_processing_kwargs":{"center_writing_weights":false},"normalize_activations":"none","runtime":"synthetic_fixture_runtime@0.1.0"},"method_family":"linear_probe","primitive_key":{"artifact_digest":"sha256:ba618eb5c0b46e04affaa94aed3e5fb2c3aa9ce41ac740b6557cb188abda3538","binding_digest":"sha256:4aa5c556c7f87468c8babaa7a593a3fa27589f828ba83680b8aa61798afd2358","curator_namespace":"aurelius.synthetic-fixtures","curator_primitive_id":"synthetic-linear-probe-refusal-v1","key_type":"operator_curated","revision":"1"},"provenance":"operator_curated","result_ref":"res_bb459af1b9a6a0635aa0fc410ced104ff3ac98c64139227770a134fbc1d2f7c3","scores":{"consumption_contract":null,"intrinsic":null}}]}],"schema_version":"0.1.0","worker_image_digest":"sha256:cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc"},"methodology":{"aggregation_policy_version":"synthetic-unrated-policy-v0","calibration_notice":"All fixture concepts are unrated; this package defines no tier meaning or confidence estimator.","claim":"Synthetic contract values only; no model-behavior or training-outcome claim."},"projection_summary":[{"calibration_tier":"unrated","concept_id":"synthetic_helpfulness","confidence":{"lower":null,"status":"unrated","upper":null},"corroboration_status":"unrated","direction":"increase","normalized_magnitude":"0.125","status":"read"},{"calibration_tier":"unrated","concept_id":"synthetic_refusal","confidence":{"lower":null,"status":"unrated","upper":null},"corroboration_status":"unrated","direction":"decrease","normalized_magnitude":"0.075","status":"read"}],"raw_delta_matrix_ref":"fixture://synthetic-nonproduction/raw-delta-matrix.json","report_id":"report_synthetic_nonproduction_001","report_schema_version":"0.1.0"}