Research phase · F710-A4

F710-A4 — Clause Role Blind Freeze

F710-A4 covers “Clause Role Blind Freeze”. The canonical record classifies it as candidate/partial, based on 3 source artifacts and 1 original source figure. The highlighted measurements are Generation Probe Cases Per Arm = 50, Decision = PARTIAL_ROLE_SEMANTICS, and Metrics Independent Beam Accuracy = 0.74.

F710-A4 Candidate/Partial Not recorded

StatusCandidate/Partial

VerdictNot recorded

Source figures1

Artifacts3

Purpose and research question

The objective is to document the recorded outcome for “Clause Role Blind Freeze”. The research question is whether the phase-specific checks support the canonical candidate/partial classification. The recorded measurements include Generation Probe Cases Per Arm = 50, Decision = PARTIAL_ROLE_SEMANTICS, and Metrics Independent Beam Accuracy = 0.74.

Method

F710-A4 is documented by 2 structured JSON records and 1 Markdown report and 1 original source figure. The reported table retains safe scalar measurements found in that phase package.

Documentation basis

3 source artifacts and 1 phase-specific figure are recorded for this phase.

Evidence highlights

Generation Probe Cases Per Arm: 50
Decision: PARTIAL_ROLE_SEMANTICS
Metrics Independent Beam Accuracy: 0.74
Metrics Pseudo Word Ratio: 0
Metrics Authorization Blind Accuracy: 0.9
Metrics Role Blind Accuracy: 0.9006666666666666
Metrics Latency P50 Ms: 172.225400005118
Metrics Latency P95 Ms: 210.15100000659004

Recorded evidence

Generation Probe Cases Per Arm	50
Decision	PARTIAL_ROLE_SEMANTICS
Metrics Independent Beam Accuracy	0.74
Metrics Pseudo Word Ratio	0
Metrics Authorization Blind Accuracy	0.9
Metrics Role Blind Accuracy	0.9006666666666666
Metrics Latency P50 Ms	172.225400005118
Metrics Latency P95 Ms	210.15100000659004

Public evidence and source figures

F710-A4 benchmark evidence — Learned Clause Role Control 710a4 — **Learned Clause Role Control 710a4**
Original source figure · 1760 × 800 px

These are original source figures from the existing public evidence set, not editorial HTML visualizations.

Result and interpretation

F710-A4 Clause Role Blind Freeze Not recorded

F710-A4 retains its candidate/partial classification and the recorded decision no explicit terminal verdict. The recorded measurements include Generation Probe Cases Per Arm = 50, Decision = PARTIAL_ROLE_SEMANTICS, Metrics Independent Beam Accuracy = 0.74, and Metrics Pseudo Word Ratio = 0.

Key results

Generation Probe Cases Per Arm: 50
Decision: PARTIAL_ROLE_SEMANTICS
Metrics Independent Beam Accuracy: 0.74
Metrics Pseudo Word Ratio: 0
Metrics Authorization Blind Accuracy: 0.9
Metrics Role Blind Accuracy: 0.9006666666666666
Metrics Latency P50 Ms: 172.225400005118
Metrics Latency P95 Ms: 210.15100000659004

Evidence sources

Values on this page are reproduced from the archived phase record and the source artifacts below. No new benchmark was run to create this page.

CLAUSE ROLE BLIND FREEZE 710A4 (JSON)
LEARNED CLAUSE ROLE CONTROL 710A4 (JSON)
LEARNED CLAUSE ROLE CONTROL 710A4 (MD)

Research topics

Word / World Document · Training · Evaluation