Evidence map clause
4.1 Provider port and test seams
This clause and its evidence summary follow the authored EVIDENCE-MAP.md section.
Evidence map clause · EVIDENCE-MAP.md
Map clause: Provider protocol, streaming, retries, tools, and fake provider.
Source heading anchors: spec/04-provider.md#41-provider-port-and-test-seams; spec/04-provider.md#42-chatgpt-responses-wire; spec/04-provider.md#43-streaming; spec/04-provider.md#44-error-classes; spec/04-provider.md#45-deadlines-retries-waits-fallback; spec/04-provider.md#46-logins-and-credentials; spec/04-provider.md#47-tools; spec/04-provider.md#48-fake-provider-conformance; spec/04-provider.md#481-endpoints; spec/04-provider.md#482-roles-and-turns; spec/04-provider.md#483-script-steps-json-lines
Spec revision / decision reference: v1.2 (pinned fragment locally verified); section-level D references are listed in the source classification. Last reviewed: 2026-10-07.
Type: Safety/correctness for protocol and retry behavior; empirical for claims about harness, tool surface, latency, or task outcomes.
Evidence and status: Harness comparison is INTERIM. Short and long latency records are DESCRIPTIVE client-visible timing, not task success. X-surface is INTERIM with no agent-output format claim. Grok frontier is NOT-RUN. Synthetic validation and conformance do not establish an authenticated provider comparison. Related hypotheses: H19–H38, H106–H110, H137–H138.
Unresolved measurement: Authenticated, current-version provider and tool-surface comparisons with matched task, model, effort, retry policy, request accounting, venue, and official grade; test protocol edge cases separately from task efficacy.