recordUntimed records an attempt with zero timestamps. These tests are about numbering, outcome and aggregation, where the derived durations are nois

stripANSI removes ANSI escape sequences from a string so tests can assert against plain text content.

rawRequest builds an *http.Request with the given body and context for middleware tests.

CompletedCount returns the number of file-level checkpoints replay recovered. Review reuse is narrower — see ReusableItem — while scan, having no mani

ContextWithSessionKey returns a context carrying the given session key. Review and scan runs bind their session history's SessionID at the top of Run

Finalize decides the request-level outcome for m exactly once. The three inputs are the recorded attempts, the error the logical request returned to

ModelListContains reports whether target matches any entry in models (trimmed).

ReadLines returns a window of lines from the file plus the total line count. startLine is 1-based; maxLines is the maximum number of lines to collect.

SetInput sets the frozen, resolved input identity (including the mandatory Mode). Finalize rejects an invalid or missing Mode.

runGitTest runs a git command in dir and fails the test on error.

GetDiff returns all changes as parsed model.Diff structs.

LoadResumeState replays a previous session JSONL into a fingerprint index. A record that cannot be parsed fails the load: with nothing to arbitrate co

LoadSession fully parses a JSONL file into a ViewSession.

MarkFailed records that the item failed. class must be a valid, specific failure class — an empty or unknown-string class is rejected rather than down

SessionFilePath returns the JSONL path for a persisted session.

SetError records an error for this task record, writes an llm_error entry to the JSONL stream, and increments the session-level LLM failure counter.

IsEnabled returns true when telemetry has been initialized with exporters.

SessionTaskKey derives the prompt-cache affinity key for one task conversation within a session. Prompt caches match on prefixes, and OCR's task types

SetResponse records the LLM response in the most recent TaskRecord of the given type. It uses actual token usage from the API response when available,

String names the stop for diagnostics — telemetry attributes, log lines and test failure messages. Without it a MainLoopStop formats as a bare integer

Build an HTTPError suitable for mocking urlopen.

NewCommentWorkerPool creates a pool with the given concurrency limit. workerCount <= 0 defaults to 8.

--- Factory --- NewLLMClient creates the appropriate client based on the resolved endpoint protocol. protocol dispatch (canonical names from protocol.

ParseCommentsWithPath is like ParseComments but uses defaultPath as a fallback when individual comment objects omit the path field. repair is non-nil

RegisterSelected records an item in the selected set (the coverage denominator). It must be called once per item during the pre-dispatch pass, after f

ResolveEndpointWithOptions resolves an endpoint by first honoring an explicit provider selection. Otherwise it processes config, environment, then she

String renders a one-line human-readable estimate, matching scan.Estimate's format so diff and scan warnings read consistently. Money is intentionally

captureStderr captures everything written to os.Stderr during fn. Mirrors captureStdout; used to assert structured usage emitted on the failure path.

groupDiffs calls the LLM with file metadata (no diff content) to produce semantic groups. Falls back to one-file-per-group on any error.

newModelTUI builds a model-only TUI for tests. It has no config path or existing config, so add/delete/persist operations are unavailable — use newMod

Manifest returns the run's coverage builder, or nil for legacy/scan sessions that did not opt in via SessionOptions.Operation. Callers must be nil-saf

NewToolCallMessage creates an assistant history message. Use this instead of NewTextMessage("assistant", ...) to preserve native replay state and reas

Output executes a git command and returns stdout only.

ValidateOptions verifies that this session can be resumed in the requested review mode at all. It deliberately does not compare the ref text the user

buildAnthropicParams converts the shared ChatRequest into Anthropic SDK parameters.

cloneProviderEntry deep-copies a ProviderEntry so callers (rollback paths, map cloning) can safely mutate the returned value without aliasing the orig

CompletionTokenLimit returns the output cap for LLM requests. Runtime prompt-limit overrides must not silently expand the model's output budget.

Event emits a structured event as a span with immediate end.

NewCommitProvider creates a Provider for commit mode: show changes introduced by a single commit.

NewWorkspaceProvider creates a Provider for workspace mode (current uncommitted changes).

NormalizeAuthHeader normalizes an auth header value to a canonical form. It returns an error for unrecognized values.

NormalizeProtocol canonicalizes protocol names. It is case-insensitive and trims whitespace. Empty string is returned as-is (the caller decides the de

Read returns the full content of a file path (relative to RepoDir), resolved according to the active review mode. - Workspace: reads directly from the

dispatchSubtasks groups items into batches per the configured strategy, then processes batches sequentially while running files within each batch conc

mustUnmarshal is a test helper that unmarshals SARIF output and fails the test on error, so malformed output produces a clear failure instead of a nil

resolveOutputWriter resolves the --output target into a writer plus a cleanup function. - "" or "-" → os.Stdout with a no-op cleanup (colors pres

Await blocks until all submitted work has completed and returns aggregated results from every Submit call so far. A panic in submitted work is recove

The deferred finalizeRequest is this client's boundary for the retry report; see the OpenAI counterpart for why it is deferred and why the results are