模块 1 · Symphony Baseline and the Comparison Frame
第 2 节,共 5 节 · 3 分钟 · 英语 · 只读
A matrix turns a pile of vendor features into a readable picture: capability categories run down the rows, candidate offerings run across the columns, and each cell records whether that offering has the capability. Shared capabilities show up as rows that fill in across every column; unique features show up as cells that differ from the rest of their row. This lesson builds that frame using Symphony's baseline categories, so later lessons can drop vendors into the columns without re-deciding the structure.
A per-vendor feature list answers "what does this offering do?" but hides the question you actually care about: "what do these offerings have in common, and where do they differ?" The matrix fixes that by forcing the same question across every offering. Rows are capability categories drawn from the Symphony baseline. Columns are candidate offerings. Each cell records whether the offering has that capability, does not have it, or has a differentiated version of it. Two patterns then become visible at a glance: a row that is filled across every column is a shared capability, and a cell that differs from the rest of its row is a unique feature. The structure is the teaching point — without shared rows, commonality is invisible no matter how many features you list.
来源支持
The rows come from what Symphony actually does, so the comparison starts from a concrete reference rather than an abstract checklist. Symphony is a long-running automation service that continuously reads work from a configured issue tracker, creates an isolated workspace for each issue, and runs a coding agent session inside that workspace. Its specification names four operational problems it solves: turning issue execution into a repeatable daemon workflow instead of manual scripts; isolating agent execution in per-issue workspaces; keeping workflow policy in-repo so teams version the agent prompt and runtime settings with their code; and providing enough observability to operate and debug multiple concurrent agent runs. It also maintains a single authoritative orchestrator state for dispatch, retries, and reconciliation, and can stop active runs when issue state changes make them ineligible. Those concerns — tracker-driven dispatch, per-issue isolation, versioned workflow policy, observability, and run lifecycle control — are the candidate rows. They are orchestration-level categories, not model-level ones, which is why they transfer to other offerings.
来源支持
来源支持
Look at the first two rows of the matrix. Both cells say yes, so those rows are shared capabilities: tracker-driven dispatch and per-issue isolated execution are not differentiators between these two offerings. Now look at the third row. Neither cell says yes, so voice-native interaction is absent from both — it is not a shared capability, it is a capability neither offering provides, which is itself useful information because it tells you where to look next. The pattern to internalize is positional: a full row means commonality, and a cell that differs from the rest of its row means a differentiator. When later lessons add offerings as columns, the same reading applies — scan rows for fullness and scan cells for outliers.
来源支持
You are adding a third column, Superset, to the matrix above. The supplied description says Superset accepts tasks from GitHub Issues and Linear, runs agents in isolated worktrees with terminals and diffs visible to the operator, and routes the unattended portion through scheduled automations with a TypeScript SDK, MCP server, and Slack bot. Fill the three existing rows for this new column and state which rows are shared and which cell is differentiated. Assume the row definitions stay exactly as written.
Row 1, tracker-driven dispatch: Superset accepts tasks from GitHub Issues and Linear, so the cell is yes — same as the other two columns. Row 2, per-issue isolated execution environment: Superset runs agents in isolated worktrees, so the cell is yes — same as the other two columns. Row 3, voice-native interaction: the supplied description says nothing about voice, so the cell is not supported in the supplied description — same as the other two columns. Reading the filled matrix: rows 1 and 2 are now full across all three columns, so tracker-driven dispatch and per-issue isolation are shared capabilities across the set. Row 3 is empty across all three columns, so voice-native interaction is a capability none of them provides. No cell in this three-row matrix is differentiated, which is a real result: on these three rows, Superset does not distinguish itself. Its differentiating features — visible terminals and diffs, scheduled automations, the SDK and Slack bot — would need their own rows to appear.
来源支持
来源支持
想一想
Take the three-row matrix from this lesson and add one row for a capability you believe only one of the two offerings has. Which cell would differ, and what evidence would you need before you could mark it as a genuine differentiator rather than a gap in your sources?