Baz vs Sonar (2026)

Stands at F and confidence normalised from a non-canonical 0.6 to 0.85. Breadth across classes is comfortably met with three source platforms, six ticketing systems, design, and two knowledge bases, all dated in the changelog. The detail worth carrying to comparison pages is the Works alongside positioning: Baz publishes dedicated solution pages for Claude Code, OpenAI Codex, Cursor and Devin, which is the super-harness framing, integrating with coding agents rather than competing with them, the same shape as sonar in this cohort.

Stands at F with a citable basis. Breadth is met across IDEs, CI systems and agent runtimes, and the eleven named agent clients are the distinguishing detail: this vendor integrates with the coding agents rather than competing with them, which is what makes it a verification layer rather than a rival. Recorded honestly: these are inbound connections where agents call Sonar, plus outbound reads of repositories and CI, rather than Sonar acting across a broad tool surface of its own.

Stands at F and confidence normalised from a non-canonical 0.6 to 0.85. Two things lift this beyond a standard multi-agent claim: the Planner loop runs BEFORE code exists, so orchestration covers plan generation, review and approval as well as implementation, and the Fixer agent closes its own loop by building an environment, validating and only committing when validation succeeds. Named infrastructure, the Agent Harness and Context Broker, is documented rather than implied, and Sessions trace the coordination.

DOWNGRADED from F. The July basis credited verification running from the agent inner loop to pre-PR to CI plus a remediation agent that detects, fixes, verifies and opens a pull request, which is real but is a fixed three-stage pipeline the vendor names Guide, Verify, Solve, not orchestration the customer composes. Nothing multi-agent, no branching, no customer-defined workflow, no agent coordination is documented. Compare qodo and cubic, both held at P for a fixed review-then-fix loop, and contrast zencoder and greptile at F for composable multi-agent pipelines. Held at P rather than N because the Remediation Agent genuinely chains detect, fix, re-verify and open a pull request without a human between steps.

Stands at F and confidence normalised from a non-canonical 0.55 to 0.85. All three trigger classes are documented and dated: events through pull request webhooks, schedules through Skills Maintainer scans every three days, weekly or fortnightly, and on-demand through comment commands, the CLI and a slash command invoked from inside another coding agent. That last one is the distinguishing detail, since it means a developer starts a Baz planning session without leaving Claude Code or Cursor.

Stands at F with a citable basis. All three trigger classes are documented: event-driven on commits and pull requests through CI and Agent Apps for GitHub, continuous inside the agent inner loop where Agentic Analysis fires on each generation, and developer-invoked through IDE, CLI and MCP tool calls. The inner-loop trigger is the genuinely unusual one, since verification fires on the agent's activity rather than on a human action, which few vendors in this lane document.

Stands at F and confidence normalised from a non-canonical 0.6 to 0.85. Grounding here is unusually broad because it spans three kinds at once: code structure through embeddings and graphs, organisational knowledge through Notion and Fibery, and product intent through six ticketing systems plus Figma. The Voyage AI code embeddings detail is a rare instance of a vendor naming its retrieval model. Cross-repository context with a visible cross repo tag on findings is the part most comparable to potpie-ai and greptile at F.

Stands at F with a citable basis. Context Augmentation is the Guide stage of the vendor's own framework and is genuinely distinct from how most of this lane grounds: rather than indexing a repository to answer questions, it pushes the organisation's standards, architecture and guidelines into the agent before it writes, so grounding is preventive rather than retrieval-time. Architecture search, call flows and dependency checks are named tools an agent calls through MCP.

Stands at F and confidence normalised from a non-canonical 0.6 to 0.85. Two independent named memory systems is more than any other vendor in this lane documents, and both are dated and mechanically described rather than asserted: Reviewer Memory accumulates from repeated feedback and rewrites the reviewer's own system prompt with version history and rollback, and Module Memory persists implementation detail per module. The prompt-versioning-and-rollback detail is the part that makes this unambiguously F, since it means memory is a managed, inspectable artefact rather than a black box.

Stands at P and confidence normalised from a non-canonical 0.6. The July basis reasoned correctly, that the baseline from the last full scan carries context between runs while persistent agent memory is not the framing, and re-retrieval supports it. What persists is analysis state, the project baseline, new-code definition, issue history and accepted or won't-fix dispositions, which is durable and genuinely affects later runs, but it is project state rather than agent memory. Kept separable from Know so one fact does not do double duty: Know rests on Context Augmentation pushing standards to agents, this on the retained baseline.

Stands at F and confidence normalised from a non-canonical 0.6 to 0.85. Plan review approvals shipped 1 July 2026 and change the picture from the July basis, which credited only suggest-without-blocking and chat: a plan now goes through a dedicated approval workflow with versioning, pinned comments and approve or request-changes before implementation starts, which is a shipped gate ahead of code generation rather than review after it. Eighth distinct oversight architecture in this lane and the only one that gates the PLAN rather than the diff or the tool call.

Stands at F with a citable basis. The quality gate is a genuine enforcement mechanism rather than advisory review: it blocks the merge, it is customer-configured, and the vendor positions the MCP server as standardising policy enforcement so gate conditions become CI checks applied consistently. That is control the agent cannot talk its way past, which is the strongest form on this axis. Seventh distinct oversight architecture in this lane and the only one that is deterministic, since a rule-based gate cannot be argued with by a more persuasive prompt.

Stands at F but the basis is materially rewritten, because the July basis graded partly from FOUNDER PEDIGREE, founded by cloud application security veterans, which is not evidence of anything the product or company does and should never have carried a grade. Re-retrieval found the real evidence the July build missed: SOC 2 certification stated in the site footer with a published trust centre at trust.baz.co and a security, privacy and compliance documentation page, plus role-based access control shipped September 2025. The conjunction is now met properly on attestation plus named controls. Confidence raised from a non-canonical 0.55 to 0.85 accordingly.

Stands at F with a citable basis. The conjunction is met on the product side rather than the corporate side, which is worth being explicit about: Sonar's security posture here is what the product does for the customer's code, SAST, secrets, SCA and compliance frameworks, not an attestation about Sonar's own operations. No SOC 2 or ISO certification for Sonar itself was retrieved on either pass; the July basis attributed SOC 2 Type II and ISO 27001 to Gitar, the acquired company, which is a different legal entity's attestation and is recorded as such rather than transferred to the parent. Flagged for a trust-page fetch at lane close.

Stands at F and confidence normalised from a non-canonical 0.55 to 0.85. Sessions are a retained, structured execution record covering invocation, tool calls, stages, outcome and cost, which is the F bar on this axis as applied to ellipsis and poolside, and they extend across every agent rather than one. The product-level timeline framing rather than raw logs is worth recording: the vendor is deliberately building auditability as a reviewable artefact. The July basis graded this F on the marketing phrase observable and explainable, not a black box; the grade was right and the basis was not.

Stands at F but confidence lowered from high to medium, because the July basis asserted auditability at high confidence in marketing language, definitive, transparent, auditable, traceable, without naming a mechanism or a page. The grade holds on a real property rather than a claim: deterministic rule-based analysis means every finding maps to a named rule and a code location and reproduces exactly, which is auditability of a stronger kind than a retained log, and it is why compliance buyers can defend the output. Recorded honestly: what is NOT documented on any page reached is an agent execution trace, a session record or an audit log of who ran what and when, so this F rests on finding-level explainability rather than on agent observability. Would benefit from a governance-page fetch at lane close.

Upgraded from N and confidence normalised from a non-canonical 0.45. The July basis said no on-premise or self-hosted option is documented, which was graded from absence and is now contradicted by two dated first-party entries the July builder did not reach: Private Mode shipped 25 December 2025, and private application review running inside the customer's own Kubernetes cluster shipped 29 July 2026. Held at P rather than F under the early access convention, since private application review is explicitly a limited technical preview, and because Private Mode governs analysis and retention rather than giving the customer a full self-hosted deployment. Would move to F when private application review reaches general availability.

Stands at F with a citable basis. Both halves of the axis are met independently: a genuine self-managed deployment in SonarQube Server, which is the long-established enterprise product rather than a bolted-on option, and data control through Gitar retaining no code after processing plus bring-your-own-Anthropic-key. The MCP server self-hosting path via official container image, and its installation as a Server extension exposing tools at the customer's own /mcp endpoint, means the agentic surface deploys inside the customer boundary too rather than only the analysis engine.

Stands at F and confidence normalised from a non-canonical 0.6 to 0.85. Eleven named prebuilt agents and reviewers is the largest shipped agent suite in this lane, and the Awesome Reviewers corpus is an unusual second layer: a free public library of 5,132 review instructions across twelve domains that a customer fetches as a markdown bundle straight into a skill folder. That is vendor-supplied adoptable assets in exactly the form this axis rewards, and it is published openly rather than gated behind the product.

Stands at F with a citable basis. Vendor-supplied rule sets and quality profiles across forty-plus languages are the largest body of prebuilt, adoptable assets of any vendor in this lane, and they are exactly what the axis rewards: the customer adopts them ready-made rather than authoring them. The agent plugins and slash commands for named harnesses are a second, newer layer of the same thing. Recorded honestly: these are rules and profiles rather than agents, so this is a broad reading of the axis, but the assets are unambiguously prebuilt, vendor-maintained and installed rather than built.

Stands at N and confidence normalised from a non-canonical 0.5 to 0.65. The July basis said Bedrock foundation models with no customer choice, which re-retrieval supports but also refines: the changelog names GPT-5-Codex with Sonnet 4.5 reflections as the reviewer stack, so the vendor mixes providers, but that is the vendor's routing decision, not the customer's. Nothing documents model selection, bring-your-own-key or a provider setting anywhere. Recorded honestly: like sonar in this cohort, a low Model grade here reflects a deliberate product design where the vendor owns model choice as part of guaranteeing review quality, and the vendor does disclose which models it uses, which is better than the provider-undisclosed case the conventions treat as N.

Stands at P and confidence normalised from a non-canonical 0.6. The July basis reasoned correctly and is preserved: core analysis is deterministic and model-free, so most of the product has no model to choose, and the LLM-touching parts, AI CodeFix and Gitar, support bring your own Anthropic key. That is a real customer-facing model control, which clears the P floor, but it is a single named provider with no selection across providers and no routing, so F is not available. Worth noting this is the one vendor in the lane where a low Model grade is a design virtue rather than a limitation, since determinism is the selling point.

DOWNGRADED from F on Mike's ruling of 2026-08-30, overturning my own August grade. Same error as zencoder: I held F while the basis itself recorded that no public REST API or SDK was retrieved, and flagged it as the weakest F on this axis in the lane rather than acting on that. Under the axis bar, full coverage means the platform is composable from outside through a documented API or SDK, with MCP as one form of evidence rather than the bar. The Baz surface is genuinely strong and that is why this is P rather than N: an MCP server exposing review tools to any IDE agent, a CLI with an interactive review loop, a slash command that starts Planner from inside another coding agent, and marketplace distribution. The Awesome Reviewers raw endpoints are a real documented API, but they serve a public corpus of review instructions rather than the Baz platform, which is the same distinction that correctly kept appfactor at P: MCP Bridge exposes the customer's systems, not AppFactor. Grade would move to F on a documented platform API or SDK. baz.ai/docs was not fetched on either pass and remains the cheap check at lane close.

Stands at F, now with a citable basis and an evidenceUrl; the July row named no page and carried a null URL. This is among the strongest Ext cells in the lane and unusual in kind: the MCP server is the product's primary agentic surface rather than an add-on, exposing a documented tool bag other vendors' agents call, and the supported-client list includes eleven named agents. Worth pairing against zed, which authored a protocol, and greptile, which hosts one alongside a REST API.

Stands at F and confidence normalised from a non-canonical 0.55 to 0.85, and this is one of the very few genuine F grades on this axis in the lane. The Eval rule is that the axis measures what the CUSTOMER can test, not what the vendor tests internally, and Prompt Playground is exactly that: a shipped agent test lab where a customer edits a reviewer's prompt and runs it against real change requests before rollout. Evaluations closes the loop by classifying every comment as accepted, rejected or unaddressed and feeding that back into reviewer behaviour. Contrast the nine vendors held at P in this lane for agents that merely verify their own output.

Stands at F with a citable basis. This is the qodo, cubic and greptile precedent in its purest form, where the axis and the product coincide: verification is not a feature of Sonar, it is Sonar. The distinguishing property against every other reviewer in this lane is determinism, since rule-based static analysis returns the same finding for the same code rather than varying by prompt, which is what makes the output auditable. Recorded honestly: the 44 percent outage-reduction and 8 percent token-reduction figures are vendor-reported from a press release with no published methodology.

Upgraded from P and confidence raised from a non-canonical 0.45 to 0.85. The July basis said computer use is limited to validation, which undersold it: this is the most thoroughly documented browser-driven capability found anywhere in this lane, built up across four dated releases from November 2025 to July 2026, and it is the product's differentiator rather than an accessory. The axis test is met exactly, since a rendered user interface has no programmatic interface and the agent operates it as a human would, clicking through flows and reading state off the screen. Contrast the thirteen downgrades in the June cohort, every one of which credited a terminal or sandbox; here the sandbox is where the browser runs, and the browser is the point.

Stands at N and confidence raised from a non-canonical 0.6 to 0.85, because this is graded from what the vendor says the product is rather than from absence. The July basis reasoned correctly, and it is one of the few in either cohort that did. Sonar analyses source code and exposes tools to agents; there is no browser, no GUI automation and no software operated without a programmatic interface anywhere in the product. Searched the AI capabilities documentation, the MCP server documentation and the product pages.