rayyagari2-create · rayyagari2-create · May 28, 2026 · May 28, 2026
diff --git a/calibration/d1-d4-rubric.md b/calibration/d1-d4-rubric.md
@@ -7,7 +7,7 @@ weighted dimensions, each scored 0-25:
 |-----------------|----------------------------------------------------|-----|
 | **D1 Correctness**  | Did the agent produce correct output on first attempt? | 25  |
 | **D2 Observability** | Did the agent emit enough telemetry, logs, and intermediate state to verify what it did? | 25  |
-| **D3 Compliance** | Did the agent follow the rules of the role scope, approvals, manifest use? | 25  |
+| **D3 Policy** | Did the agent follow the rules of the role scope, approvals, manifest use? | 25  |
 | **D4 Recurrence** | Did the agent repeat a prior mistake from its own failure library? | 25  |
 | **Total**       |                                                    | 100 |
 
@@ -38,7 +38,7 @@ Acceptable:
 ```
 D1 Correctness:  25 7/7 acceptance criteria met on first QA attempt, no rework
 D2 Observability: 22 bulletin entries at all 4 phase transitions, one missing handoff log
-D3 Compliance:   25 pre-spawn protocol followed; manifest matches ACs
+D3 Policy:   25 pre-spawn protocol followed; manifest matches ACs
 D4 Recurrence:   25 no known pattern repeated; novel task class
 Total:           97 → STANDARD (confidence band: LOW, n=6)
 ```
@@ -152,7 +152,7 @@ being trustworthy.
 
 ---
 
-## D3 Compliance
+## D3 Policy
 
 **What it measures:** did the agent follow the rules of its role?
 

diff --git a/docs/competitive-landscape.md b/docs/competitive-landscape.md
@@ -0,0 +1,127 @@
+# Competitive Landscape
+
+A short note on where AWF sits relative to four systems it is most
+often confused with. The pattern across all four: each operates at a
+different layer of the agent stack and the relationship to AWF is
+more often complementary than competing.
+
+For the layering model that grounds this document, see
+[`docs/execution-substrates.md`](execution-substrates.md).
+
+---
+
+## Maggy
+
+| Field | Value |
+|---|---|
+| **Category** | AI engineering command centre and execution substrate. |
+| **Relationship to AWF** | Execution layer beneath the authority layer. AWF authorizes over Maggy the same way it authorizes over Claude Code, Codex or Cursor. |
+| **AWF distinction** | Maggy answers "how is this work driven". AWF answers "who is allowed to do this work". |
+
+A buyer can adopt Maggy for engineering execution workflow and AWF for
+cross-runtime authorization. The two compose: Maggy drives the work,
+AWF decides whether Maggy is allowed to drive it on behalf of a given
+trust subject for a given task class.
+
+Maggy is part of the substrate roster AWF designed adapters for. An
+adapter ships post-Sprint 3 on pilot demand. Sprint 3 covers Maggy at
+the design-document level only.
+
+---
+
+## Microsoft AGT
+
+| Field | Value |
+|---|---|
+| **Category** | Runtime security enforcement layer. |
+| **Relationship to AWF** | Different layer. AGT is permission-check infrastructure at the moment of action. AWF is authority-record infrastructure across sessions and substrates. |
+| **AWF distinction** | AGT decides whether the next tool call is permitted, now, deterministically. AWF decides whether the trust subject behind the agent has earned authority over this class of work, over time. |
+
+AGT and AWF are complementary, not substitutes. AGT enforces a static
+policy at the moment of execution. AWF enforces a trust trajectory
+across sessions: a trust subject can be permitted by AGT to run a
+query yet `BLOCKED` by AWF because that trust subject has not
+accumulated enough D1 to D4 evidence on `db_migration` to earn the
+required tier.
+
+A production deployment with both layers gets the strengths of each:
+deterministic permission checks at action time from AGT and earned
+authority across sessions from AWF. Neither layer's job is the other's.
+
+For more on how AGT fits the broader governance stack, see
+[`docs/architecture/three-layer-stack.md`](architecture/three-layer-stack.md).
+
+---
+
+## Superlog
+
+| Field | Value |
+|---|---|
+| **Category** | Application observability for AI applications. |
+| **Relationship to AWF** | Different domain. Superlog instruments the application; AWF governs authority over the agents that produced the application's behaviour. |
+| **AWF distinction** | Superlog answers "what did the application do at runtime". AWF answers "was the agent that built this allowed to". |
+
+Superlog occupies the same conceptual slot for AI applications that
+Datadog or Honeycomb occupy for traditional services. It is consumed
+*after* an agent has shipped code. AWF is consumed *before* the agent
+runs, with the feedback loop closing only when telemetry from the
+running application feeds back into the agent's trust signal via
+Process Intelligence (Sprint 4).
+
+The two systems can compose. Superlog's runtime signal becomes input
+to AWF's trust-update logic. The signals flow in one direction, from
+runtime back to authority. There is no overlap at the authority layer
+itself.
+
+---
+
+## Pentagon
+
+| Field | Value |
+|---|---|
+| **Category** | Agent team workspace and execution layer. |
+| **Relationship to AWF** | Execution layer, similar to Maggy. AWF authorizes over Pentagon, not against it. |
+| **AWF distinction** | Pentagon answers "where does an agent team coordinate, share state and hand off work". AWF answers "is this team's chosen runtime authorized for this class of work". |
+
+Pentagon is a workspace product. Multiple agents and their human
+operators live inside it and pass work back and forth. The runtime
+authorization question still applies: when a Pentagon-resident agent
+attempts a task class, *that agent's trust subject* is what AWF scores
+authority against. Pentagon is the surface. AWF is the authority
+record behind the surface.
+
+A buyer can adopt Pentagon for cross-agent collaboration and AWF for
+the authority layer that decides what those agents are allowed to do.
+Same complementary pattern as the others.
+
+---
+
+## The general shape
+
+The four systems above are confused with AWF for the same reason. Each
+operates in or near the agent stack. Each speaks the language of
+governance or accountability. Each ships an artifact that *looks* like
+an authority decision from a distance.
+
+Up close they are not. The test is the question each system answers.
+
+| System | Question it answers |
+|---|---|
+| Maggy, Pentagon, Claude Code, Codex, Cursor | Can this work be done? And if so, how? |
+| Microsoft AGT | Is this specific tool call permitted, right now? |
+| Superlog | What did the application do at runtime? |
+| AWF | Has the agent's trust subject earned authority over this class of work, on this substrate? |
+
+Different question, different system. AWF composes with all of them.
+It replaces none.
+
+---
+
+## Related
+
+- `docs/execution-substrates.md`: the layering model that places each
+  of the systems above on its appropriate layer.
+- `docs/architecture/three-layer-stack.md`: runtime governance vs
+  scheduled automation vs behavioural accountability. The framing AGT
+  fits inside.
+- AWF Sprint Plan v4.4.2, Competitive Landscape: Maggy.