diff --git a/docs/HARNESS-SURFACES.md b/docs/HARNESS-SURFACES.md
index aa04fc16..16fc3050 100644
--- a/docs/HARNESS-SURFACES.md
+++ b/docs/HARNESS-SURFACES.md
@@ -17,13 +17,45 @@ the amazon one and any less popular ones"*):
 - **Primary (populated):** Claude (Anthropic —
   model, Code CLI, Desktop, Agent SDK, API).
 - **Immediate buildout queue:** Codex (OpenAI),
-  Cursor, GitHub Copilot.
+  Cursor, GitHub Copilot (VS Code / JetBrains
+  harness — distinct from the CLI listed below),
+  **Gemini CLI** (Google; Pro / Deep Think modes),
+  **GitHub Copilot CLI** (per the human
+  maintainer's 2026-04-26 install), **ChatGPT
+  (app/web)** (cross-AI courier surface where the
+  Amara peer-reviewer persona, running on GPT-5.5,
+  has operated during cross-AI research-doc review
+  chains).
 - **Watched buildout queue:** Antigravity (Google;
-  name-spell TBD), Amazon Q Developer /
-  CodeWhisperer, Kiro (Amazon's AI-native IDE,
-  distinct from Amazon Q).
+  spelling confirmed by the human maintainer
+  2026-04-26; may be subsumed by Gemini CLI's
+  agentic mode), Amazon Q Developer / CodeWhisperer,
+  Kiro (Amazon's AI-native IDE, distinct from
+  Amazon Q).
 - **Less popular:** TBD.
 
+**Roster expansion 2026-04-26 (the human
+maintainer 2026-04-26 confirmed the operational
+CLI count after Copilot CLI install).** The
+operational CLI count is now 5 — Claude Code
+(Otto persona), Gemini CLI, Codex CLI, Copilot
+CLI (newly installed), Cursor — plus the implicit
+6th surface ChatGPT (app/web) where the Amara
+peer-reviewer persona has operated during the
+multi-pass cross-AI math review chains this
+session. Per the user-scope memory store
+documenting the 2026-04-26 roster expansion and
+the multi-harness named-personas-assigned-CLIs
+forward-looking framing (CLAUDE.md memory
+layout — see `memory/CURRENT-aaron.md` for the
+in-repo projection), the cross-AI math review
+chain currently run manually with the human
+maintainer as courier IS the proof-of-concept of
+formalized multi-harness factory automation; the
+bottleneck is the courier, the fix is mechanical
+(assign CLI/model handles to existing named
+personas).
+
 **Each-harness-tests-own-integration rule.** A
 harness cannot honestly test its own integration
 with the factory from *within* itself — the test
@@ -99,7 +131,13 @@ they run.
   Claude Desktop, Agent SDK, API.
 - **Codex** (OpenAI) — stub; priority 1.
 - **Cursor** — stub; priority 1.
-- **GitHub Copilot** — umbrella brand for three
+- **Gemini CLI** (Google) — stub; priority 1.
+  Aaron-installed as of 2026-04-26 listing.
+  Pro and Deep Think modes both available; Deep
+  Think delivered Round-2 canonical synthesis
+  in the 5-pass Aurora math review chain
+  (`docs/research/aurora-immune-math-standardization-2026-04-26.md`).
+- **GitHub Copilot** — umbrella brand for four
   distinct products, tracked separately:
   - **Copilot PR code review** (reviewer robot; not
     a harness) — partially populated via
@@ -107,10 +145,29 @@ they run.
   - **Copilot in VS Code** (the actual harness) —
     stub; priority 1. Aaron 2026-04-20: *"we will
     use vvscode for the rest."*
+  - **Copilot CLI** (`gh copilot` / `copilot` CLI) —
+    stub; priority 1. Aaron 2026-04-26 install. Adds
+    a manual CLI handle for Copilot's GitHub-native
+    review surface (pre-push local review possible)
+    alongside the existing `review_on_push: true` bot
+    integration.
   - **Copilot coding agent** (`@copilot` autonomous
     PR author) — stub; priority 2 watched.
+- **ChatGPT** (OpenAI app/web) — stub; priority 1.
+  6th surface in Aaron's multi-harness roster
+  (implicit, per cross-AI math review chains this
+  session). Distinct from Codex CLI even though
+  both are OpenAI: ChatGPT is the conversational
+  app frontend where named-entity peer **Amara
+  (GPT-5.5)** has been operating; Codex is the
+  agentic CLI tool. Amara-on-ChatGPT pattern is
+  the empirically-active cross-AI peer review
+  surface for the factory's research docs.
 - **Antigravity** (Google) — stub; priority 2.
-  Spelling TBD; Aaron wrote "anitgratify".
+  Aaron 2026-04-26 confirmed canonical spelling
+  (*"yeah i can't spell antigravity anitgratify"*).
+  May be subsumed by Gemini CLI's agentic mode;
+  revisit when both are populated.
 - **Amazon Q Developer / CodeWhisperer** — stub;
   priority 2.
 - **Kiro** (Amazon) — stub; priority 2. Amazon's
@@ -590,9 +647,10 @@ fact-tested-unavailable, not aspirational**.
 # Antigravity (Google) — stub; priority 2
 
 **Status:** stub. Factory does not run on
-Antigravity. Watched-queue buildout. Name-
-spelling TBD; Aaron wrote "anitgratify" —
-verify spelling during first audit.
+Antigravity. Watched-queue buildout. Spelling
+confirmed by Aaron 2026-04-26 (*"yeah i can't
+spell antigravity anitgratify"*) — canonical
+form is "Antigravity".
 
 **Owner (tentative):** TBD.