From 4026b261f75141d0d677e662caa8ceac4374ec85 Mon Sep 17 00:00:00 2001
From: Tejas Polakam <tejas.polakam@scale.com>
Date: Fri, 2 Oct 2026 13:04:43 -0400
Subject: [PATCH] chore: scrub internal identifiers from tests, fixtures and
 comments

Replace internal eval names, run and instance ids, task layouts, team and
image names in tests and fixtures with neutral placeholders, and drop
comments that only carried internal context: references to internal repos
and pipelines, internal PR and review numbers, and production measurements.
No behaviour change.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 .../artifacts/environment_universe.py         |  9 +++--
 src/agent_env/env/env.py                      |  2 +-
 .../env/envs/service_db/Dockerfile.db-web     |  5 ++-
 src/agent_env/env/gateway/constants.py        |  9 +++--
 .../src/components/rubric-grading-results.tsx |  2 +-
 .../ui/src/components/rubric-matrix.tsx       |  2 +-
 .../ui/src/lib/bar-downloads.smoke.ts         |  8 ++---
 .../lib/parse-trajectory-claude-cli.smoke.ts  |  6 ++--
 .../lib/parse-trajectory-opencode.smoke.ts    |  3 +-
 .../parse-trajectory-openinference.smoke.ts   |  2 +-
 .../ui/src/lib/step-attempt-failures.smoke.ts |  7 ++--
 .../sandbox_providers/modal_sandbox.py        |  2 +-
 .../snapshot_utils/agent_state_capture.py     |  4 +--
 .../verify_a2a_modalities.py                  |  9 +++--
 .../task_steps/apply_server_config.py         |  5 ++-
 .../task_step/task_steps/collect_artifacts.py |  6 ++--
 .../verify_spec_conformance.py                |  4 ---
 .../task_step/task_steps/prompt_agent.py      |  2 +-
 .../test_task_instance_retry_resume.py        |  6 ++--
 tst/unit/env/envs/test_multi_env_vm_sizing.py |  4 +--
 tst/unit/env/gateway/triggers_time_test.py    |  4 +--
 tst/unit/env/test_env.py                      |  2 +-
 .../env_gateway_provider_test.py              |  2 +-
 .../sandbox_providers/modal_sandbox_test.py   | 10 +++---
 .../sandbox_providers/vm_sandbox_test.py      |  2 +-
 .../golden/env/gateway_server.json            |  2 +-
 .../serialization/golden/env/mcp_server.json  |  2 +-
 tst/unit/serialization/golden/env/multi.json  |  2 +-
 .../serialization/golden/env/service_db.json  |  2 +-
 .../serialization/golden/env/website.json     |  2 +-
 .../serialization/test_golden_wire_format.py  | 10 +++---
 tst/unit/store/test_config_image_store.py     |  8 ++---
 tst/unit/store/test_secret_logging.py         |  2 +-
 tst/unit/task_step/test_collect_artifacts.py  | 34 +++++++++----------
 .../test_load_artifact_env_binding.py         |  4 +--
 .../task_step/test_manifest_cli_builder.py    |  3 +-
 .../test_parallel_deploy_env_binding.py       | 10 +++---
 .../task_step/test_prompt_agent_snapshots.py  |  2 +-
 .../test_snapshot_env_env_binding.py          |  6 ++--
 .../task_step/test_universe_comparison.py     |  7 ++--
 40 files changed, 98 insertions(+), 115 deletions(-)

diff --git a/src/agent_env/artifact/artifacts/environment_universe.py b/src/agent_env/artifact/artifacts/environment_universe.py
index 1d620e1e..813eac6e 100644
--- a/src/agent_env/artifact/artifacts/environment_universe.py
+++ b/src/agent_env/artifact/artifacts/environment_universe.py
@@ -59,11 +59,10 @@ def _serialize(self, handler: Any) -> dict[str, Any]:
             data.pop("service_artifact_ids", None)
             data.pop("metadata", None)
             # Dual-write: the pinned refs get a twin, and only them.
-            # The scope line also asked for a `service_artifact_ids` twin on these
-            # modern docs — declined. That key means "unpinned, resolve at latest"
-            # to every reader, and v0.9.1 (pinned permanently by
-            # synthetic-artifacts-pipeline) acts on it, so writing it onto a doc
-            # that IS pinned tells that worker to ignore the pins. Nothing legacy
+            # No `service_artifact_ids` twin on these modern docs: that key means
+            # "unpinned, resolve at latest" to every reader, older releases
+            # included, so writing it onto a doc that IS pinned tells those
+            # readers to ignore the pins. Nothing legacy
             # is dropped here: unpinned docs never reach this branch and still
             # round-trip `service_artifact_ids` / `metadata` above, and
             # get_environment_artifacts still falls back to them.
diff --git a/src/agent_env/env/env.py b/src/agent_env/env/env.py
index 009f2106..f083b9a8 100644
--- a/src/agent_env/env/env.py
+++ b/src/agent_env/env/env.py
@@ -81,7 +81,7 @@ class DeployedEnv:
     expires_at_utc: Optional[str] = None
 
     def __post_init__(self) -> None:
-        # Only records from before #1137 (2026-09-24) lack their card: they keep the stored values.
+        # Only older records lack their card: they keep the stored values.
         if self.environment_card and self.environment_card_url:
             self.mcp_url = _mcp_url(self.environment_url, self.environment_card)
             self.mcp_server_name = self.environment_card.get("name")
diff --git a/src/agent_env/env/envs/service_db/Dockerfile.db-web b/src/agent_env/env/envs/service_db/Dockerfile.db-web
index c55bb637..3e4540ed 100644
--- a/src/agent_env/env/envs/service_db/Dockerfile.db-web
+++ b/src/agent_env/env/envs/service_db/Dockerfile.db-web
@@ -1,8 +1,7 @@
 # Official pgweb image, pulled from its public GitHub Container Registry mirror.
 # We avoid Docker Hub because anonymous pulls are rate-limited per IP (429s from
-# CodeBuild's shared NAT). ghcr.io hosts the same image (identical digest) with
-# no anonymous rate limit and no auth required, so no private ECR mirror is
-# needed. The digest matches docker.io/sosedoff/pgweb:0.16.2. To bump: pick a
+# CI runners behind a shared NAT). ghcr.io hosts the same image (identical
+# digest) with no anonymous rate limit and no auth required. The digest matches docker.io/sosedoff/pgweb:0.16.2. To bump: pick a
 # new tag from https://github.com/sosedoff/pgweb/pkgs/container/pgweb and update
 # the tag + digest here.
 FROM ghcr.io/sosedoff/pgweb:0.16.2@sha256:8f1ed22e10c9da0912169b98b62ddc54930dc39a5ae07b0f1354d2a93d44c6ed
diff --git a/src/agent_env/env/gateway/constants.py b/src/agent_env/env/gateway/constants.py
index 270fa154..5a2256d0 100644
--- a/src/agent_env/env/gateway/constants.py
+++ b/src/agent_env/env/gateway/constants.py
@@ -14,9 +14,8 @@
 
 # Max end-to-end time for a single service's data load (seconds).
 #
-# The previous value of 600 was sized above a then-slowest prod load of ~430s.
-# That is no longer the ceiling: a large universe's gmail component now takes 498s and slack
-# 499s, leaving under 20% headroom on a number chosen to have plenty.
+# A single component of a large universe can take close to 500s, so this leaves
+# wide headroom.
 #
 # The margin matters more than the average because of how a load fails. A
 # component that exceeds this does not retry — the environment comes up, reports
@@ -33,8 +32,8 @@
 # Extra seconds granted per GB of payload, on top of the floor.
 #
 # Explicitly a heuristic, not a model. Measured load cost tracks RECORD COUNT, not bytes:
-# github moved 3.5GB in 89s while gmail took 206s for 409MB and slack 207s for 86.6MB (~20x
-# and ~94x worse per byte). So size is a weak proxy -- it over-grants to the big-blob
+# a blob-heavy service can move GBs in about a minute while a record-heavy one takes minutes
+# for well under 1GB. So size is a weak proxy -- it over-grants to the big-blob
 # services that never needed it and under-grants to the small record-heavy ones that do.
 # It is still strictly better than a flat cap, because the flat cap's failure mode was
 # killing healthy work. A real fix is a stall timeout (fail on no progress rather than on
diff --git a/src/agent_env/explorer/ui/src/components/rubric-grading-results.tsx b/src/agent_env/explorer/ui/src/components/rubric-grading-results.tsx
index 464a8e71..1bc85f3d 100644
--- a/src/agent_env/explorer/ui/src/components/rubric-grading-results.tsx
+++ b/src/agent_env/explorer/ui/src/components/rubric-grading-results.tsx
@@ -35,7 +35,7 @@ export interface VerificationResult {
 }
 
 export interface VerifierOutput {
-  /** Stamped by agent-env rubrics_verifier (PR1 judge output format registry). */
+  /** Stamped by agent-env rubrics_verifier (judge output format registry). */
   format?: JudgeOutputFormat;
   results: VerificationResult[];
   score: number;
diff --git a/src/agent_env/explorer/ui/src/components/rubric-matrix.tsx b/src/agent_env/explorer/ui/src/components/rubric-matrix.tsx
index 34be137f..271f0d68 100644
--- a/src/agent_env/explorer/ui/src/components/rubric-matrix.tsx
+++ b/src/agent_env/explorer/ui/src/components/rubric-matrix.tsx
@@ -30,7 +30,7 @@ function rubricLabel(rubric: RubricCriterion): string {
  * Pick the verifier output that actually grades the configured rubrics,
  * not just whichever happens to come first in dict iteration order.
  *
- * Tasks like `mm-avatar-full-urls` write multiple entries to
+ * A task can write multiple entries to
  * `metadata.verifications` — `verify-fs` (sandbox), `verify-response`
  * (rubric), `aggregate`. Prefer entries stamped with
  * `format: "rubric_binary"`; fall back to max rubric-id overlap
diff --git a/src/agent_env/explorer/ui/src/lib/bar-downloads.smoke.ts b/src/agent_env/explorer/ui/src/lib/bar-downloads.smoke.ts
index 4645fc0a..5aa1cf9e 100644
--- a/src/agent_env/explorer/ui/src/lib/bar-downloads.smoke.ts
+++ b/src/agent_env/explorer/ui/src/lib/bar-downloads.smoke.ts
@@ -164,17 +164,17 @@ function main(): void {
     );
   }
 
-  // The reported case: demo-bench-proof-c2, many steps + results.
+  // A task with many steps + results.
   {
     const many = Array.from({ length: 12 }, (_, i) => ({
-      agent_trajectory_s3_uri: `obj://halt/${i}`,
+      agent_trajectory_s3_uri: `obj://bucket/${i}`,
       step_id: `step-${i}`,
     }));
-    const downloads = buildBarDownloads(many, 'obj://halt/results.json');
+    const downloads = buildBarDownloads(many, 'obj://bucket/results.json');
     assert(downloads.length === 13, 'all 12 trajectories plus results kept');
     assert(
       shouldCollapseBarDownloads(downloads.length),
-      'the reported task collapses into a dropdown',
+      'a many-step task collapses into a dropdown',
     );
     assert(
       new Set(downloads.map(d => d.key)).size === downloads.length,
diff --git a/src/agent_env/explorer/ui/src/lib/parse-trajectory-claude-cli.smoke.ts b/src/agent_env/explorer/ui/src/lib/parse-trajectory-claude-cli.smoke.ts
index 5202ddbf..012e2da5 100644
--- a/src/agent_env/explorer/ui/src/lib/parse-trajectory-claude-cli.smoke.ts
+++ b/src/agent_env/explorer/ui/src/lib/parse-trajectory-claude-cli.smoke.ts
@@ -4,10 +4,8 @@
  * and assert that classification + event extraction + sub-agent folding
  * produce sensible output.
  *
- * Fixture default: `/tmp/traj.json` — pulled from the real run
- * a captured agent run while diagnosing the
- * "Cannot read properties of undefined (reading 'openinference.span.kind')"
- * crash. Override with `CLAUDE_CLI_FIXTURE=<path>`.
+ * Fixture default: `/tmp/traj.json`, a captured agent run. Override with
+ * `CLAUDE_CLI_FIXTURE=<path>`.
  *
  * Runner: plain TS, throws on assertion failure. Run with any TS executor:
  *   npx tsx src/lib/parse-trajectory-claude-cli.smoke.ts
diff --git a/src/agent_env/explorer/ui/src/lib/parse-trajectory-opencode.smoke.ts b/src/agent_env/explorer/ui/src/lib/parse-trajectory-opencode.smoke.ts
index 86989fbe..dd332d1c 100644
--- a/src/agent_env/explorer/ui/src/lib/parse-trajectory-opencode.smoke.ts
+++ b/src/agent_env/explorer/ui/src/lib/parse-trajectory-opencode.smoke.ts
@@ -4,8 +4,7 @@
  * `looksLikeOpenCodeStreamJson`) and assert the event extraction contract.
  *
  * Default fixture is inline + synthetic but faithful to the real record shape
- * (verified against the live run `task-v2env-186-778c8c2c-opencode-y5hm9olj`:
- * records are `{type, timestamp, sessionID, part}`; tool calls are
+ * (records are `{type, timestamp, sessionID, part}`; tool calls are
  * self-contained with `part.state={status,input,output,time}`; a turn ends on a
  * `step_finish` whose `part.reason === 'stop'`). Point at a real pulled
  * trajectory with `OPENCODE_FIXTURE=<path>`.
diff --git a/src/agent_env/explorer/ui/src/lib/parse-trajectory-openinference.smoke.ts b/src/agent_env/explorer/ui/src/lib/parse-trajectory-openinference.smoke.ts
index f444ce3c..00313cb3 100644
--- a/src/agent_env/explorer/ui/src/lib/parse-trajectory-openinference.smoke.ts
+++ b/src/agent_env/explorer/ui/src/lib/parse-trajectory-openinference.smoke.ts
@@ -5,7 +5,7 @@
  * output.
  *
  * The fixture is a 22-span trajectory captured from a 3-sub-agent kimi-k2p5
- * run via `/tmp/smoke_otel_bridge.py`. It exercises the OpenInference
+ * run. It exercises the OpenInference
  * adapter on every span kind (AGENT/CHAIN/TOOL/LLM), nested sub-agent
  * runs, and tool calls.
  *
diff --git a/src/agent_env/explorer/ui/src/lib/step-attempt-failures.smoke.ts b/src/agent_env/explorer/ui/src/lib/step-attempt-failures.smoke.ts
index 802ae939..5e0d3e9e 100644
--- a/src/agent_env/explorer/ui/src/lib/step-attempt-failures.smoke.ts
+++ b/src/agent_env/explorer/ui/src/lib/step-attempt-failures.smoke.ts
@@ -25,7 +25,7 @@ function assert(cond: unknown, msg: string): void {
 }
 
 function main(): void {
-  // The real prod entry from aeh-0000000000000-000000000000000a (completed run, one retry).
+  // A completed run's entry, one retry.
   const real = {
     attempt: 1,
     step_id: 'verify-response',
@@ -80,9 +80,8 @@ function main(): void {
   }
 
   {
-    // Timestamps are minute-precision, so a step failing in seconds produces ties.
-    // Shape from aeh-0000000000000-000000000000000b, whose verify-response attempts took
-    // 83s / 11s / 22s.
+    // Timestamps are minute-precision, so a step failing in seconds produces ties:
+    // here three verify-response attempts that took 83s / 11s / 22s.
     const parsed = parseStepAttemptFailures([
       { ...real, attempt: 3, occurred_at_utc: '2026-07-29 00:49 UTC' },
       { ...real, attempt: 2, occurred_at_utc: '2026-07-29 00:49 UTC' },
diff --git a/src/agent_env/providers/sandbox_providers/modal_sandbox.py b/src/agent_env/providers/sandbox_providers/modal_sandbox.py
index 86808e64..fe817c76 100644
--- a/src/agent_env/providers/sandbox_providers/modal_sandbox.py
+++ b/src/agent_env/providers/sandbox_providers/modal_sandbox.py
@@ -467,7 +467,7 @@ async def _log_sandbox_started(
 ) -> None:
     """Log one structured line mapping Modal's container id to the sandbox and its tags.
 
-    Modal's Datadog metrics carry only ``container_id`` (the ``ta-`` task id), not sandbox tags, so
+    Modal's exported metrics carry only ``container_id`` (the ``ta-`` task id), not sandbox tags, so
     this line is the join key from a container's usage to its run and pipeline step. Never raises:
     the task id comes from a private Modal method.
     """
diff --git a/src/agent_env/task_step/snapshot_utils/agent_state_capture.py b/src/agent_env/task_step/snapshot_utils/agent_state_capture.py
index 594eecf4..3e969d42 100644
--- a/src/agent_env/task_step/snapshot_utils/agent_state_capture.py
+++ b/src/agent_env/task_step/snapshot_utils/agent_state_capture.py
@@ -220,8 +220,8 @@ def upload_trajectory(
     Always uploaded from the worker side, never by the agent: sandbox VMs hold
     static STS env vars with a fixed expiry, and long runs exceed it.
 
-    ``name`` makes the object addressable by an id the caller already records
-    (#731); omit it to get a random one, which is what writing repeatedly under a
+    ``name`` makes the object addressable by an id the caller already records;
+    omit it to get a random one, which is what writing repeatedly under a
     single prefix needs, since a fixed name would overwrite the last upload.
     """
 
diff --git a/src/agent_env/task_step/task_steps/a2a_agent_validator/verify_a2a_modalities.py b/src/agent_env/task_step/task_steps/a2a_agent_validator/verify_a2a_modalities.py
index 98491d36..22160a21 100644
--- a/src/agent_env/task_step/task_steps/a2a_agent_validator/verify_a2a_modalities.py
+++ b/src/agent_env/task_step/task_steps/a2a_agent_validator/verify_a2a_modalities.py
@@ -15,11 +15,10 @@
 - `probe_step_did_not_run`: no matching response — the upstream
   `PromptAgentTaskStep` crashed before `context.prompt_responses.append(...)`.
 
-The gemini wrapper currently passes all 12 modality probes (text, image
-{png,jpeg,gif}, audio {wav,mpeg,mp4,ogg}, PDF, video/mp4, plus image/png
-delivered via `FileWithUri` over `s3://` and presigned `https://`). New
-A2A wrappers (claude-code, codex, grok) inherit this same probe set and
-will report which subset they actually ingest.
+There are 12 modality probes (text, image {png,jpeg,gif}, audio
+{wav,mpeg,mp4,ogg}, PDF, video/mp4, plus image/png delivered via
+`FileWithUri` over `s3://` and presigned `https://`). Every A2A agent gets
+the same probe set, and the results report which subset it ingests.
 """
 
 from __future__ import annotations
diff --git a/src/agent_env/task_step/task_steps/apply_server_config.py b/src/agent_env/task_step/task_steps/apply_server_config.py
index a760e3af..f827d0dc 100644
--- a/src/agent_env/task_step/task_steps/apply_server_config.py
+++ b/src/agent_env/task_step/task_steps/apply_server_config.py
@@ -57,9 +57,8 @@ def __init__(self, service: str, uri: str, args: dict[str, Any]):
 
     def to_dict(self) -> dict:
         # Dual-write. This feeds the `tasks` and `task_steps`
-        # documents — NOT the delivered bundle. The bundle mints its own dict
-        # from these attributes in export/exporter.py, so the permanently frozen
-        # `{'service': ...}` Harbor applier input is unaffected by the twin here.
+        # documents only. An exporter builds its own dict from these attributes,
+        # so a frozen `{'service': ...}` export format is unaffected by the twin here.
         return {"service": self.service, "environment": self.service, "uri": self.uri, "args": self.args}
 
     @classmethod
diff --git a/src/agent_env/task_step/task_steps/collect_artifacts.py b/src/agent_env/task_step/task_steps/collect_artifacts.py
index a42d1f80..5a602113 100644
--- a/src/agent_env/task_step/task_steps/collect_artifacts.py
+++ b/src/agent_env/task_step/task_steps/collect_artifacts.py
@@ -39,8 +39,8 @@
     }
 
 With seed CSV:
-    name,platform,expected_artifacts
-    VPC Endpoints,Terraform,"Dockerfile,run.sh,codebase.patch,test.patch,gold.patch,prompt.md,before.json,after.json,p2p_tests.json,f2p_tests.json,validation.json,progress.md"
+    name,expected_artifacts
+    example-task,"Dockerfile,run.sh,solution.patch,report.md"
 
 Or with static fallback (no seed data needed):
     {
@@ -613,7 +613,7 @@ async def _collect_via_controller(self, context, store, artifact_id, version):
         # Read files via the CUA MCP server (cua_get_file / cua_bash) over the
         # gateway's step/v1. These are MCP tools registered on the gateway on both
         # Ubuntu and macOS — NOT endpoints on the macOS controller sidecar, whose
-        # /step only accepts a ScaleCuaAction and 500s on a call_tool payload.
+        # /step only accepts a computer-use action and 500s on a call_tool payload.
         logger.info(f"Collecting artifacts via CUA MCP server (env_id={self.env_id}, gateway={deployed_env.environment_url})")
 
         items = self._resolve_items(context)
diff --git a/src/agent_env/task_step/task_steps/mcp_env_validator/verify_spec_conformance.py b/src/agent_env/task_step/task_steps/mcp_env_validator/verify_spec_conformance.py
index 4d20403e..30acf47d 100644
--- a/src/agent_env/task_step/task_steps/mcp_env_validator/verify_spec_conformance.py
+++ b/src/agent_env/task_step/task_steps/mcp_env_validator/verify_spec_conformance.py
@@ -14,10 +14,6 @@
 enforcement by passing ``spec_conformance`` in ``ValidationGateAggregatorStep``'s
 ``required_gates`` (a step arg, not env metadata). No baked spec = ``skipped``. Execution
 errors (unreachable MCP) still raise.
-
-Keep the diff logic in sync with the universe-generation pipeline's
-``synthetic_mcp_server_generation/validation/spec_conformance.py`` (same diff at generation
-time); tool shape differs — MCP objects here, dicts there.
 """
 
 from __future__ import annotations
diff --git a/src/agent_env/task_step/task_steps/prompt_agent.py b/src/agent_env/task_step/task_steps/prompt_agent.py
index 154fcf4e..4c23de17 100644
--- a/src/agent_env/task_step/task_steps/prompt_agent.py
+++ b/src/agent_env/task_step/task_steps/prompt_agent.py
@@ -711,7 +711,7 @@ async def _fetch_trajectory(self, a2a_url: str, traj_ext: dict, a2a_server_task_
             )
         # Fetched by a2a_server_task_id, the id the A2A server produced; stored under
         # target_a2a_task_id, the client message id persisted on the conversation as
-        # a2a_task_id, so the FE can resolve it (#731).
+        # a2a_task_id, so the FE can resolve it.
         upload = None
         if mode == "objects":
             upload = await asyncio.to_thread(
diff --git a/tst/integration/env/task_step/test_task_instance_retry_resume.py b/tst/integration/env/task_step/test_task_instance_retry_resume.py
index a8355383..a2c64177 100644
--- a/tst/integration/env/task_step/test_task_instance_retry_resume.py
+++ b/tst/integration/env/task_step/test_task_instance_retry_resume.py
@@ -194,9 +194,9 @@ async def test_a_failed_steps_partial_writes_reach_the_doc_but_not_the_journal()
 
     A step that writes and then fails gets its writes onto the instance through
     ``record_task_failure``'s wholesale ``context`` write, which is not journaled. Replay
-    therefore omits them and the journal is *not* a complete record of a failed run. The
-    stacked retry PR is expected to journal the failing step with ``status=failure`` first;
-    when it does, this test should start failing and be tightened to assert equality.
+    therefore omits them and the journal is *not* a complete record of a failed run. Once
+    the failing step is journaled with ``status=failure`` first, this test should start
+    failing and be tightened to assert equality.
     """
     task_id = f"retry-resume-partial-test-{uuid.uuid4().hex[:8]}"
     instance_id = f"{task_id}-deterministic"
diff --git a/tst/unit/env/envs/test_multi_env_vm_sizing.py b/tst/unit/env/envs/test_multi_env_vm_sizing.py
index 3e9866bc..f8c3c8b2 100644
--- a/tst/unit/env/envs/test_multi_env_vm_sizing.py
+++ b/tst/unit/env/envs/test_multi_env_vm_sizing.py
@@ -1,11 +1,11 @@
 """A MultiEnv does not size its own box; cpu and memory both defer to the provider.
 
-The providers already floor both per backend (#1045: 0.125 vCPU on Modal, where cpu is a
+The providers already floor both per backend (0.125 vCPU on Modal, where cpu is a
 burstable reservation, 0.5 on the VM backends; 8GB memory on either), and a number invented
 here could only override a better-informed default. In container mode it was actively
 wrong: the derived size landed on the gateway, a router, as if it hosted every service.
 
-7 days of production logs showed nothing depending on the derived cpu except that gateway,
+Nothing depended on the derived cpu except that gateway,
 and the two callers that leaned on the derived memory were getting it from `cpu x 4096` --
 an 11-service env on 2GB. An explicit value from the caller still wins on both axes.
 """
diff --git a/tst/unit/env/gateway/triggers_time_test.py b/tst/unit/env/gateway/triggers_time_test.py
index dbee0300..b08ebbbb 100644
--- a/tst/unit/env/gateway/triggers_time_test.py
+++ b/tst/unit/env/gateway/triggers_time_test.py
@@ -256,7 +256,7 @@ async def test_event_anchored_never_fires_without_anchor():
 ], ids=["one-shot", "bounded-recurrence"])
 @pytest.mark.asyncio
 async def test_terminating_time_anchor_does_not_strand_dependent(anchor_when):
-    """Greptile #765: a TIME anchor that retires must still resolve its dependents, even though
+    """A TIME anchor that retires must still resolve its dependents, even though
     `_fire_time` recomputes `_has_time_triggers` (in `finally`) *before* calling `_resolve_dependents`,
     which early-returns on a cleared flag.
 
@@ -534,7 +534,7 @@ async def test_rearm_reanchors_trigger_that_was_firing():
 
 @pytest.mark.asyncio
 async def test_rearm_drops_capped_backlog_and_says_so(monkeypatch):
-    """Greptile P1c: a re-arm that lands while a capped catch-up batch is firing drops the leftover
+    """A re-arm that lands while a capped catch-up batch is firing drops the leftover
     backlog — by design (those marks are on a timeline the new clock never reached) — but the drop must
     NOT be silent: it emits a `reanchored` event naming the dropped mark, and the trigger re-anchors to
     the new t0 rather than replaying the old axis."""
diff --git a/tst/unit/env/test_env.py b/tst/unit/env/test_env.py
index a6a4a7f9..7c52efd9 100644
--- a/tst/unit/env/test_env.py
+++ b/tst/unit/env/test_env.py
@@ -186,7 +186,7 @@ def test_a_record_takes_its_mcp_url_and_server_name_from_its_card():
 
 @pytest.mark.parametrize("card, card_url", [(None, _CARD_URL), (_CARD, None)], ids=["no-card", "no-card-url"])
 def test_a_record_without_its_card_keeps_its_stored_mcp_url_and_name(card, card_url):
-    """Records written before #1137 carry no card, so their stored values stand."""
+    """Older records carry no card, so their stored values stand."""
     record = DeployedEnv.from_dict({"env_id": "e", "env_version": 1, "gateway_url": "https://old", "sandbox_id": "sb",
                                     "mcp_url": "https://old/mcp", "mcp_server_name": "old", "environment_card": card, "environment_card_url": card_url})
     assert (record.mcp_url, record.mcp_server_name) == ("https://old/mcp", "old")
diff --git a/tst/unit/providers/env_providers/env_gateway_provider_test.py b/tst/unit/providers/env_providers/env_gateway_provider_test.py
index 880abe09..92c27146 100644
--- a/tst/unit/providers/env_providers/env_gateway_provider_test.py
+++ b/tst/unit/providers/env_providers/env_gateway_provider_test.py
@@ -209,7 +209,7 @@ async def _spy_acquire(self, ctx):
 
     assert len(gp._container_sandboxes) == 4
 
-    # PR 2: the deploy routes DB state through the provider.
+    # The deploy routes DB state through the provider.
     # servicedb container env comes from the provider's store-spec
     db_call = next(c for c in create_calls if c["port"] == SERVICE_DB_PORT)
     assert db_call["env"] == LocalPostgresStateProvider().store_spec(["slack", "email"]).env
diff --git a/tst/unit/providers/sandbox_providers/modal_sandbox_test.py b/tst/unit/providers/sandbox_providers/modal_sandbox_test.py
index 76080487..a095df55 100644
--- a/tst/unit/providers/sandbox_providers/modal_sandbox_test.py
+++ b/tst/unit/providers/sandbox_providers/modal_sandbox_test.py
@@ -42,7 +42,7 @@ async def test_create_container_wraps_modal_create_failure_with_context(_patched
 
         with pytest.raises(RuntimeError) as ei:
             await _patched_provider.create_container(
-                image_name="123456789012.dkr.ecr.us-west-2.amazonaws.com/agent-env/agent-images/mcp-server-airtable:v4",
+                image_name="123456789012.dkr.ecr.us-west-2.amazonaws.com/example/mcp-server:v1",
                 port=18765,
                 env={},
                 cpu=2.0,
@@ -54,7 +54,7 @@ async def test_create_container_wraps_modal_create_failure_with_context(_patched
     msg = str(ei.value)
     assert "Modal sandbox create failed" in msg
     assert "_BareSandboxTimeoutError" in msg
-    assert "mcp-server-airtable:v4" in msg
+    assert "example/mcp-server:v1" in msg
     assert "port=18765" in msg
     assert "region=us-east-1" in msg
     assert "i6pn=True" in msg
@@ -75,7 +75,7 @@ async def test_create_container_wraps_post_create_failure_and_terminates(_patche
 
         with pytest.raises(RuntimeError) as ei:
             await _patched_provider.create_container(
-                image_name="registry.example/mcp-server-openclaw-strava:v6",
+                image_name="registry.example/mcp-server-example:v1",
                 port=18765,
                 env={},
             )
@@ -83,7 +83,7 @@ async def test_create_container_wraps_post_create_failure_and_terminates(_patche
     msg = str(ei.value)
     assert "Modal sandbox post-create failed" in msg
     assert "sb_id=sb-FAKEFAKEFAKEFAKEFAKE" in msg
-    assert "mcp-server-openclaw-strava:v6" in msg
+    assert "mcp-server-example:v1" in msg
     fake_sb.terminate.aio.assert_awaited_once()
 
 
@@ -141,7 +141,7 @@ def mint(self, host):
             return RegistryAuth(host, "u", "p")
 
     ecr_host = "123456789012.dkr.ecr.us-west-2.amazonaws.com"
-    ecr = "123456789012.dkr.ecr.us-west-2.amazonaws.com/agent-env/agent-images/x:v1"
+    ecr = "123456789012.dkr.ecr.us-west-2.amazonaws.com/example/x:v1"
     _patched_provider._ecr_pull_secret_name = "agent-env-ecr-reader"
     p = lambda n: patch(f"agent_env.providers.sandbox_providers.modal_sandbox.modal.{n}")  # noqa: E731
 
diff --git a/tst/unit/providers/sandbox_providers/vm_sandbox_test.py b/tst/unit/providers/sandbox_providers/vm_sandbox_test.py
index d340e7b1..775125b1 100644
--- a/tst/unit/providers/sandbox_providers/vm_sandbox_test.py
+++ b/tst/unit/providers/sandbox_providers/vm_sandbox_test.py
@@ -62,7 +62,7 @@ async def test_load_docker_images_downloads_to_file_before_load(signing_store):
 
     Retrying `curl ... | gunzip | docker load` directly corrupts the stream
     because curl can't rewind bytes already piped to stdout, so the retry flags
-    must only apply to a `-o file` download (Greptile P1 on PR #430).
+    must only apply to a `-o file` download.
     """
     sandbox = _RecordingVmSandbox(images_stdout="myimage\n")
     artifact = SimpleNamespace(tar_gz_object_url="s3://bucket/img.tar.gz", image_name="myimage:latest")
diff --git a/tst/unit/serialization/golden/env/gateway_server.json b/tst/unit/serialization/golden/env/gateway_server.json
index 8049f74b..eab5f047 100644
--- a/tst/unit/serialization/golden/env/gateway_server.json
+++ b/tst/unit/serialization/golden/env/gateway_server.json
@@ -6,7 +6,7 @@
   },
   "id": "gateway-env",
   "metadata": {
-    "owner": "env-pod"
+    "owner": "example-team"
   },
   "type": "gateway_server",
   "version": 11
diff --git a/tst/unit/serialization/golden/env/mcp_server.json b/tst/unit/serialization/golden/env/mcp_server.json
index b24dbd55..7c15783c 100644
--- a/tst/unit/serialization/golden/env/mcp_server.json
+++ b/tst/unit/serialization/golden/env/mcp_server.json
@@ -8,7 +8,7 @@
   "environment_name": "slack",
   "id": "slack-env",
   "metadata": {
-    "owner": "env-pod"
+    "owner": "example-team"
   },
   "service_name": "slack",
   "type": "mcp_server",
diff --git a/tst/unit/serialization/golden/env/multi.json b/tst/unit/serialization/golden/env/multi.json
index 930cb21d..08009cec 100644
--- a/tst/unit/serialization/golden/env/multi.json
+++ b/tst/unit/serialization/golden/env/multi.json
@@ -9,7 +9,7 @@
     }
   ],
   "metadata": {
-    "owner": "env-pod"
+    "owner": "example-team"
   },
   "type": "multi",
   "version": 6,
diff --git a/tst/unit/serialization/golden/env/service_db.json b/tst/unit/serialization/golden/env/service_db.json
index 34051b9d..65465b50 100644
--- a/tst/unit/serialization/golden/env/service_db.json
+++ b/tst/unit/serialization/golden/env/service_db.json
@@ -16,7 +16,7 @@
   },
   "id": "service-db-env",
   "metadata": {
-    "owner": "env-pod"
+    "owner": "example-team"
   },
   "type": "service_db",
   "version": 3
diff --git a/tst/unit/serialization/golden/env/website.json b/tst/unit/serialization/golden/env/website.json
index 23f83880..c58fccd1 100644
--- a/tst/unit/serialization/golden/env/website.json
+++ b/tst/unit/serialization/golden/env/website.json
@@ -13,7 +13,7 @@
   },
   "id": "shop-site-env",
   "metadata": {
-    "owner": "env-pod"
+    "owner": "example-team"
   },
   "service_name": "shop",
   "type": "website",
diff --git a/tst/unit/serialization/test_golden_wire_format.py b/tst/unit/serialization/test_golden_wire_format.py
index 689af5df..66f0e0ec 100644
--- a/tst/unit/serialization/test_golden_wire_format.py
+++ b/tst/unit/serialization/test_golden_wire_format.py
@@ -69,7 +69,7 @@ def _mcp_server_env() -> MCPServerEnv:
         version=4,
         docker_image_artifact=_docker_image("slack-mcp-image", 5, "slack-mcp:v5"),
         environment_name="slack",
-        metadata={"owner": "env-pod"},
+        metadata={"owner": "example-team"},
     )
 
 
@@ -80,7 +80,7 @@ def _website_env() -> WebsiteEnv:
         backend_docker_image_artifact=_docker_image("shop-backend-image", 3, "shop-backend:v3"),
         frontend_docker_image_artifact=_docker_image("shop-frontend-image", 3, "shop-frontend:v3"),
         environment_name="shop",
-        metadata={"owner": "env-pod"},
+        metadata={"owner": "example-team"},
     )
 
 
@@ -156,7 +156,7 @@ def _website_env() -> WebsiteEnv:
         id="gateway-env",
         version=11,
         docker_image_artifact=_docker_image("gateway-image", 12, "gateway:v12"),
-        metadata={"owner": "env-pod"},
+        metadata={"owner": "example-team"},
     ),
     "mcp_server": _mcp_server_env,
     "multi": lambda: MultiEnv(
@@ -164,7 +164,7 @@ def _website_env() -> WebsiteEnv:
         version=6,
         mcp_server_envs=[_mcp_server_env()],
         website_envs=[_website_env()],
-        metadata={"owner": "env-pod"},
+        metadata={"owner": "example-team"},
     ),
     "service_db": lambda: ServiceDBEnv(
         id="service-db-env",
@@ -172,7 +172,7 @@ def _website_env() -> WebsiteEnv:
         db_web_docker_image_artifact=_docker_image("pgweb-image", 1, "pgweb:v1"),
         db_mcp_docker_image_artifact=_docker_image("db-mcp-image", 1, "db-mcp:v1"),
         version=3,
-        metadata={"owner": "env-pod"},
+        metadata={"owner": "example-team"},
     ),
     "website": _website_env,
 }
diff --git a/tst/unit/store/test_config_image_store.py b/tst/unit/store/test_config_image_store.py
index 4bbc1b62..348937b6 100644
--- a/tst/unit/store/test_config_image_store.py
+++ b/tst/unit/store/test_config_image_store.py
@@ -93,10 +93,10 @@ def test_env_override_beats_config_toml(monkeypatch, tmp_path):
 def test_ecr_from_config_builds_store_lazily_without_network(monkeypatch):
     store = EcrImageStore.from_config(
         registry_host="123.dkr.ecr.us-west-2.amazonaws.com",
-        repository_prefix="agent-env/agent-images-dev",
+        repository_prefix="example/images-dev",
     )
     assert isinstance(store, EcrImageStore)
-    assert store.image_ref("foo", "v1") == "123.dkr.ecr.us-west-2.amazonaws.com/agent-env/agent-images-dev/foo:v1"
+    assert store.image_ref("foo", "v1") == "123.dkr.ecr.us-west-2.amazonaws.com/example/images-dev/foo:v1"
 
 
 def test_ecr_rejects_a_non_ecr_credential_provider():
@@ -115,11 +115,11 @@ def test_ecr_rejects_a_non_ecr_credential_provider():
 def test_ecr_store_exposes_its_registry_host():
     store = EcrImageStore.from_config(
         registry_host="123.dkr.ecr.us-west-2.amazonaws.com",
-        repository_prefix="agent-env/agent-images-dev",
+        repository_prefix="example/images-dev",
     )
     assert store.registry_host == "123.dkr.ecr.us-west-2.amazonaws.com"
     # Host only — no repository prefix, unlike image_ref().
-    assert "agent-images-dev" not in store.registry_host
+    assert "images-dev" not in store.registry_host
 
 
 def test_registry_host_is_the_host_component_of_image_ref():
diff --git a/tst/unit/store/test_secret_logging.py b/tst/unit/store/test_secret_logging.py
index 48a13f6c..dd578604 100644
--- a/tst/unit/store/test_secret_logging.py
+++ b/tst/unit/store/test_secret_logging.py
@@ -79,7 +79,7 @@ def test_level_restored_even_when_the_call_raises():
 
 
 def test_an_overlapping_fetch_stays_silenced_when_the_first_finishes():
-    """The race Greptile flagged, forced rather than raced for.
+    """The overlap race, forced rather than raced for.
 
     Two fetches overlap and the one that started FIRST finishes first. It
     captured DEBUG on the way in, so a naive restore puts DEBUG back while the
diff --git a/tst/unit/task_step/test_collect_artifacts.py b/tst/unit/task_step/test_collect_artifacts.py
index cc2e70fd..8992e41f 100644
--- a/tst/unit/task_step/test_collect_artifacts.py
+++ b/tst/unit/task_step/test_collect_artifacts.py
@@ -85,15 +85,15 @@ def test_defaulted_agent_name_with_env_id_tolerated(self):
 def _macos_cua_context():
     """A macOS CUA env: gateway_url (CUA MCP server, 18765) plus a distinct
     cua_controller_url (the osworld sidecar, 18768). cua_get_file must go to the
-    gateway, not the sidecar — the sidecar's /step only accepts a ScaleCuaAction
-    and 500s on a call_tool payload."""
+    gateway, not the sidecar — the sidecar's /step only accepts a computer-use
+    action and 500s on a call_tool payload."""
     return TaskStepContext(instance_id="inst-1", deployed_envs=[_macos_record()])
 
 
 class TestControllerPath:
     def test_reads_route_to_gateway_mcp_not_controller_sidecar(self):
         """Regression: reads must use gateway_url (CUA MCP server), even when a
-        macOS cua_controller_url is present. Routing them to the sidecar (#693)
+        macOS cua_controller_url is present. Routing them to the sidecar
         made every read 500."""
         step = CollectArtifactsTaskStep(
             id="collect", version=None, env_id="macos-cua",
@@ -503,31 +503,31 @@ class TestManifestResolution:
 
     def test_manifest_mode_keys_by_logical_name_and_filters(self):
         step = CollectArtifactsTaskStep(
-            id="collect", version=None, base_path="/app/swe_atlas",
+            id="collect", version=None, base_path="/app/workspace",
             manifest_step_id="m9", exclude_basenames=["base_image.tar", "base_image.tar.sha256"],
         )
         ctx = _manifest_ctx({
-            "dockerfile":     "/app/swe_atlas/Dockerfile",
-            "gold_patch":     "/app/swe_atlas/patches/gold_patch.diff",
-            "base_image_tar": "/app/swe_atlas/base_image.tar",        # excluded basename
-            "m7_iter_logs":   "/app/swe_atlas/logs/m7_iter_*.json",   # unresolved glob -> skipped
-            "coverage_state": "target_reached",                        # literal value, not a path -> skipped
-            "harbor_zip":     "/app/swe_atlas/ruff_harbor_v1.zip",
+            "dockerfile":     "/app/workspace/Dockerfile",
+            "gold_patch":     "/app/workspace/patches/gold_patch.diff",
+            "base_image_tar": "/app/workspace/base_image.tar",        # excluded basename
+            "run_logs":       "/app/workspace/logs/run_*.json",        # unresolved glob -> skipped
+            "status":         "complete",                              # literal value, not a path -> skipped
+            "bundle_zip":     "/app/workspace/bundle_a.zip",
         })
         items = {k: (src, obj) for k, src, obj in step._resolve_items(ctx)}
-        assert set(items) == {"dockerfile", "gold_patch", "harbor_zip"}
+        assert set(items) == {"dockerfile", "gold_patch", "bundle_zip"}
         # logical name -> (absolute source path, base_path-relative object name)
-        assert items["gold_patch"] == ("/app/swe_atlas/patches/gold_patch.diff", "patches/gold_patch.diff")
-        assert items["dockerfile"] == ("/app/swe_atlas/Dockerfile", "Dockerfile")
+        assert items["gold_patch"] == ("/app/workspace/patches/gold_patch.diff", "patches/gold_patch.diff")
+        assert items["dockerfile"] == ("/app/workspace/Dockerfile", "Dockerfile")
 
     def test_enumeration_drops_excluded_basenames(self):
         step = CollectArtifactsTaskStep(
-            id="collect", version=None, base_path="/app/swe_atlas",
+            id="collect", version=None, base_path="/app/workspace",
             exclude_basenames=["base_image.tar", "base_image.tar.sha256"],
         )
         enumerated = ["Dockerfile", "patches/gold_patch.diff", "base_image.tar",
-                      "base_image.tar.sha256", "245a_harbor_v1.zip"]
-        assert step._drop_excluded(enumerated) == ["Dockerfile", "patches/gold_patch.diff", "245a_harbor_v1.zip"]
+                      "base_image.tar.sha256", "bundle_b.zip"]
+        assert step._drop_excluded(enumerated) == ["Dockerfile", "patches/gold_patch.diff", "bundle_b.zip"]
 
     def test_manifest_uses_latest_response_when_step_reran(self):
         # On retry/resume a step can append more than once; the latest output wins.
@@ -723,7 +723,7 @@ def test_collects_through_the_named_container(self):
 
 class TestControllerWire:
     def test_controller_call_posts_call_tool_to_the_cards_step_endpoint(self, monkeypatch):
-        """The #693 intent at the wire: call_tool goes to the gateway's step/v1 (18765), never the macOS sidecar (18768)."""
+        """The same routing at the wire: call_tool goes to the gateway's step/v1 (18765), never the macOS sidecar (18768)."""
         sent = _mock_gateway(monkeypatch, {"content": [{"type": "text", "text": "aGk="}]})
         step = CollectArtifactsTaskStep(id="collect", version=None, env_id="macos-cua", artifact_paths=["/x"])
 
diff --git a/tst/unit/task_step/test_load_artifact_env_binding.py b/tst/unit/task_step/test_load_artifact_env_binding.py
index 40ae3900..a4f377c4 100644
--- a/tst/unit/task_step/test_load_artifact_env_binding.py
+++ b/tst/unit/task_step/test_load_artifact_env_binding.py
@@ -1,7 +1,7 @@
 """Which deployment of an env a `load_artifact` step loads into.
 
 Matching on `env_id` alone sends every branch's load to one deployment, leaving the
-others deployed but empty — their agents then bind correctly (#808) to an env with
+others deployed but empty — their agents then bind correctly to an env with
 no data.
 """
 
@@ -17,7 +17,7 @@
 from agent_env.task_step.task_steps.deploy_env import DeployEnvTaskStep
 from agent_env.task_step.task_steps.load_artifact import LoadArtifactTaskStep
 
-ENV_ID = "multi-env-0gr8479d"
+ENV_ID = "multi-env"
 
 
 def _fake_env(sandbox_id: str):
diff --git a/tst/unit/task_step/test_manifest_cli_builder.py b/tst/unit/task_step/test_manifest_cli_builder.py
index 46894bac..d760ed5a 100644
--- a/tst/unit/task_step/test_manifest_cli_builder.py
+++ b/tst/unit/task_step/test_manifest_cli_builder.py
@@ -451,8 +451,7 @@ def test_root_command_collisions_rejected(entities, match):
 
 
 def test_manifest_live_schema_drift_warns(fake_gateway, tmp_path):
-    # The universe-generation pipeline's manifest validation keeps these in sync, but the renderer parses whatever is
-    # served: surface drift instead of dropping it silently.
+    # The renderer parses whatever is served: surface drift instead of dropping it silently.
     drifted = _manifest(
         [
             {
diff --git a/tst/unit/task_step/test_parallel_deploy_env_binding.py b/tst/unit/task_step/test_parallel_deploy_env_binding.py
index 886732be..5041db9a 100644
--- a/tst/unit/task_step/test_parallel_deploy_env_binding.py
+++ b/tst/unit/task_step/test_parallel_deploy_env_binding.py
@@ -1,8 +1,6 @@
-"""The `haltbench-mvp-whichenv` DAG in-process: two `deploy_env` of one `env_id`
-running concurrently, then an agent per env.
-
-Instance haltbench-mvp-whichenv-7c9fb4o9: both envs deployed, both agents wired
-to the first.
+"""A two-branch DAG in-process: two `deploy_env` of one `env_id` running
+concurrently, then an agent per env. Each agent must bind to its own branch's
+deployment, not both to the first.
 """
 
 from __future__ import annotations
@@ -17,7 +15,7 @@
 from agent_env.task_step.task_steps.deploy_agent import DeployAgentTaskStep
 from agent_env.task_step.task_steps.deploy_env import DeployEnvTaskStep
 
-ENV_ID = "multi-env-0gr8479d"
+ENV_ID = "multi-env"
 
 
 def _fake_env_deploying_to(mcp_url: str, sandbox_id: str):
diff --git a/tst/unit/task_step/test_prompt_agent_snapshots.py b/tst/unit/task_step/test_prompt_agent_snapshots.py
index 4ac3c318..ff7559cc 100644
--- a/tst/unit/task_step/test_prompt_agent_snapshots.py
+++ b/tst/unit/task_step/test_prompt_agent_snapshots.py
@@ -700,7 +700,7 @@ async def test_the_final_capture_publishes_the_universe_for_load_artifact(monkey
     await _series(env_id="env-1").finish(ctx)
 
     # Keyed on the step id, so a judge step can be wired to this prompt step
-    # without knowing it snapshots. `load_artifact` resolves it once #746 lands.
+    # without knowing it snapshots. `load_artifact` resolves it.
     assert ctx.metadata["env_snapshotted_universes"]["solve"] == {
         "id": f"snapshot-env-1-{DISCRIMINATOR}",
         "version": 4,
diff --git a/tst/unit/task_step/test_snapshot_env_env_binding.py b/tst/unit/task_step/test_snapshot_env_env_binding.py
index 6c18e5e2..990ca780 100644
--- a/tst/unit/task_step/test_snapshot_env_env_binding.py
+++ b/tst/unit/task_step/test_snapshot_env_env_binding.py
@@ -2,8 +2,8 @@
 
 Matching on `env_id` alone sends every branch's snapshot to one deployment: the
 first-appended env gets snapshotted k times while the other rollouts' state is
-never captured. Same defect #808 fixed for `deploy_agent` and #817 for
-`load_artifact`, on the step that reads the state back out.
+never captured. The same defect as in `deploy_agent` and `load_artifact`, on the
+step that reads the state back out.
 
 `snapshot_env` has three resolution modes rather than one, so the filter is
 exercised against each: `env_instance_id` (exact, untouched), `env_id`, and the
@@ -22,7 +22,7 @@
 from agent_env.task_step.task_steps.deploy_env import DeployEnvTaskStep
 from agent_env.task_step.task_steps.snapshot_env import EnvSnapshotResult, SnapshotEnvTaskStep
 
-ENV_ID = "haltbench-rollout-env-0gr8479d"
+ENV_ID = "rollout-env"
 
 
 def _fake_env(sandbox_id: str):
diff --git a/tst/unit/task_step/test_universe_comparison.py b/tst/unit/task_step/test_universe_comparison.py
index c266164e..e2267279 100644
--- a/tst/unit/task_step/test_universe_comparison.py
+++ b/tst/unit/task_step/test_universe_comparison.py
@@ -1,7 +1,6 @@
 """Unit tests for the universe load/export roundtrip comparison.
 
-Locks in the semantic-comparison behavior, including the two correctness fixes
-from PR #403 review:
+Locks in the semantic-comparison behavior, including two correctness cases:
   * large financial amounts in the epoch range must NOT be coerced to timestamps
     (otherwise distinct amounts collapse to the same second — a silent false-negative);
   * a DROPPED non-empty nested-dict key is real data loss and must fail, while an
@@ -16,7 +15,7 @@
 
 
 class TestFinancialAmountsNotTimestamps:
-    """Greptile P1: numbers in the epoch range are amounts, not timestamps."""
+    """Numbers in the epoch range are amounts, not timestamps."""
 
     def test_distinct_amounts_in_epoch_range_differ(self):
         # Both fall in the old 1.4e9–2.0e9 "epoch seconds" window; sub-second precision
@@ -34,7 +33,7 @@ def test_numeric_string_matches_number(self):
 
 
 class TestNestedDictKeyAsymmetry:
-    """Greptile P1: dropped universe keys fail; added export keys are enrichment."""
+    """Dropped universe keys fail; added export keys are enrichment."""
 
     def test_dropped_nonempty_key_fails(self):
         assert _vals_equal(