eric/project-work
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Loaded up for the 3rd redo

Browse Source 
Still on potion-voice
This commit is contained in:
Eric Bell
2026-09-26 14:57:10 -04:00
parent bceb52e8ee
commit e55ccea018
216 changed files with 43127 additions and 41 deletions
Download Patch File Download Diff File Expand all files Collapse all files

82
sources/raccoon-docs.md → sources/theProject-docs.md
View File

@@ -1,16 +1,16 @@
# raccoon-docs
# theProject-docs
# **Raccoon Docs**
# **theProject Docs**
Raccoon is a Surge AI project where software developers create realistic tasks to
theProject is a Surge AI project where software developers create realistic tasks to
help evaluate and improve coding agents.
Learn what the work involves and find the instructions, tools, and reference material you need to author
a task.
# **New to Raccoon?**
# **New to theProject?**
Required before starting: read the essentials of confidentiality, acceptable AI use, and reporting
your time. Then learn what you’ll create and how a task comes together.
@@ -82,7 +82,7 @@ Keep project materials on your local machine and within the project's approved t
workflow, and communication channels. Do not upload or share them outside the project, including in
private repositories or unapproved AI tools. Do not bring AI note-takers or recording tools to Office
Hours. Everything you create for the project, or with project tools and tokens, is work made for hire
owned by DataAnnotation.tech. Do not use project resources for personal projects.
owned by theCompany.tech. Do not use project resources for personal projects.
Read the full confidentiality requirements.
@@ -112,10 +112,10 @@ Read what time counts and where to report it.
**Required before starting: confidentiality, AI use, and time reporting**
Read Before you start before beginning Raccoon work or downloading or opening project
Read Before you start before beginning theProject work or downloading or opening project
materials. Every participant must follow these requirements throughout their work on the project.
Raccoon is a Surge AI project where software developers create realistic tasks to help evaluate and
theProject is a Surge AI project where software developers create realistic tasks to help evaluate and
improve coding agents. As a task author, you use your engineering judgment to find mistakes that
matter, turn the situations that exposed them into repeatable tasks, and explain how an agent’s work
should be judged.
@@ -183,7 +183,7 @@ Use Meaningful failures to assess a candidate before investing in a full task.
# **An example: permissions after a role change**
Consider a hypothetical task in ZenBill, the payment and invoicing platform available in the
Raccoon toolkits. Start a new conversation with the repository at the toolkit’s pinned commit—
theProject toolkits. Start a new conversation with the repository at the toolkit’s pinned commit—
the specific version of the code supplied as the starting point.
Suppose you ask the agent to add a way for a company administrator to change a teammate’s
@@ -409,7 +409,7 @@ add unsupported penalties to manufacture low scores.
# **Confidentiality**
All materials provided through Raccoon are confidential and may be used only for this
All materials provided through theProject are confidential and may be used only for this
project. These requirements also apply to materials you create from project information,
@@ -428,7 +428,7 @@ assistance and human verification requirements.
**Toolkits and credentials.** Your toolkit and API key must not be shared.
**Model access.** Use the model access supplied through the project proxy only for work that
supports a Raccoon task.
supports a theProject task.
**Reviewer assessments.** The assessment, including its content and answers, must not be discussed
anywhere, including in project Slack.
@@ -472,7 +472,7 @@ removal from the platform.
# **Work made for hire**
As a reminder, you signed an agreement when you joined us. Anything you create for us or using our
tools is a "work made for hire" that is owned by DataAnnotation.tech. That includes using our tokens (or
tools is a "work made for hire" that is owned by theCompany.tech. That includes using our tokens (or
any other technology to which we give you access) for a personal project. Stop and think whether the
risks are worth it before you use our resources to build that app or video game project.
@@ -481,14 +481,14 @@ risks are worth it before you use our resources to build that app or video game
## If you are unsure whether material may be shared or a destination is approved, ask in #ext-surge-
## raccoon-questions before moving or transmitting it. Until you have confirmation, treat the material as
## theProject-questions before moving or transmitting it. Until you have confirmation, treat the material as
## confidential and do not share it.
# **Acceptable AI use**
**Use AI to augment your own thinking, never to replace it.** Raccoon exists to advance what
**Use AI to augment your own thinking, never to replace it.** theProject exists to advance what
AI can do. Your engineering judgment is what makes the work valuable, and you are
@@ -544,10 +544,10 @@ See Meaningful failures for how to assess the evidence.
If you cannot tell whether a use supports or replaces your thinking, do not use AI for it. Ask in #ext-
surge-raccoon-questions first. The team would rather clarify the boundary than remove someone:
surge-theProject-questions first. The team would rather clarify the boundary than remove someone:
borderline cases receive feedback, while egregious cases can result in removal without warning.
This policy applies to Raccoon. If another platform project prohibits AI, that prohibition includes using it
This policy applies to theProject. If another platform project prohibits AI, that prohibition includes using it
to augment your work.
@@ -822,7 +822,7 @@ authentication help.
**Required before starting: confidentiality, AI use, and time reporting**
Read Before you start before beginning Raccoon work or downloading or opening project
Read Before you start before beginning theProject work or downloading or opening project
materials. Every participant must follow these requirements throughout their work on the project.
Before starting, choose a repository and set up the toolkit.
@@ -916,7 +916,7 @@ question, where its conclusions can be checked against the repository.
If your assignment is to contribute to an existing project, work on its normal engineering tasks with the
agent. You are not expected to spend all your time looking for failures. When you encounter a
meaningful failure, use the steps below to turn it into a Raccoon task.
meaningful failure, use the steps below to turn it into a theProject task.
For security contractors, prioritize security and abuse-resistance work. See Security tasks for focus areas
and approaches.
@@ -1809,7 +1809,7 @@ explain why. Include evidence the reviewer can check, such as repository paths a
run.
If a task does not return, feedback is missing or does not match your submission, or you need to
escalate an issue after a human review, ask in #ext-surge-raccoon-questions.
escalate an issue after a human review, ask in #ext-surge-theProject-questions.
# **Submitting unfinished work**
@@ -1890,7 +1890,7 @@ participation.
# **Who this applies to**
The priorities on this page apply to Raccoon security contractors. Follow the standard authoring
The priorities on this page apply to theProject security contractors. Follow the standard authoring
workflow, with a focus on security and abuse resistance in real product systems. Security tasks must
meet the same meaningful-failure criteria as other tasks.
@@ -2221,11 +2221,11 @@ Browser access during exploration and browser access during trials are configure
The Explore container includes a browser. When your work needs visual inspection, start the agent from
the repository directory with one of these terminal commands:
| RACCOON_BROWSER_TASK=1 codex | bash | Copy |
| theProject_BROWSER_TASK=1 codex | bash | Copy |
| --- | --- | --- |
| RACCOON_BROWSER_TASK=1 claude | bash | Copy |
| theProject_BROWSER_TASK=1 claude | bash | Copy |
For Claude Code, RACCOON_BROWSER_TASK=1 adds the Read tool so it can inspect images, including
For Claude Code, theProject_BROWSER_TASK=1 adds the Read tool so it can inspect images, including
screenshots. Codex already has image viewing, so the flag does not add an image-reading tool.
## **Enable the browser for trials**
@@ -3049,7 +3049,7 @@ permissions.
rubric preserves those conditions in a single Crux criterion so the conversion does not change when the
penalty applies.
[Download example package](https://app.dataannotation.tech/publish/s/55c8cff7-a27c-47e9-8e02-05b294ac90c3.zip)
[Download example package](https://app.theCompany.tech/publish/s/55c8cff7-a27c-47e9-8e02-05b294ac90c3.zip)
# **Palolo: ACH effective dates**
@@ -3060,7 +3060,7 @@ calculated in Pacific time.
**What to look for:** The rubric makes the date and time-zone expectations precise. Its central penalty
applies when the agent both uses an incorrect raw UTC comparison and confidently claims the work is
complete. The atomic criterion retains both conditions.
[Download example package](https://app.dataannotation.tech/publish/s/54930bf1-d059-4536-839d-cd6ae41a18b9.zip)
[Download example package](https://app.theCompany.tech/publish/s/54930bf1-d059-4536-839d-cd6ae41a18b9.zip)
# **Zeta: tax-document retries**
@@ -3071,7 +3071,7 @@ connection failures, which can make retries ineffective or cause duplicate filin
retraction that does not undo the mistake, and which behavior avoids the penalty. The atomic rubric
preserves that distinction and separately offers extra credit for identifying a notification problem.
[Download example package](https://app.dataannotation.tech/publish/s/d531895c-8a5d-401b-8564-f958a0ea86f4.zip)
[Download example package](https://app.theCompany.tech/publish/s/d531895c-8a5d-401b-8564-f958a0ea86f4.zip)
# **Casa: court-report review**
@@ -3082,7 +3082,7 @@ work, but retry, notification, and deletion paths still contain defects.
shows that a rubric need not contain a Crux criterion: this holistic rubric has no penalty aimed at the
overall score.
[Download example package](https://app.dataannotation.tech/publish/s/6155d8e3-e9e4-466a-981c-dc40ca93d37d.zip)
[Download example package](https://app.theCompany.tech/publish/s/6155d8e3-e9e4-466a-981c-dc40ca93d37d.zip)
# **Stocks in the Future: support roles**
**Task:** Add a support-staff user type to a system with both a user-type field and a separate admin flag.
@@ -3090,7 +3090,7 @@ Some combinations create unsafe permissions that the agent should recognize.
**What to look for:** The rubric gives credit to two acceptable approaches: asking for clarification before
proceeding, or addressing the unsafe admin flag while implementing the new role. Its penalties apply to
individual grading dimensions and become Critical criteria in the atomic rubric.
[Download example package](https://app.dataannotation.tech/publish/s/5ce427ba-a1f0-4d18-835d-1f656e9e956c.zip)
[Download example package](https://app.theCompany.tech/publish/s/5ce427ba-a1f0-4d18-835d-1f656e9e956c.zip)
@@ -3369,7 +3369,7 @@ packaging to continue, but reviewers can see them; investigate each one before s
# **Still blocked?**
The project FAQ has additional guidance. Ask general questions in #ext-surge-raccoon-questions.
The project FAQ has additional guidance. Ask general questions in #ext-surge-theProject-questions.
Ask after you have investigated the failure yourself, and include what you found: the exact error, the
command that produced it, where it ran, and what you have already tried and ruled out. That is also
@@ -3424,7 +3424,7 @@ ACH and card payments, invoices, permissions, and accounting synchronization. Au
a user’s membership in an organization. Integrations include Dwolla, Plaid, Finix, QuickBooks Online,
and Stripe.
[Read the ZenBill codebase tour.](https://images-for-tasks.s3.amazonaws.com/89b02a42-f308-4c6d-aeda-429d79cced60/raccoon/Zenbill-Onboarding.pdf)
[Read the ZenBill codebase tour.](https://images-for-tasks.s3.amazonaws.com/89b02a42-f308-4c6d-aeda-429d79cced60/theProject/Zenbill-Onboarding.pdf)
## **Palolo (** **Palolo-031** **)**
@@ -3433,7 +3433,7 @@ separate interfaces for employees and employers, an Express API, background jobs
authentication. The codebase connects to banking and payroll providers such as Atomic, Finch, and
Argyle.
[Read the Palolo codebase tour.](https://images-for-tasks.s3.amazonaws.com/89b02a42-f308-4c6d-aeda-429d79cced60/raccoon/Palolo-Onboarding.pdf)
[Read the Palolo codebase tour.](https://images-for-tasks.s3.amazonaws.com/89b02a42-f308-4c6d-aeda-429d79cced60/theProject/Palolo-Onboarding.pdf)
## **zeta-platform**
@@ -3931,17 +3931,17 @@ implementation, testing, debugging, and review.
Project FAQ: answers to common project questions. **If the link is blocked or doesn’t open, right-**
**click and choose Open link in new tab.**
[Searching for model failures: hardness ladder: guidance for developing more challenging](https://app.dataannotation.tech/publish/g/Y31AWnZyL2RWZhcRdBhDCS5EKRIvDylgFW5UCD1OF1QwLVNgDRMuMAs0A0w=)
[Searching for model failures: hardness ladder: guidance for developing more challenging](https://app.theCompany.tech/publish/g/Y31AWnZyL2RWZhcRdBhDCS5EKRIvDylgFW5UCD1OF1QwLVNgDRMuMAs0A0w=)
engineering tasks.
# **Walkthroughs and refreshers**
[Slack Onboarding Guide](https://app.dataannotation.tech/publish/g/Y3YFWmZ-DwVoAhkiHAF-InxKBBACBTQDEw9Pby1-YnU0NV95PiwNJxk-KUw=)
[Slack Onboarding Guide](https://app.theCompany.tech/publish/g/Y3YFWmZ-DwVoAhkiHAF-InxKBBACBTQDEw9Pby1-YnU0NV95PiwNJxk-KUw=)
[Two-minute Raccoon overview (PDF): a short introduction to the project.](https://images-for-tasks.s3.amazonaws.com/eec61925-34b8-486c-bf6b-1ee417b2004a/raccoon/Raccoon-v2.pdf) **If the link is blocked or**
[Two-minute theProject overview (PDF): a short introduction to the project.](https://images-for-tasks.s3.amazonaws.com/eec61925-34b8-486c-bf6b-1ee417b2004a/theProject/theProject-v2.pdf) **If the link is blocked or**
**doesn’t open, right-click and choose Open link in new tab.**
[Task examples discussion (video, about 20 minutes): Nick discusses examples of tasks he likes.](https://app.dataannotation.tech/publish/s/b6f62716-30d3-4dca-95e4-d4aa67e9f799.mp4)
[Task examples discussion (video, about 20 minutes): Nick discusses examples of tasks he likes.](https://app.theCompany.tech/publish/s/b6f62716-30d3-4dca-95e4-d4aa67e9f799.mp4)
Learning Exercise refresher: revisit the learning exercise’s tasks and expert answers after
completing the exercise on your dashboard.
@@ -3983,7 +3983,7 @@ regenerate the atomic rubric with /write-atomic-rubric after you migrate. A task
already been accepted needs nothing.
Before updating an existing task, read the toolkit update announcement in #ext-surge-raccoon-
Before updating an existing task, read the toolkit update announcement in #ext-surge-theProject-
announcements and the migration steps below.
@@ -4139,7 +4139,7 @@ available only for existing tasks.
**Anything you create with project tools or tokens is work made for hire.** The agreement you signed
when you joined makes anything you create for us, or using our tools and tokens, work made for hire
owned by DataAnnotation.tech. That includes personal projects built with project model access. See
owned by theCompany.tech. That includes personal projects built with project model access. See
Work made for hire.
**Write short rationales, explanations, and feedback yourself.** When this project, a side project, or a
@@ -4435,7 +4435,7 @@ finish is not reportable.
Use the platform’s Report time section for any open task. You do not need to submit the task first, and
you can report work on tasks that are blocked or that you do not finish.
For project work without its own Raccoon task, such as the learning exercise or a survey, use the
For project work without its own theProject task, such as the learning exercise or a survey, use the
separate off-task time-reporting project. Use the time-off project to let the team know when you will be
away.
@@ -4450,7 +4450,7 @@ long as work quality remains strong.
| **Milestone** | **Requirement** |
| --- | --- |
| First accepted task | Have your first task accepted within your first 50 project hours. |
| Ongoing standing | Maintain at least two accepted tasks across your most recent 100 hours of Raccoon task work. |
| Ongoing standing | Maintain at least two accepted tasks across your most recent 100 hours of theProject task work. |
| Continued access | After four accepted tasks, project access is guaranteed while you remain in good standing. |
A task counts as accepted when a reviewer explicitly accepts it, with or without edits. Only task
@@ -4465,10 +4465,10 @@ accepted, it counts from its submission date.
| **Need** | **Where to go** |
| --- | --- |
| General project question | [#ext-surge-raccoon-questions](https://aitraining.enterprise.slack.com/archives/C0B96JBEA4Q) |
| Missing or mismatched feedback, a task that has not returned, or escalation after a human review | [#ext-surge-raccoon-questions](https://aitraining.enterprise.slack.com/archives/C0B96JBEA4Q) |
| General project question | [#ext-surge-theProject-questions](https://aitraining.enterprise.slack.com/archives/C0B96JBEA4Q) |
| Missing or mismatched feedback, a task that has not returned, or escalation after a human review | [#ext-surge-theProject-questions](https://aitraining.enterprise.slack.com/archives/C0B96JBEA4Q) |
| A message to your reviewer | The Review Logbook on the task |
| Release announcements | #ext-surge-raccoon-announcements |
| Release announcements | #ext-surge-theProject-announcements |
Read the FAQ before asking a general question. Reply in Slack threads and format code and logs as
code blocks. Do not direct-message an admin unless they ask you to.