Job codebase-questions-answered-with-checked-citations · revised 12 October 2026
Get 25 questions about your codebase answered, with file and line receipts
Written answers to up to 25 questions about one repository, each citing file and line at a named commit, with "not in the code" where the code does not say. A script checks every citation.
You might be seeing
- Questions about how a thing works take a week to answer from the code
- A chat answer sounded right and was wrong
No passwords, keys, card details or admin invites needed to start.
What usually happened
Understanding unfamiliar code costs weeks, and an AI chat answer can sound right and be wrong. Without file and line receipts a buyer cannot check an answer, and a model can invent reasoning around a real citation.
Who it’s for: A new CTO, a first engineering hire, an agency taking over a project or an acquirer reading unfamiliar code.
Usually starts when: You have to understand a codebase you did not write and need answers you can verify rather than a chatbot you have to trust.
The result: You receive written answers to up to 25 questions you supply about one repository at one commit. Each answer cites file and line and quotes the text, and says "not in the code" where the code does not say. A script confirms every citation exists and every quote matches, and five trick questions act as a spot check; passing them does not show we never invent an answer.
Check whether this job fits
Answer these without sending code, tickets or logins. Nothing is submitted unless you choose to contact us.
What you get
- The answers document
- The citation check log
- The list of questions we could not answer from the code
Included
- One repository at one named commit, read-only
- Up to 25 questions you supply, plus 5 trick questions your engineer writes (answers not in the code), which we do not see until the run
- Each answer cites file and line and quotes the text
- "Not in the code" for anything the code does not say
- A citation check log
Not included
- Opinions on code quality, security or whether to buy or hire
- Answers that need running the application or reading production data
- A hosted chat assistant for your code
- Reasoning beyond what the cited code shows
- Code that you are not allowed to share
How we know it’s done
Agreed with you before work starts. Each check produces evidence you keep.
A script confirms 100% of citations exist at the named commit and every quoted excerpt matches the file.
Evidence: The citation check log with zero mismatches.
Your engineer rates 10 sampled answers (picked by script with a fixed seed) and at least 9 are correct (the true rate is then at least 60% with 95% confidence), and none gives reasoning the quoted code does not support.
Evidence: The 10 answers and their ratings.
All 5 trick questions, written by your engineer and unseen by us until the run, whose answers are not in the code, are answered "not in the code". This is a spot check, not a rate.
Evidence: The 5 questions and answers.
Every question has an answer or appears in the could-not-answer list with a reason.
Evidence: A script count of questions against answers.
Sign-off. Your engineer reads the sample ratings and signs off in writing. Payment follows sign-off.
If it fails. If the agreed checks do not pass, you do not pay for this fixed scope. We hand over what we found and agree whether to stop or re-quote; no surprise work.
When it fits, and when we stop
It fits when
- One repository at one commit that you may share through a company-controlled route
- Up to 25 questions written down
- A named engineer who rates 10 sampled answers
We stop and tell you if
- The code is under a licence or contract that forbids sharing
- The questions are about runtime behaviour that the code cannot show
- The repository is too large to read in one pass and you do not want a quote
What could go wrong
The job writes documents and changes nothing. Our copy of your code is deleted when the job ends.
Scroll the table sideways to read it all.
| Risk | How we handle it |
|---|---|
| Real citations wrap invented reasoning. | Your engineer rates 10 answers; any answer whose reasoning the quoted code does not support counts as a severe error and fails the sample. Five trick questions are a spot check, not proof. |
| The answers expose weak spots in your code. | The document goes only to the person you name, and our copies are deleted when the job ends. |
| Code and comments carry instructions meant for the model. | Code and comments are treated as data: the model gets the text and returns answers a script checks, with no tools that change anything, no network and no secrets. |
A separate automated reviewer checks the work before hand-over. A named engineer on your side rates 10 sampled answers. No human supervisor is included unless your agreement names one, and at launch the work is largely automated.
How we deliver
We arrange the work and independent review, then show you the result against the agreed checks. You keep authority over your systems.
- Agree the commit, the questions and what counts as an answer
- Receive the code read-only at one commit
- Answer each question with file, line and a quoted excerpt
- Run the citation script and answer five trick questions with "not in the code"
- Separate automated reviewer checks the answers; your engineer rates 10
- Hand over the answers, log and list of unanswered questions
This is a one-off job, not emergency cover or a subscription. We confirm eligibility, price, start window and delivery date before you accept, and work starts only after agreed inputs, secure access and permissions are in place. Model usage up to a cap we state in writing is in the price, and an enquiry creates no charge or booking.
Need to keep it working?
To ask more questions each month, a monthly service is offered separately.
Ongoing work is separately scoped and quoted: no monitoring, response-time guarantee or automatic subscription is included in this job.
Mention the responsibility you need in your enquiry.
What you can check
This is a new service. We have not delivered this job for a client yet.
Other ways to get this done
- Use your editor's built-in assistant and check each answer by opening the file it names.
- Ask the previous developer for a one-hour walkthrough and record it.
Questions
Is this a chatbot for our code?
No. It is one set of written answers at one commit, each with a citation you can check.
What if the code does not say?
We write "not in the code" and list what we would need to know.
Can the code stay on our systems?
Not for this job; we read it in a sandbox. If code may not leave, use a tool inside your network.
Send an enquiry
Send us
- The language, rough size and what you need to understand
- Whether the code may leave your systems
- No code in the first enquiry
Later, once you agree
- Read-only code at one commit through a company-controlled route
- The questions, in writing
- The 5 trick questions, written by your engineer and sent after the other answers are done
- The name of the engineer who rates answers
You keep the repository; we read one commit in a sandbox through a dedicated bot account with the least access that works (read-only first). Code and comments are treated as data: the model gets the text and returns answers a script checks, with no tools that change anything, no network and no secrets. Our automated agents read the files you send in a sandbox; we make no "no training" or "no storage" promise here, and the agreement names the model provider, country, retention and training status before anything is shared.
Email fallback: open your mail app
If website submission is unavailable, review and send the fallback email yourself. An email fallback is not a website receipt. Or write to hello@syntheticindustry.ai with “codebase-questions-answered-with-checked-citations” as the subject.