Can the check tell a real read of the private file from a mention of one?

Alice answers the second review with pull request #120, the newest commit in the Client. Her checker no longer takes the open count on trust: it counts reads of the private file in the trace itself, which each arm’s record now carries (independent_trace_text). The committed transcript, canary-reads.txt, stays in the repository, and from here on no check reads it, so Bob’s trace questions now edit the trace the checker does read. Alice says deleting or forging that trace “is now caught; it wasn’t before”, and that the upload and the configuration are now checked for all four arms.

Select Solve for Bob’s third review. Its own text was never found: what it said survives only second-hand, in the message of Alice’s next commit, and the lesson labels it so. Bob asks the seven questions again, word for word, and three more. Each replaces the real read of the private file, in the arm that does read it, with something that only looks like one: a prose mention, a failed open, an open of a file with the same name somewhere else. Alice made no claim about those; they are new.

Here the checker catching 6 of Bob’s 10 edits, and still passing every assertion for the rest: deleting the trace; replacing the real read with a prose mention of the private file; replacing the real read with a failed open of the private file; replacing the real read with an open of a same-named file elsewhere; a reproduction of what each edit did is the lens’s last row. Open the lens on an accepted line: Peek opens the record as Bob edited it, at the trace he changed.

Powered by WebContainers
Files
Preparing Environment
  • npm install
  • npm run dev