Extend practice feedback and problem bank checks

[jserv]

This branch widens what a practice session gives back and tightens what the problem bank has to prove. Candidates can write their own test cases and see what their code returns, practice without a camera when the interview is not recorded, and get a post-interview debrief that states the hint rung they reached and the level they practiced for, while the report is refused if it names the published problem. The lobby keeps a longer history whose reports can be reopened, keeps due reviews across level changes, resets a failed review to one day, and shows progress by topic. The bank now accepts original exercises, every judge needs five cases including a boundary one, and a C starter that returns an array has to say who frees it. On the server, a provider whose credentials are refused leaves rotation, and report-model reviews during quiet pauses are capped per interview. CI gains a check that the release image matches the manifest's Rust version and a weekly run that requires the pinned avatar model, and the legacy migration paths for the git hooks and vendored files are removed. The interview contract moves to bundle 8, live prompt 3 and report prompt 8, with each step recorded in docs/interview-contract-versions.md.

Verified with ./scripts/test.sh at each of the 18 commits, not only the tip, and all pass; the tip runs 672 Rust tests and 523 browser tests with nothing skipped. The guarantees the commits claim were checked by breaking the guarded line and confirming a test failed: a prompt edit without a version bump, the interim review cap and its limit of 72, 401 and 403 provider refusal, the published-name check, judge case coverage, the C ownership note, and the version parity check. The candidate case fixes were reproduced before and after in Chromium. One upstream test this branch does not touch, setup_page_accepts_credentials_and_writes_the_primary_config_file in tests/cli.rs, failed once under load and passed on every rerun.

Deliberately left for follow-up: a report for an original exercise can still mention LeetCode, because the published-name check keys off a source title those exercises do not have; a sixth candidate case blocks the test run instead of being refused on its own; a null expected value breaks the palindrome checker; a mistyped class-case method breaks the whole C++ or Java compile; only four of the five candidate cases reach the reviewer; the parity check passes when neither version can be read; and the failed-review test cannot fail as written. A checkout that still holds web/vendor/avatar/jim.vrm from before the model moved to the browser cache now fails verify-vendor.sh until the file is deleted.

---

Summary by cubic

Gives practice sessions richer feedback and tightens what the problem bank has to prove. Candidates can add their own typed test cases, skip the camera when the interview isn't recorded, and get a debrief with the hint rung, volunteered hints, and practiced level; reports that name the published problem are repaired rather than delivered.

Practice and reports

• Browser-observed candidate cases run beside judge cases and reach both the interviewer and the report prompt; class judges read return types from the contract so a case without an expected value still builds.
• The lobby keeps a compact history of 500 attempts, reopens reports past the full-report cap, keeps due reviews across level changes, resets a failed review to one day only when it is the newest review outcome, and shows progress by topic.
• Valid review history and candidate-case checker failures stay visible when neighboring entries are malformed.
• Quiet-pause interim reviews are capped per interview (CODETRIAL_MAX_INTERIM_REVIEWS, default 12, max 72).
• The interview contract moves to bundle 8, live prompt 3, report prompt 8, and report schema 2; schema 1 reports remain scored.

Problem bank and tooling

• The bank accepts original problems, adding the original fixed-capacity-ring-buffer; every judge needs five cases including a boundary case, and C starters that return arrays must say who frees them.
• Tree contracts now admit empty trees, and malformed bank input is refused consistently.
• Provider projects rejected with 401 or 403 leave rotation until their credential is replaced.
• CI checks that the release Rust image matches Cargo.toml's rust-version and runs a weekly avatar-model check.
• Legacy git-hook and vendored-avatar migration paths are removed; a stale web/vendor/avatar/jim.vrm must be deleted before verify-vendor.sh passes.

^{Written for commit ebf1d89. Summary will update on new commits.}

[cubic-dev-ai]

2 issues found across 4 files (changes from recent commits).

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="problem-bank/problems.json">

<violation number="1">
P2: The new bound contradicts the scenario this problem ships with. The constraint now claims an empty tree is valid (0 nodes), but problem-bank/variants.json for this problem states the opposite: the contract reads "receives a non-empty tree of 1 to 10^4 nodes" and clarification 3 says "Between 1 and 10^4 stations, so there is always at least the gateway." The judge in problem-bank/judges.json also has no empty-tree case. Because src/agent/prompts.rs joins variant.constraints into the interviewer's prompt alongside the contract, the interviewer now reads conflicting input domains, and a candidate may be penalized either for failing to handle `root = []` or for assuming it can occur. Either keep the published-style "1 <= number of nodes" bound, or actually support empty trees: update the variant contract and clarification and add a judge case for `root = []` (expected "[]").</violation>
</file>

<file name="src/agent/problem_variants.rs">

<violation number="1" location="src/agent/problem_variants.rs:903">
P2: The constraints line for Hop Depth Signal Means now allows an empty tree (0 nodes), but the contract still says the tree is non-empty (1 to 10^4 nodes) and the clarification says there is always at least the gateway. That contradiction tells candidates an empty input may be tested while the contract rules it out. Align them: either keep 1 as the lower bound, or update the contract and clarification (and define the empty return) if allowing an empty tree is intended.</violation>
</file>

<sub>Tip: Review your code locally with the cubic CLI to iterate faster.

Re-trigger cubic</sub>

[cubic-dev-ai]

P2: The constraints line for Hop Depth Signal Means now allows an empty tree (0 nodes), but the contract still says the tree is non-empty (1 to 10^4 nodes) and the clarification says there is always at least the gateway. That contradiction tells candidates an empty input may be tested while the contract rules it out. Align them: either keep 1 as the lower bound, or update the contract and clarification (and define the empty return) if allowing an empty tree is intended.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At src/agent/problem_variants.rs, line 903:

<comment>The constraints line for Hop Depth Signal Means now allows an empty tree (0 nodes), but the contract still says the tree is non-empty (1 to 10^4 nodes) and the clarification says there is always at least the gateway. That contradiction tells candidates an empty input may be tested while the contract rules it out. Align them: either keep 1 as the lower bound, or update the contract and clarification (and define the empty return) if allowing an empty tree is intended.</comment>

<file context>
@@ -900,7 +900,7 @@ pub const PROBLEM_VARIANTS: &[(&str, ProblemVariant)] = &[
         brief: &["A field sensor network relays data through a tree of stations, each forwarding to at most two stations below it. Every station reports a signal reading, and the dashboard shows the mean reading for all stations that are the same number of hops from the gateway.", "Implement meanReadingPerHop(root), where root is the TreeNode for the gateway and each node's value is that station's reading, and return the mean reading for each hop distance, starting with the gateway."],
         contract: "meanReadingPerHop(root) receives a non-empty tree of 1 to 10^4 nodes with 32-bit signed values and returns a list of the unrounded arithmetic mean of node values at each depth, root depth first. Means are compared as exact numbers, so integer-valued means like 3 and 3.0 both match.",
-        constraints: &["1 <= number of nodes <= 10^4", "-2^31 <= Node.val <= 2^31 - 1"],
+        constraints: &["0 <= number of nodes <= 10^4", "-2^31 <= Node.val <= 2^31 - 1"],
         clarifications: &[("Should the means be rounded or truncated?", "No. Return each mean as a floating-point value."), ("How large can readings be?", "Each reading is any 32-bit signed integer, from -2^31 to 2^31 - 1."), ("How many stations are there?", "Between 1 and 10^4 stations, so there is always at least the gateway."), ("What order should the means be in?", "From the gateway outward: the gateway's own reading first, then one hop away, and so on.")],
         follow_ups: &["The dashboard now wants the median reading per hop distance instead of the mean. What changes?", "Stations report continuously and readings update. How would you keep the per-hop means current?", "The tree is too wide to hold a full hop level in memory. What would you do?"],
</file context>

Suggested change

	constraints: &["0 <= number of nodes <= 10^4", "-2^31 <= Node.val <= 2^31 - 1"],
	constraints: &["1 <= number of nodes <= 10^4", "-2^31 <= Node.val <= 2^31 - 1"],