Stop scoring hints as a numeric deduction

Bundle 10 retires the instruction that every hint should meaningfully
reduce the communication score. A hint is evidence about independence,
but how much to subtract was left to the model, so two interviews with
the same hint count were marked differently for it. Hints stay as
context the reviewer weighs against everything else.

The register in docs is now read back by the bundle test, because it is
what a maintainer consults before a bump and it had nothing holding it
to the constants.

Author: jserv

docs/interview-contract-versions.md

@@ -8,10 +8,11 @@ can select it.
 
 ## The active bundle
 
-Bundle 9: live prompt 3, report prompt 8, rubric 1, report schema 2.
+Bundle 10: live prompt 3, report prompt 9, rubric 1, report schema 2.
 
 | Bundle | Introduced |
 |---|---|
+| 10 | Hints remain qualitative context for candidate independence rather than automatic numeric communication-score deductions. |
 | 9 | Reports keep the fixed mid-level hiring bar and state the optional level the candidate practiced for beside it. |
 | 8 | Candidate-authored test cases reach the live interviewer and report brief, while judge pass totals remain separate. |
 | 7 | The post-interview server stamp adds optional debrief, topics, and practice level fields. `tests/golden/report-schema.json` remains the model output shape only; server-stamped fields are versioned at the browser sanitizer. |

src/agent.rs

@@ -115,9 +115,9 @@ const ROUND_TRANSITION_SKEW: std::time::Duration = std::time::Duration::from_sec
 /// `the_time_warning_threshold_is_the_same_number_on_both_sides`.
 pub const TIME_WARNING_S: u64 = 300;
 
-pub const INTERVIEW_CONTRACT_BUNDLE_VERSION: u32 = 9;
+pub const INTERVIEW_CONTRACT_BUNDLE_VERSION: u32 = 10;
 pub const LIVE_PROMPT_VERSION: u32 = 3;
-pub const REPORT_PROMPT_VERSION: u32 = 8;
+pub const REPORT_PROMPT_VERSION: u32 = 9;
 pub const RUBRIC_VERSION: u32 = 1;
 pub const REPORT_SCHEMA_VERSION: u32 = 2;
 

src/agent/prompts.rs

@@ -840,8 +840,7 @@ Score two independent dimensions from 0 to 100:
    optimization, and accurately answered follow-ups. Also consider completeness
    of Situation, Task, personal Action, and Result only if the interviewer actually
    asked a behavioral question. If none was asked, say behavioral communication
-   was not assessed and do not deduct for it. Consider independence too: each hint
-   should meaningfully reduce this score; {} hint(s) were given."#,
+   was not assessed and do not deduct for it."#,
         input.duration_min,
         input.elapsed_min,
         input.problem.title,
@@ -857,8 +856,7 @@ Score two independent dimensions from 0 to 100:
         input.hints_used,
         input.hint_rung,
         volunteered_hints,
-        test_summary,
-        input.hints_used
+        test_summary
     )
 }
 

tests/agent.rs

@@ -740,6 +740,10 @@ fn prompt_golden_digest_matches_versions() {
             (3, 8),
             "1be28b91ea6148c278f262c01d61670563ff4752b37ad517675284e9304a5825",
         ),
+        (
+            (3, 9),
+            "4f9e1aac81d6479abbe454d3eeb8dad27f414c2b311bc6cf89165dad9566ce1f",
+        ),
     ]
     .into_iter()
     .find_map(|(candidate, digest)| (candidate == versions).then_some(digest));
@@ -5323,17 +5327,36 @@ fn generated_problem_metadata_exposes_no_private_rubric() {
 
 #[test]
 fn interview_contract_versions_are_one_closed_bundle() {
-    assert_eq!(INTERVIEW_CONTRACT_BUNDLE_VERSION, 9);
+    // The doc is the register a maintainer reads before bumping anything, so a
+    // bump that leaves it behind is worse than no register at all. Read the
+    // active-bundle line back rather than trusting the two to be edited
+    // together, and require the table to carry a row for that bundle.
+    let doc = std::fs::read_to_string("docs/interview-contract-versions.md")
+        .expect("the contract register is readable");
+    assert!(
+        doc.contains(&format!(
+            "Bundle {INTERVIEW_CONTRACT_BUNDLE_VERSION}: live prompt {LIVE_PROMPT_VERSION}, \
+             report prompt {REPORT_PROMPT_VERSION}, rubric {RUBRIC_VERSION}, report schema \
+             {REPORT_SCHEMA_VERSION}."
+        )),
+        "docs/interview-contract-versions.md does not name the active bundle"
+    );
+    assert!(
+        doc.contains(&format!("\n| {INTERVIEW_CONTRACT_BUNDLE_VERSION} | ")),
+        "the bundle table has no row for {INTERVIEW_CONTRACT_BUNDLE_VERSION}"
+    );
+
+    assert_eq!(INTERVIEW_CONTRACT_BUNDLE_VERSION, 10);
     assert_eq!(LIVE_PROMPT_VERSION, 3);
-    assert_eq!(REPORT_PROMPT_VERSION, 8);
+    assert_eq!(REPORT_PROMPT_VERSION, 9);
     assert_eq!(RUBRIC_VERSION, 1);
     assert_eq!(REPORT_SCHEMA_VERSION, 2);
     assert_eq!(
         interview_contract_json(),
         json!({
-            "bundleVersion": 9,
+            "bundleVersion": 10,
             "livePromptVersion": 3,
-            "reportPromptVersion": 8,
+            "reportPromptVersion": 9,
             "rubricVersion": 1,
             "reportSchemaVersion": 2,
         })

tests/browser/replay-render.test.js

@@ -523,7 +523,7 @@ test("the report card this page renders names no finding either", () => {
     said,
     new Set([
       "(.md)", "(Sxlang)", "(editor", "(none", "-", "/", "0", "01:05", "1", "10",
-      "100", "2", "2.", "2;", "3", "37", "7", "70", "8", "9", "95%",
+      "100", "2", "2.", "2;", "3", "37", "7", "70", "8", "95%",
       "Approach", "Chain", "CodeTrial.", "Coding", "Committee", "Common",
       "Communication", "Contract", "Done", "Download", "Evidence",
       "FACE_MISSING", "Follow-ups", "Framework", "Given:", "HIRE", "Held",

tests/fixtures/framework-evaluation-cases.json

@@ -51,7 +51,7 @@
     {
       "id": "explicit-hint-use",
       "coverage": ["explicit-hint-use"],
-      "reaction": {"kind":"report", "code":"def two_sum(nums, target): return []", "required":["each hint", "2 hint(s) were given"], "forbidden":["0 hint(s)"]},
+      "reaction": {"kind":"report", "code":"def two_sum(nums, target): return []", "required":["2 total", "never as a numeric deduction"], "forbidden":["meaningfully reduce this score", "0 total"]},
       "transcript":"Interviewer suggested tracking prior complements; candidate then described a map.",
       "evidence":[{"phase":"algorithm","source":"candidate_speech","kind":"inferred","confidence":70,"summary":"Described the map after an explicit interviewer hint."}],
       "hintsUsed":2, "round":"none", "assessed":["Algorithm"]

tests/golden/prompts.json

@@ -421,9 +421,9 @@ const textEncoder = new TextEncoder();
 /// function-local, moving it left the whole suite green with the supported-card
 /// branch no longer rendering, which is the defect a local constant invites.
 export const ACTIVE_CONTRACT = {
-  bundleVersion: 9,
+  bundleVersion: 10,
   livePromptVersion: 3,
-  reportPromptVersion: 8,
+  reportPromptVersion: 9,
   reportSchemaVersion: 2,
   rubricVersion: 1,
 };