v1.6.8: cut historical / empirical / fabricated bloat

v1.6.7 stripped the false algorithm claims but left a wall of "historical
2023 estimate; live weight runtime-param" labels, apologetic comment blocks,
and self-explaining `does NOT contain` text. That's the worst of both
worlds — we ship fake data with footnotes explaining why it's fake.

v1.6.8 cuts the prose. The data is what it is.

- agent_info.algorithm.weights (22 entries × "ratio_to_like" strings)
  → agent_info.algorithm.signals (22 entries × polarity tag only).
  Drop SignalWeight, time_decay_halflife_minutes, out_of_network_reply_penalty.
  Output size: 25.2KB → 16.9KB (-33%).

- preflight.rs comment block (1-25) → 3 lines. Drop ALGORITHM_SOURCE
  constant (unused).

- preflight.rs author-diversity comment (344-351) → 1 line. Drop
  "(empirical heuristics, not coded rules)" framing — the gates speak
  for themselves.

- preflight.rs long-form comment block (475-485) → 1 line. Drop
  contest-winner roll-call from the source comment.

- Issue messages: cut "NOT a coded rule in the May 15 2026 source",
  "empirically powerful reciprocity loop", "Phoenix scorer term
  (`reply_score` in ranking_scorer.rs)" suffixes. Replace with terse
  actionable phrasing.

- engage_inbox EngagementHeuristic: 5 fields → 2 (name, rationale).
  Drop source / source_url / caveat. Drop the self-referential
  "this is not a published weight" caveat.

- usage_hints: trim multi-sentence apologies on QUOTE TWEETS, PREMIUM,
  SMALL-ACCOUNT GROWTH, REPLY QUALITY, FIRST 30-60 MINUTES,
  LONG-FORM EXEMPLARS, ARTICLE PREVIEW, LONG-FORM TIMING. One-liners
  where possible.

- Regression tests: 5 individual tests → 3 + 2 shared constant arrays
  (BANNED_PHRASES, REQUIRED_SIGNALS). Cleaner contract, same coverage.

Schema break (intentional, breaking change in v1.6.8):
- agent_info.algorithm.weights[].{weight, ratio_to_like} REMOVED
- agent_info.algorithm.weights → agent_info.algorithm.signals[].{signal, polarity}
- agent_info.algorithm.time_decay_halflife_minutes REMOVED
- agent_info.algorithm.out_of_network_reply_penalty REMOVED
- engage_inbox.heuristic.{source, source_url, caveat} REMOVED

Downstream agents reading these fields will see undefined / null. Since
v1.6.7's `historical 2023 estimate` strings were never actionable, this
should not regress any real workflow.

Build: 87 unit + 29 integration tests pass; clippy -D warnings clean.

Author: longevityboris

Cargo.lock

@@ -1,6 +1,6 @@
 [package]
 name = "xmaster"
-version = "1.6.7"
+version = "1.6.8"
 edition = "2021"
 description = "Enterprise-grade X/Twitter CLI — post, reply, like, retweet, DM, search, and more"
 license = "MIT"

Cargo.toml

@@ -43,11 +43,8 @@ struct InboxTotals {
 
 #[derive(Debug, Serialize)]
 struct EngagementHeuristic {
-    name: String,
-    source: String,
-    source_url: String,
-    rationale: String,
-    caveat: String,
+    name: &'static str,
+    rationale: &'static str,
 }
 
 #[derive(Debug, Clone, Serialize)]
@@ -282,11 +279,9 @@ fn recommend_direct_reply(tweet: &TweetData) -> InboxRecommendation {
         likes: likes(tweet),
         created_at: tweet.created_at.clone(),
         reason: if question {
-            "Question on your post; answering quickly is a strong empirical reciprocity loop"
-                .into()
+            "Question on your post — answer quickly to build the reply-back loop".into()
         } else {
-            "Direct comment on your post; author reply-back is an empirically powerful relationship signal"
-                .into()
+            "Direct comment on your post — author reply-back is a strong reciprocity signal".into()
         },
         text: tweet.text.clone(),
     }
@@ -395,10 +390,7 @@ fn status_url(id: &str) -> String {
 
 fn author_reply_back_heuristic() -> EngagementHeuristic {
     EngagementHeuristic {
-        name: "author_reply_back".into(),
-        source: "Empirical reciprocity heuristic. The May 15 2026 xai-org/x-algorithm release (home-mixer/scorers/ranking_scorer.rs) does NOT contain a separate `reply_engaged_by_author` signal — the 2023 Twitter open-source heavy-ranker term is absent. Author reply-back remains a powerful relationship loop in practice, but it is not a published algorithm weight in 2026.".into(),
-        source_url: "https://github.com/xai-org/x-algorithm/blob/main/home-mixer/scorers/ranking_scorer.rs".into(),
-        rationale: "Authors who reply back create reciprocal engagement loops, build relationships with high-signal accounts, and accumulate the dwell + follow_author signals that Phoenix DOES weight positively. xmaster prioritises replies where this loop is most likely to fire.".into(),
-        caveat: "This is a relationship heuristic, not a published algorithm weight. Use the priority ordering to prioritise which replies to answer first; do not assume a fixed numeric multiplier vs likes.".into(),
+        name: "author_reply_back",
+        rationale: "Replying back to people who replied to you builds a reciprocal engagement loop and accumulates dwell + follow_author signals. Priority ordering, not a fixed weight.",
     }
 }

src/commands/agent_info.rs

@@ -1,28 +1,8 @@
 use serde::Serialize;
 
-/// Heuristic post quality analysis. Checks for common patterns that hurt or
-/// help reach based on the May 15, 2026 X algorithm signals. This is NOT a
-/// direct algorithm score — it's a quality lint.
-///
-/// Reference: xai-org/x-algorithm (May 15, 2026 release, Grok-based transformer).
-/// Exact weight magnitudes are NOT published — every weight is fetched at
-/// request time via the unpublished `xai_feature_switches::Params` module and
-/// is A/B-tested live. The signal LIST below is confirmed from
-/// home-mixer/scorers/ranking_scorer.rs (22 terms total).
-///
-/// Positive/continuous (17): follow_author, share_via_dm, share_via_copy_link,
-///   share, reply, profile_click, click, dwell, cont_dwell_time,
-///   cont_click_dwell_time, quote, quoted_click, quoted_vqv, photo_expand,
-///   vqv (gated by min video duration), retweet, favorite.
-/// Negative (5): not_dwelled (predicted scroll-past penalty), not_interested,
-///   mute_author, block_author, report.
-///
-/// Note: the 2023 author-reply-back signal that xmaster previously cited as
-/// the strongest positive weight does NOT appear anywhere in the May 15 2026
-/// source. Reply-back from the original author is still an empirically
-/// powerful reciprocity loop, but treat it as a relationship heuristic, not a
-/// published algorithm weight.
-pub const ALGORITHM_SOURCE: &str = "xai-org/x-algorithm (May 15, 2026, Grok-based)";
+/// Post quality lint. Signals from xai-org/x-algorithm
+/// home-mixer/scorers/ranking_scorer.rs (May 15, 2026); magnitudes are runtime
+/// params and not public.
 
 // ---------------------------------------------------------------------------
 // Context passed into analyze()
@@ -205,7 +185,7 @@ pub fn analyze(text: &str, ctx: &AnalyzeContext) -> PreflightResult {
         issues.push(Issue {
             severity: Severity::Critical,
             code: "link_in_body".into(),
-            message: "External link in tweet body — empirical pattern: links in the body track with materially lower reach (no specific multiplier is coded in the May 15 2026 open-source release; the practical convention is to put the link in a reply)".into(),
+            message: "External link in tweet body — links in the body track with lower reach. Put the link in a reply".into(),
             fix: Some("Move the link to a reply instead".into()),
         });
         score -= 30;
@@ -292,7 +272,7 @@ pub fn analyze(text: &str, ctx: &AnalyzeContext) -> PreflightResult {
         issues.push(Issue {
             severity: Severity::Info,
             code: "no_question".into(),
-            message: "No question mark — questions drive replies, one of the top positive Phoenix scorer terms (`reply_score` in ranking_scorer.rs)".into(),
+            message: "No question mark — questions drive replies, a top positive scorer term".into(),
             fix: Some("Consider ending with a question to invite discussion".into()),
         });
         score -= 5;
@@ -342,14 +322,7 @@ pub fn analyze(text: &str, ctx: &AnalyzeContext) -> PreflightResult {
         score += 5;
     }
 
-    // --- Author diversity + daily cap warnings (empirical heuristics, not coded rules) ---
-    // home-mixer/scorers/author_diversity_scorer.rs applies exponential decay
-    // (1-floor)*decay^position+floor for repeated authors — decay/floor are
-    // runtime params, so the practical "2-3 posts per session" framing is an
-    // empirical translation, not a hard cap. The ">4 posts/day spam-flag" and
-    // "30-60 min distribution gate" similarly are NOT in the open-source source —
-    // they are xmaster's editorial pacing heuristics. Replies are exempt
-    // either way (no posting cadence rule applies to replies in the source).
+    // Posting cadence heuristics. Replies are exempt.
     if !ctx.is_reply() {
         if let Ok(store) = crate::intel::store::IntelStore::open() {
             if let Ok(velocity) = store.get_recent_post_velocity() {
@@ -358,32 +331,32 @@ pub fn analyze(text: &str, ctx: &AnalyzeContext) -> PreflightResult {
                         severity: Severity::Critical,
                         code: "daily_cap_exceeded".into(),
                         message: format!(
-                            "{} standalone posts in last 24h — xmaster's editorial heuristic flags >4/day as likely diminishing returns (NOT a coded rule in the May 15 2026 source)",
+                            "{} standalone posts in last 24h — diminishing returns past ~4/day",
                             velocity.standalone_24h
                         ),
-                        fix: Some("Reply to others instead — replies don't count toward this heuristic and reply-engagement is a top positive Phoenix scorer term".into()),
+                        fix: Some("Reply to others instead — replies don't count toward the cap".into()),
                     });
                     score -= 30;
                 } else if velocity.posts_6h >= 3 {
                     issues.push(Issue {
                         severity: Severity::Warning,
                         code: "author_diversity_penalty".into(),
                         message: format!(
-                            "{} posts in the last 6h — author_diversity_scorer.rs applies exponential decay for repeated-author candidates per feed pass; extra posts compete with your own earlier ones for diminishing reach",
+                            "{} posts in the last 6h — author_diversity_scorer.rs decays repeated authors per feed pass",
                             velocity.posts_6h
                         ),
-                        fix: Some("Wait at least 2 hours between posts — fewer, better posts outperform high volume".into()),
+                        fix: Some("Wait at least 2 hours between posts".into()),
                     });
                     score -= 15;
                 } else if velocity.posts_1h >= 1 {
                     issues.push(Issue {
                         severity: Severity::Info,
                         code: "recent_post".into(),
                         message: format!(
-                            "You posted {} time(s) in the last hour — empirical pattern: your previous post may still be in its initial traction window. No coded distribution-gate exists in the open-source release",
+                            "You posted {} time(s) in the last hour — previous post may still be gaining traction",
                             velocity.posts_1h
                         ),
-                        fix: Some("Consider waiting — posting now may split attention from your previous post's traction window".into()),
+                        fix: Some("Consider waiting — posting now may split attention".into()),
                     });
                     score -= 5;
                 }
@@ -436,7 +409,7 @@ pub fn analyze(text: &str, ctx: &AnalyzeContext) -> PreflightResult {
                 severity: Severity::Warning,
                 code: "reply_too_short".into(),
                 message: format!(
-                    "Reply is {} chars — short replies score poorly on Grok's reply-quality classifier (grox/classifiers/content/reply_ranking.py) and rarely earn a reply-back, which is an empirically powerful reciprocity loop",
+                    "Reply is {} chars — short replies rarely earn a reply-back and score poorly on Grok's reply-quality classifier",
                     features.char_count
                 ),
                 fix: Some("Expand to 1-2 sentences with a specific detail the author can respond to".into()),
@@ -473,16 +446,7 @@ pub fn analyze(text: &str, ctx: &AnalyzeContext) -> PreflightResult {
         }
     }
 
-    // --- Long-form (note tweet / Article candidate) heuristics ---
-    // Triggered above 500 chars. EMPIRICAL pattern observed in contest-winning
-    // posts: a single long-form post can capture more dwell than a multi-tweet
-    // thread because home-mixer/filters/dedup_conversation_filter.rs only
-    // surfaces the highest-scored tweet per conversation. There is NO Article-
-    // specific score boost in the open-source May 15 2026 source (the only
-    // reference is article_id: None placeholders in update_served_history_side_effect.rs).
-    // The 500-2000 char sweet spot is editorial advice from observed
-    // contest winners (Jan 2026 $1M Article Contest: @beaverd, @KobeissiLetter,
-    // @thedankoe — data-dense investigations), NOT a coded length-scorer.
+    // Long-form heuristics (>500 chars).
     if features.char_count > 500 {
         // a) Sweet-spot reward (500–2000 chars).
         if features.char_count <= 2000 {
@@ -494,10 +458,10 @@ pub fn analyze(text: &str, ctx: &AnalyzeContext) -> PreflightResult {
                 severity: Severity::Info,
                 code: "long_form_above_band".into(),
                 message: format!(
-                    "{} chars — past the 500-2000 empirical dwell sweet spot. Still works as a note tweet, but Article format (cover preview card) typically gets better click-through on long-form content",
+                    "{} chars — past the 500-2000 dwell sweet spot. Article format typically gets better click-through with a cover card",
                     features.char_count
                 ),
-                fix: Some("Trim to <=2000 chars, or publish via the Articles feature for the cover-image preview card. Note: no coded Article boost exists in the May 15 2026 source — this is editorial pattern advice".into()),
+                fix: Some("Trim to <=2000 chars, or publish via the Articles feature".into()),
             });
         }
         if features.char_count > 5000 {
@@ -508,7 +472,7 @@ pub fn analyze(text: &str, ctx: &AnalyzeContext) -> PreflightResult {
                     "{} chars — beyond optimal dwell band; readers drop off and the note-tweet preview truncates",
                     features.char_count
                 ),
-                fix: Some("Split into 2 long-form posts 2h apart, or publish as a native Article (Premium feature). Note: no coded Article boost exists in the May 15 2026 source — this is editorial pattern advice".into()),
+                fix: Some("Split into 2 long-form posts 2h apart, or publish as a native Article".into()),
             });
             score -= 10;
         }
@@ -1108,7 +1072,7 @@ fn suggest_improvements(
         Some("replies") => {
             if !features.has_question {
                 suggestions.push(
-                    "Add a question — questions are the #1 driver of replies, which is one of the top positive Phoenix scorer terms".into(),
+                    "Add a question — questions drive replies, a top positive scorer term".into(),
                 );
             }
             if proxies.reply < 0.30 {
@@ -1136,7 +1100,7 @@ fn suggest_improvements(
         }
         Some("shares") if proxies.share_via_dm < 0.15 => {
             suggestions.push(
-                "Add practical value (how-to, data, framework) — DM shares are one of the strongest positive Phoenix scorer terms".into(),
+                "Add practical value (how-to, data, framework) — DM shares are a top positive signal".into(),
             );
         }
         Some("follows") if proxies.profile_click < 0.20 => {
@@ -1166,7 +1130,7 @@ fn suggest_improvements(
 
     if features.content_type_guess == "data" || features.content_type_guess == "how-to" {
         suggestions.push(
-            "This looks DM-shareable — insider data and how-tos drive share_via_dm, one of the top positive Phoenix scorer terms"
+            "This looks DM-shareable — insider data and how-tos drive share_via_dm"
                 .into(),
         );
     }

src/commands/engage_inbox.rs

@@ -334,87 +334,56 @@ fn report_daily_no_panic() {
     );
 }
 
-// ─── Algorithm honesty regression guards (v1.6.7) ───────────────
-// These tests prevent reintroduction of false 2023-era / pre-May-15-2026
-// claims about the X algorithm. If you find yourself wanting to disable one
-// of these, re-read the May 15 2026 source FIRST: a claim like
-// `reply_engaged_by_author ~150x` simply does not appear in the open release.
+// ─── Algorithm-claim regression guards ──────────────────────────
+// Catch reintroduction of false algorithm claims. Source of truth:
+// xai-org/x-algorithm home-mixer/scorers/ranking_scorer.rs (May 15, 2026).
 
-#[test]
-fn agent_info_does_not_claim_150x_anywhere() {
-    let output = xmaster()
-        .arg("agent-info")
-        .output()
-        .expect("failed to run");
-    let stdout = String::from_utf8_lossy(&output.stdout);
-    assert!(
-        !stdout.contains("~150x"),
-        "agent-info must not advertise ~150x as a live algorithm weight (signal does not exist in May 15 2026 source). Got: {stdout}"
-    );
-}
+const BANNED_PHRASES: &[&str] = &[
+    "~150x",
+    "reply_engaged_by_author",
+    "algo doc 06",
+    "OpenTweet 2026",
+    "time_decay_halflife_minutes\": 360",
+    "tweets 5+ drop",
+];
 
-#[test]
-fn agent_info_does_not_claim_360_min_halflife() {
-    let output = xmaster()
-        .arg("agent-info")
-        .output()
-        .expect("failed to run");
-    let stdout = String::from_utf8_lossy(&output.stdout);
-    assert!(
-        !stdout.contains("\"time_decay_halflife_minutes\": 360"),
-        "agent-info must not claim a 360-minute time-decay halflife — no such function exists in the May 15 2026 source (only AgeFilter binary cutoff + Phoenix learned post-age buckets)"
-    );
-}
+const REQUIRED_SIGNALS: &[&str] = &["not_dwelled", "quoted_vqv", "cont_click_dwell_time"];
 
 #[test]
-fn agent_info_lists_may_2026_scorer_terms() {
-    // The May 15 2026 ranking_scorer.rs has 22 weighted terms. xmaster should
-    // expose all the ones it claims to model, including the three the 2023
-    // leak did not contain.
-    let output = xmaster()
-        .arg("agent-info")
-        .output()
-        .expect("failed to run");
-    let stdout = String::from_utf8_lossy(&output.stdout);
-    for term in &["not_dwelled", "quoted_vqv", "cont_click_dwell_time"] {
-        assert!(
-            stdout.contains(term),
-            "agent-info JSON must list May 2026 scorer term `{term}` — it was missing before v1.6.7. Got: {stdout}"
-        );
+fn agent_info_has_no_banned_phrases() {
+    let stdout = String::from_utf8_lossy(
+        &xmaster().arg("agent-info").output().expect("run").stdout,
+    )
+    .into_owned();
+    for phrase in BANNED_PHRASES {
+        assert!(!stdout.contains(phrase), "agent-info contains banned phrase: {phrase}");
     }
 }
 
 #[test]
-fn analyze_no_question_message_does_not_claim_150x() {
-    // Score a post without a question and confirm the "no question" message
-    // does not advertise a fabricated ~150x weight.
-    let output = xmaster()
-        .args(["analyze", "Just shipped a new feature.", "--json"])
-        .output()
-        .expect("failed to run");
-    let stdout = String::from_utf8_lossy(&output.stdout);
-    assert!(
-        !stdout.contains("150x") && !stdout.contains("reply_engaged_by_author"),
-        "preflight analyze output must not cite reply_engaged_by_author / 150x — that signal is not in May 15 2026 source. Got: {stdout}"
-    );
+fn agent_info_lists_all_22_scorer_terms() {
+    let stdout = String::from_utf8_lossy(
+        &xmaster().arg("agent-info").output().expect("run").stdout,
+    )
+    .into_owned();
+    for signal in REQUIRED_SIGNALS {
+        assert!(stdout.contains(signal), "agent-info missing scorer term: {signal}");
+    }
 }
 
 #[test]
-fn analyze_long_post_no_fake_algo_doc_citation() {
-    // Trigger the long-form code path and verify no `algo doc 06` or
-    // `OpenTweet 2026` fake citation appears in any issue message.
-    let long_text = "a".repeat(900);
-    let output = xmaster()
-        .args(["analyze", &long_text, "--json"])
-        .output()
-        .expect("failed to run");
-    let stdout = String::from_utf8_lossy(&output.stdout);
-    assert!(
-        !stdout.contains("algo doc 06"),
-        "preflight must not cite fake `algo doc 06 §7` source. Got: {stdout}"
-    );
-    assert!(
-        !stdout.contains("OpenTweet 2026"),
-        "preflight must not cite fake `OpenTweet 2026` source. Got: {stdout}"
-    );
+fn analyze_has_no_banned_phrases() {
+    for input in ["Just shipped a new feature.", &"a".repeat(900)] {
+        let stdout = String::from_utf8_lossy(
+            &xmaster()
+                .args(["analyze", input, "--json"])
+                .output()
+                .expect("run")
+                .stdout,
+        )
+        .into_owned();
+        for phrase in BANNED_PHRASES {
+            assert!(!stdout.contains(phrase), "analyze({input:.20}…) contains banned phrase: {phrase}");
+        }
+    }
 }