diff --git a/plugins/linkedin-studio/references/algorithm-signals-reference.md b/plugins/linkedin-studio/references/algorithm-signals-reference.md
index 3d129bc..7a53adc 100644
--- a/plugins/linkedin-studio/references/algorithm-signals-reference.md
+++ b/plugins/linkedin-studio/references/algorithm-signals-reference.md
@@ -1,216 +1,190 @@
-# LinkedIn Algorithm Signals Reference (April 2026)
+# LinkedIn Algorithm Signals Reference (2026)
 
-Quick reference for ranking signals, weights, and penalties. For detailed context, see SKILL.md.
+**Single source of truth** for what the 2026 LinkedIn feed-ranking system rewards.
+Every other file in this plugin cites this one — do not restate magnitudes elsewhere,
+link here instead.
 
-## Positive Signals (Ranked by Weight)
+## How to read this file
 
-### Tier 1: Highest Impact
+The 2026 feed is ranked by an **LLM-based relevance system** (live in 2026; LinkedIn has
+**no publicly verifiable production name or go-live date** — see the model note below).
+Almost every "coefficient" circulating in the creator community is **third-party,
+observational, and moves year-to-year.** So this reference encodes **ordering + the two
+officially-named signals + directional magnitudes with a source and a confidence per
+claim** — never hard coefficients to optimize against.
 
-| Signal | Weight | Notes |
-|--------|--------|-------|
-| Saves | 10x | Top signal - content worth returning to. Users who save = 130% higher chance of following. Only ~3% of posts reach save-worthy status. |
-| Shares to feed | 8x | Public endorsement, strong distribution signal |
-| Expert comments | 7-9x | Comments from verified domain experts trigger broader reach |
-| Shares via DM | 6x | Private sharing still heavily valued |
-| Comments 15+ words | 2.5x | Quality engagement, 2x impact vs shorter comments |
-| Skill endorsements | +43% | Relevant skill endorsements boost content distribution |
+- **Confidence: high** = officially confirmed by LinkedIn, or convergent across multiple
+  large-N studies.
+- **Confidence: medium** = single credible large-N source, or convergent direction with a
+  contested magnitude.
+- **Confidence: low / directional** = practitioner heuristic, no primary source. Treat as
+  a hypothesis to test on your own account, not a fact.
 
-### Tier 2: High Impact
+**Rule of thumb: trust the *ordering*, test the *number*.**
 
-| Signal | Weight | Notes |
-|--------|--------|-------|
-| Profile-content match | +35-40% | 360Brew validation - profile must prove expertise on post topic |
-| Dwell time >30s | +25% | Reading/watching content signals quality |
-| Comments (any) | 1x | Base engagement unit - 15x more reach than likes, 5x more than reshares |
-| First-hour velocity | +20-30% | 15+ engagements in first hour unlocks broader distribution |
+## Officially-named ranking signals (the only two LinkedIn confirms by name)
 
-### Tier 3: Moderate Impact
+| Signal | Direction | Source | Confidence |
+|--------|-----------|--------|------------|
+| Dwell time | Time spent on a post is a ranking input (active vs passive tasks; long-dwell modeled). No public weight. | LinkedIn Eng — "Leveraging Dwell Time" (2024) | high |
+| Topic / interest relevance | Content matched to a viewer's interests is distributed — *including beyond your network* for strong content. | Tim Jurka, Head of Feed AI (2025-08-11) | high |
 
-| Signal | Weight | Notes |
-|--------|--------|-------|
-| Delayed engagement (24-72h) | 4-6x boost | Algorithm resurfaces quality content days after publication |
-| Profile views from post | +10-15% | Interest signal, potential follower conversion |
-| Click "see more" | +5-10% | Hook worked, engagement signal |
-| Reactions (all types) | 0.2x | 5x less valuable than comments |
-| Connection relevance | Variable | New connections get priority visibility for ~1 week |
+Everything below this line is direction + sourced estimate, not officially-weighted.
 
-## Negative Signals (Penalties)
+## Engagement order (not coefficients)
 
-| Signal | Penalty | Notes |
-|--------|---------|-------|
-| 5+ hashtags | -68% | Spam signal, triggers AI classifier |
-| AI-generated comments | -30% reach, -55% engagement | Detected and penalized — use human comments only |
-| Engagement pods | Shadow-ban | LinkedIn VP: goal to make pods "entirely ineffective". Comment velocity + account relationship analysis active |
-| Third-party script comments | Removed | Comments via automation tools removed from "Most Relevant" feed |
-| Off-topic for profile | -40-60% | 360Brew failure - profile doesn't validate expertise |
-| External link in body | -25-40% | Platform retention focus - use first comment instead |
-| Engagement bait phrases | -30-50% | "Comment YES if...", "Tag someone who...", "Type 1 for..." |
-| Multiple posts <3 hours | -25% | Spam-like behavior |
-| Posting gap >5 days | -15-25% | Consistency break, algorithm loses trust |
-| Posts under 1,000 chars | -25% | Too short for meaningful dwell time |
-| Posts over 2,500 chars | -32% | Diminishing returns, reader fatigue |
-| Promotional carousels | -60-70% | Algorithm penalizes selling vs. educating |
-| Link preview cards | -30-40% | Always remove preview if including links |
-| Generic connection requests | Variable | Low acceptance rate hurts profile score |
+The defensible spine is the **order**, not the multiplier:
 
-## Engagement Velocity Targets
+> **saves > shares > quality comments > reactions/likes**
 
-| Time | Target | Warning Zone | Action if Warning |
-|------|--------|--------------|-------------------|
-| 5 min | 2-3 | 0 | Check if posted at wrong time |
-| 15 min | 5-8 | <3 | Engage in comments to spark activity |
-| 30 min | 10-15 | <5 | Consider if hook is working |
-| 60 min | 15-25 | <10 | Post may not reach Stage 3 distribution |
-| 90 min | 25-40 | <15 | Limited viral potential |
+| Signal | Direction / estimate | Source | Confidence |
+|--------|----------------------|--------|------------|
+| Saves | Top engagement signal; also a follow-graph signal (saving a post raises the author's next-post feed odds). ≈ 5x a like / ≈ 2x a comment in single-vendor data. | AuthoredUp, Vertebrae, van der Blom (1.8M) | medium |
+| Shares (feed + DM) | Strong distribution signal; public endorsement. | van der Blom (1.8M) | medium |
+| Quality comments (15+ words) | Substantive comments outweigh short ones; **comment ≈ 2x a like** (quality-scored, single vendor). The popular "comment = many-x a like" claim is **unverified folklore** — dropped. | AuthoredUp (NLP-quality-scored) | medium |
+| Reactions / likes | Baseline engagement unit (≈ 1x). | van der Blom (1.8M) | medium |
 
-**Critical threshold:** 15+ engagements in first hour = unlocks 2nd/3rd degree distribution.
+> **Note on the old "comment = 15x" / "= 5x" framing:** there is no primary source for it.
+> The "5x" was the **saves** figure mis-assigned to comments. Encode the order above; do
+> not quote a comment multiplier.
 
-## Content Format Multipliers (2026)
+## Content format
 
-| Format | Reach Multiplier | Engagement Rate | Best For |
-|--------|------------------|-----------------|----------|
-| PDF/Carousel | 3.4x reach | 1.92% engagement | Frameworks, guides, step-by-step. 7 slides optimal (5-10 range), 25-50 words/slide. 35% click-through minimum or penalty |
-| Multi-image | 1.3x reach | 6.60% engagement | Before/after, comparisons, processes. Best for 5K-10K follower accounts |
-| Polls | 1.64x reach (declining) | 1.5-2% | Audience research only. Declining effectiveness in 2026 |
-| Video (60s) | 1.4x reach | Variable | Personal connection. Vertical 9:16 gets distribution boost. 30% completion rate minimum or zero reach. Always add captions (85% watch muted) |
-| Text-only | 1.17x reach | 3-5% | Thought leadership, stories, opinions. Generates best comment quality |
-| Link posts | -25-40% | <1% | Avoid if possible. Use first comment for links |
+| Format | Direction / estimate | Source | Confidence |
+|--------|----------------------|--------|------------|
+| Documents / carousels | **Top organic format (~7%, Socialinsider, company-page per-impression).** "Carousel" = PDF document post (LinkedIn removed native carousels Dec 2023). The 7% / 21.8% / 49.5% spread across studies is a **denominator artifact**, not disagreement about the winner. | Socialinsider (1.3M), Buffer (2M), Metricool (673K) | high (rank) / medium (number) |
+| Native video | #2 format and **declining**; add captions (most watch muted). No hard aspect-ratio gate — 4:5 / 1:1 preferred, captions are the enforceable spec. | Socialinsider; van der Blom | medium |
+| Text-only | Most resilient format; generates the best comment quality. | Buffer, van der Blom | medium |
+| Multi-image | Strong, slightly below documents. | Socialinsider | medium |
+| Polls | Declining effectiveness; audience research only. | van der Blom | low / directional |
+| Link posts (link in body) | Underperform — see external links below. | Ordinal (900K) | medium |
 
-## Optimal Ranges
+> The personal-profile **per-post baseline** (~2.0–2.6%) is a *different denominator* from
+> a format benchmark — never present an account baseline as a carousel rate.
 
-| Element | Optimal | Warning |
-|---------|---------|---------|
-| Post length | 1,200-1,800 chars | <1,000 (-25%) or >2,500 (-32%) |
-| Hook length | <140 chars | >140 truncated on mobile "see more" |
-| Hashtags | 3-4 | 5+ triggers -68% penalty |
-| Video length | 60 seconds | <30s low dwell, >90s retention drops. 30% completion gate |
-| Posting frequency | 3-5x/week | <2x loses consistency, >2x/day can fatigue |
-| Carousel slides | 7 slides | <5 too short, >10 diminishing returns, >15 completion drops 40% |
-| Caption (carousel) | <500 chars | Focus attention on slides |
-| About section | 2,600 chars | Use all available space, front-load keywords |
-| Headline | 220 chars | Include target audience + outcome |
+## External links (in post body)
 
-## 2026 Reach Context
+| Claim | Statement | Source | Confidence |
+|-------|-----------|--------|------------|
+| Reach effect | **Correlational reach reduction (~38% in 2026; contested band ~19–60% across studies).** It is a moving number, not a flat tax. | Ordinal (900K, p<0.001); van der Blom; DigitalApplied | medium |
+| Intent | **LinkedIn denies an *intentional* penalty** (Sr. Director Product, Aug 2025): no penalty "if the post leads with value" — the effect is engagement-driven. | Matt Navarra (relaying LinkedIn) | medium |
+| First comment | **Neither a magic fix nor a confirmed penalty** — contested. Lead with standalone value; native formats are the durable answer. | Ordinal; practitioner blogs (no large-N) | low |
 
-Overall organic reach declined significantly in 2026. This affects everyone — focus on relative performance (your posts vs your baseline), not absolute numbers.
+> **Design rule:** value-first matters more than link location. Soften any enforcing hook
+> from a hard "−X% penalty" mechanic to "body links correlate with lower reach — prefer a
+> first comment, but lead with value either way."
 
-| Metric | Change | Notes |
-|--------|--------|-------|
-| Total reach | -47% YoY | Platform-wide decline |
-| Video content | -72% YoY | Poor video penalized harder, good video still rewarded |
-| Text posts | -34% YoY | Most resilient format |
-| Company pages | ~1.6% of followers | Personal profiles outperform company pages 8x |
-| Posting cadence | 2-5x/week | Sweet spot unchanged despite reach decline |
+## Early-engagement window + evergreen resurfacing
 
-**Implication:** The algorithm rewards precision over broadcast. Smaller, engaged audiences outperform large but passive ones. 1:1 connections are now more valuable than follower count.
+| Claim | Statement | Source | Confidence |
+|-------|-----------|--------|------------|
+| Golden window | **60–90 min** (90 is the 2026 consensus); the **first 15–30 min** is the highest-leverage sub-window (~70% of reach decided there). | Buffer; Expandi; van der Blom | high |
+| First-hour velocity | Strong early engagement unlocks broader distribution. Directional, not a fixed threshold. | van der Blom | medium |
+| Evergreen resurfacing | The relevance model **can resurface strong-save / high-dwell posts days-to-weeks later** on viewer intent (posts now live 2–3 weeks vs days). **No** confirmed fixed "24–72h reinjection" rule — it is intent-driven and irregular. | AuthoredUp | medium (direction) / low (timing) |
 
-## Posting Time Windows (CET/European Audience)
+## Profile / topic alignment
 
-| Day | Peak Time | Notes |
-|-----|-----------|-------|
-| Tuesday | 8-9 AM, 10-11 AM | Best overall day |
-| Wednesday | 8 AM, 12 PM | Second best |
-| Thursday | 9 AM-1 PM | Extended peak window, highest engagement day |
-| Friday | Before 3 PM | Drops sharply after 3 PM |
-| Monday | 7-9 AM | Good for weekly kickoff content |
-| Weekend | Avoid | -50%+ engagement vs weekdays |
+| Claim | Statement | Source | Confidence |
+|-------|-----------|--------|------------|
+| Topic alignment is a ranking input | Real and officially confirmed (qualitatively): topic/interest relevance drives distribution, including beyond your network. | Tim Jurka (2025-08-11) | high |
+| Off-topic reach reduction magnitude | **No primary source** states a discrete off-topic reach-reduction figure. Treat profile/topic alignment as a real input; **do not quote a percentage.** | — | n/a (figure removed) |
 
-**For global audiences:** Post 8-11 AM your local time to catch multiple zones.
+## AI-content down-rank (officially confirmed — justifies the de-AI gate)
 
-**Industry variations:**
-- B2B Professional Services: 7-9 AM, 12-1 PM, 5-6 PM
-- Healthcare/Education: 10 AM-2 PM concentrated
-- Financial Services: Wed-Thu 10 AM-12 PM
+| Claim | Statement | Source | Confidence |
+|-------|-----------|--------|------------|
+| AI-slop suppression | LinkedIn confirmed an **active program** suppressing (1) generic AI-written posts/comments, (2) automation tools, (3) attention-bait video. ML models distinguish "original thinking" from "posts lacking substance"; flagged posts are **reach-suppressed (reportedly to first-degree), not deleted.** | VP & Exec Editor Laura Lorenzetti (2026-05-19) | high |
+| Correlational engagement gap | Likely-AI posts saw ~45% less engagement (correlational). | Originality.ai (8,795 posts) | medium |
+| Engagement-pod crackdown | Auto-comments demoted out of "Most Relevant", scoped to own network; repeat offenders restricted. | VP Product Gyanda Sachdeva (2026-02-16) | high |
 
-## Four-Stage Distribution Model
+> **Enforce what LinkedIn *named*** — personal substance, original thinking, concrete
+> specifics, genuine voice — not an unverified SEO "tell-list."
 
-| Stage | Timing | What Happens | Your Action |
-|-------|--------|--------------|-------------|
-| 1. Quality Classifier | 0-30s | AI spam/quality check + 360Brew profile validation | Ensure profile matches post topic |
-| 2. Initial Test | 0-90min | 6-10% of connections see post | Stay active, respond to all comments |
-| 3. Extended Distribution | 1-24h | 2nd/3rd degree if velocity good | Continue engagement, add value in comments |
-| 4. Long-tail | 24-72h+ | Evergreen circulation. Delayed engagement now yields 4-6x better performance. Algorithm resurfaces high-quality older content | Let compound effects work — high-dwell posts stay active up to 7 days |
+## Buzzwords
 
-**Stage 2 threshold:** 15+ engagements in first hour = unlock Stage 3.
+Buzzword avoidance is **editorial guidance for clarity, not a measured reach mechanic.**
+No primary source ties specific words to a reach penalty. Keep the buzzword list (it
+improves writing); do not justify it as "reduces reach."
 
-## Depth Score (2026)
+| Claim | Source | Confidence |
+|-------|--------|------------|
+| Specific phrasing reads better than corporate generic | Inc. (editorial) | low / directional |
+| A semantic ranker *may* indirectly favor specific over generic phrasing | inferred | low (not confirmed) |
 
-LinkedIn's primary content ranking metric. Measures actual engagement duration, not surface interactions. The feed now uses LLM-generated embeddings and transformer-based Generative Recommender models for semantic relevance scoring.
+## The deployed ranking model — what we can and cannot say
 
-| Factor | Impact | Notes |
-|--------|--------|-------|
-| Time spent reading/watching | Primary signal | Replaced likes as #1 ranking factor |
-| Slide completion (carousel) | High | Each slide click = engagement signal. 7 slides optimal for completion |
-| Video watch percentage | High | 30% minimum completion or zero distribution |
-| Scroll-back behavior | Medium | Re-reading = strong quality signal |
-| Save after reading | Highest | Save + high dwell = maximum distribution boost |
+> **An LLM-based relevance-ranking system is live on LinkedIn in 2026.**
+> **No public name. No deployment date.**
 
-**Distribution impact:**
-- High-dwell posts: active in feeds up to **7 days**
-- Low-dwell posts: dead after **24 hours**
-- First-hour dwell time determines post lifecycle
-- Minimalist carousel design: +12% completion rate vs complex backgrounds
+| Claim | Statement | Source | Confidence |
+|-------|-----------|--------|------------|
+| A live LLM relevance system exists | Confirmed in direction by LinkedIn's 2026 communications. | LinkedIn comms (2026) | high |
+| Production name | **Not publishable as fact.** The most-cited arXiv paper (2501.16450) is a Jan-**2025** *pre-production research* model (V1.0, 150B params, offline parity only), **withdrawn 2025-08-23**. A circulating "Generative Recommender / Hristo Danchev" engineering-post citation was independently flagged as **likely fabricated** — do not propagate. | arXiv 2501.16450; Gemini provenance flag | high (on the negative claim) |
+| Deployment date | No primary source. The "early-2026" date is third-party extrapolation from the paper's Jan-**2025** date. **Do not assert a date.** | — | n/a |
 
-## 360Brew Profile Validation (January 2026)
+## Operational heuristics (directional — test per account)
 
-**The algorithm validates your profile BEFORE distributing content.**
+These are creator-community heuristics with no primary-source weights. Use as starting
+hypotheses, not targets. Confidence: **low / directional** for every row.
 
-| Validation Criteria | What It Checks | Fix If Failing |
-|---------------------|----------------|----------------|
-| About Section | Specific expertise claims, domain terminology | Rewrite with concrete expertise statements |
-| Experience Section | Impact statements with metrics | Add quantified achievements |
-| Content History | Previous posts on this topic, anecdotal evidence | Requires 90 days of aligned posting for full expertise categorization. Topic mismatch limits reach directly |
-| Network Quality | Connected to professionals in your field | Connect with relevant domain experts |
-| Engagement Patterns | Do you comment on posts in your expertise area? | Daily: 3-5 thoughtful comments in your domain |
+### Engagement velocity (first 90 min)
 
-**Profile-content mismatch = -40-60% reach before anyone sees your post.**
+| Time | Rough target | If well below |
+|------|--------------|---------------|
+| 15 min | a few | check timing / hook |
+| 30 min | building | engage in comments |
+| 60–90 min | momentum | golden window closing |
 
-## Quick Decision Rules
+### Posting time windows (CET / European audience)
+
+| Day | Commonly-cited peak |
+|-----|---------------------|
+| Tue | 8–11 AM (often best overall) |
+| Wed | 8 AM, 12 PM |
+| Thu | 9 AM–1 PM (extended) |
+| Fri | before 3 PM |
+| Mon | 7–9 AM |
+| Weekend | weaker |
+
+*For global audiences: post 8–11 AM local to catch multiple zones.*
+
+### Quick decision rules
 
 | Situation | Decision |
 |-----------|----------|
-| Linking? | First comment, not body |
+| Linking? | First comment, lead with value either way |
 | Multiple ideas? | Split into separate posts |
-| Off your usual topic? | Update profile first, or skip the post |
-| Haven't posted in 5+ days? | Post simple text-only first to rebuild momentum |
-| Using a poll? | Add substantial caption (300+ chars) |
-| Tagging people? | Max 5, must be genuinely relevant |
-| Including hashtags? | 3-4 max, specific > generic |
-| Video or text? | Text for thought leadership, video for personal connection |
-| Carousel or text? | Carousel for frameworks/guides, text for stories/opinions |
-| Comment or like first? | Always comment first (higher algorithmic value) |
-| New connection posted? | Comment within first week (algorithm priority window) |
+| Off your usual topic? | Topic alignment is a real input — stay on-domain or accept lower reach |
+| Video or text? | Text for thought leadership, video (captioned, 4:5/1:1) for connection |
+| Carousel or text? | Documents for frameworks/guides, text for stories/opinions |
+| Comment or like first? | Comment (higher in the engagement order) |
 
-## Comment Strategy Quick Reference
+### Comment strategy (CEA)
 
-| Target | Why | How Often |
-|--------|-----|-----------|
-| Your posts (self-comment) | Spark discussion, add resources | Wait 10 min, then 2-4 over 90 min |
-| Whale posts (100K+ followers) | Visibility exposure | 2-3 early comments daily |
-| New connections | Algorithm priority window | Within first week of connecting |
-| Inner circle (5-10 peers) | Mutual support network | Daily genuine engagement |
-| Ideal customers | Warm outreach pipeline | Find in relevant comment sections |
+1. **Compliment** — a specific point you appreciated
+2. **Expand** — your insight or related experience
+3. **Ask** — a question to continue dialogue
 
-**Comment formula (CEA):**
-1. **Compliment** - Specific point you appreciated
-2. **Expand** - Your insight or related experience
-3. **Ask** - Question to continue dialogue
+Minimum quality: 15+ words, genuine perspective. AI-generated / "Great post!" comments are
+actively suppressed (see AI-slop down-rank).
 
-**Minimum quality:** 15+ words, adds genuine perspective.
+## 2026 reach context
 
-## Red Flags to Avoid
-
-- Engagement pods (LinkedIn VP: goal to make pods "entirely ineffective" — comment velocity analysis and account relationship patterns actively detect manufactured engagement)
-- Pitch-slapping in DMs
-- Posting same content as company page
-- Random topics outside demonstrated expertise
-- "Great post!" style generic comments (harm reach even without pod involvement)
-- Excessive self-promotion (>20% of content)
-- Tagging unrelated people for reach
-- Using AI-generated comments (55% engagement penalty)
+Organic reach declined platform-wide in 2026 — focus on **relative performance** (your
+posts vs your own baseline), not absolute numbers. Smaller engaged audiences outperform
+large passive ones. (Direction: high confidence; exact YoY %: directional, varies by
+source.)
 
 ---
 
-*Last updated: April 2026*
+*Last updated: 2026-05. Maintained as the single canonical algorithm statement; cite, do
+not restate.*
 
-*Sources: Research synthesis from Richard van der Blom (Algorithm Research 2025), Lara Acosta (SLAY Framework), 360Brew algorithm analysis, LinkedIn Engineering Blog, Buffer (2M+ post analysis), Sprout Social (2.5B engagements), Justin Welsh, Jasmin Alic, Sahil Bloom case studies. April 2026 update: ALM Corp (LLM architecture analysis), Botdog (360Brew deep dive), DesignACE (engagement signal weights), ContentIn (format strategy guide), UseVisuals (carousel statistics 2026), Visla (video format 2026)*
+*Sources (per-claim quality/confidence noted inline): arXiv 2501.16450 (pre-production
+research paper, withdrawn 2025-08-23); LinkedIn Engineering — "Leveraging Dwell Time" (2024); Tim Jurka, Head of Feed
+AI (2025-08-11); Laura Lorenzetti, VP & Exec Editor (2026-05-19); Gyanda Sachdeva, VP
+Product (2026-02-16); Matt Navarra relaying LinkedIn Sr. Director Product (Aug 2025);
+Ordinal link-penalty study (900K, p<0.001); Socialinsider (1.3M); Buffer (2M+); Metricool
+(673K); AuthoredUp (621K, NLP-quality-scored); van der Blom Algorithm Insights 2025 (1.8M);
+Originality.ai (8,795 posts); Inc. (buzzword editorial). Full provenance: research brief
+`docs/remediation/research/01-linkedin-algorithm-signals.md`.*