v1.20.16 - SCORE VARIANCE EXPLANATION
Nov 28, 2025
- 🎯 NEW FEATURE: Full Report must explain score differences from Discovery
- - Added discovery_scores to research_context (shows Discovery's initial scores)
- - Full Report now sees Discovery scores in prompt
- - If Full Report scores any dimension >0.5 different, must explain in score_variance_explanation field
- 💡 BENEFIT: Understand WHY scores changed
- - Valid: "Discovery found 10+ snaps but deeper research shows it's only special teams"
- - Invalid: General skepticism without evidence
- 🔍 Improved transparency - see exactly what new evidence caused score changes
v1.20.15 - PASS RESEARCH CONTEXT: Discovery → Full Report
Nov 28, 2025
- 🐛 PROBLEM: v1.20.14 made prompts identical but scores STILL varied wildly
- - Discovery: 3.5/5 (found some data) vs Full Report: 1.4/5 (found NO data)
- 🔍 ROOT CAUSE: Identical prompts don't guarantee identical research - separate API calls find different data
- ✅ SOLUTION: Pass Discovery's research facts to Full Report
- 1. Discovery captures key facts in new research_context field (TE/FB usage, coaching, transfers, etc.)
- 2. research_context cached in localStorage when Discovery returns
- 3. Full Report checks cache and uses SAME facts for scoring
- 💡 BENEFIT: Both features score THE SAME DATA = identical scores guaranteed
- 🎯 Faster Full Report (no re-research), lower API costs, perfect consistency
v1.20.14 - TRULY IDENTICAL PROMPTS (Still Failed)
Nov 28, 2025
- 🐛 PROBLEM: v1.20.13 FAILED - Still scoring differently (Discovery 3.5, Full Report 2.5)
- 🔍 ROOT CAUSE: v1.20.13 only copied 2 sections, but 6 MAJOR differences remained
- ✅ FIX #1: Role description - Changed from "comprehensive analyst" to "SCREEN OUT bad fits"
- ✅ FIX #2: Added explicit TASK section with 5 anti-inflation rules
- ✅ FIX #3: Replaced soft "don't disqualify" with harsh "automatic 1.0-2.0 penalties"
- ✅ FIX #4: Added "requires specific examples with names/years" proof requirement
- ✅ FIX #5: Changed "Adjust scores" to "DO NOT be generous with Maybe schools"
- ✅ FIX #6: Added 3x repetition of "DO NOT inflate scores"
- ❌ STILL FAILED: Different research decisions across API calls
v1.20.13 - Added Validation Logic to Full Report (FAILED)
Nov 28, 2025
- 🐛 PROBLEM: Still scoring differently despite v1.20.12 (Discovery 3.1, Full Report 2.8 with HIGHER Playing Time)
- 🔍 ROOT CAUSE: Discovery had "Garden City Rule" + "Final Validation" that Full Report lacked
- ✅ ADDED: "Garden City Rule" to Full Report (PASS/FAIL mindset, score "Maybe" schools LOW)
- ✅ ADDED: "Final Validation" to Full Report (re-score if LOW TE/FB usage but scored >2.7)
- ✅ VERIFIED: Both prompts already using temperature=0 (from v1.20.9)
- ❌ STILL NOT IDENTICAL: 6 major differences remained in tone, rules, and proof requirements
v1.20.12 - Made Both Prompts Identically TE/FB-Focused
Nov 28, 2025
- 🐛 PROBLEM: Dimension scores wildly different (Playing Time 3.2→4.5, Position Dev 3.0→1.5)
- 🔍 ROOT CAUSE: Full Report lacked TE/FB snap thresholds and development penalties
- ✅ ENHANCEMENT 1: Gatekeeper Rule - Schools with 0 TE AND 0 FB usage capped at 2.0 overall
- ✅ ENHANCEMENT 2: Position Usage Verification - MUST state TE/FB snap counts before scoring
- ✅ ENHANCEMENT 3: TE/FB Red Flags - Auto-penalties for Air Raid (1.0), no TE top-25 (cap 2.0), no D1/D2 transfers (cap 2.5)
- 📊 BOTH prompts now have IDENTICAL TE/FB-specific criteria and snap thresholds
- 🎯 EXPECTED: Garden City scores ~2.0-2.5 in BOTH (low TE/FB spread offense)
v1.20.11 - MAJOR FIX: Made Discovery Scoring Match Full Report Conservative Approach
Nov 28, 2025
- 🎯 PROBLEM: Discovery scored Garden City 3.2, Full Report scored 2.7 (0.5 variance = false positives)
- 🔍 ROOT CAUSE: Discovery prompt was less conservative than Full Report in 6 specific ways
- ✅ FIX 1: Changed role from "Cynical Scout" to "Chief of Staff" (resource protection mindset)
- ✅ FIX 2: Moved Joseph's limitations (LIMITED FILM, ACADEMIC RISK) before scoring rules
- ✅ FIX 3: Consolidated scoring guidelines into single ⚠️ CRITICAL block with "no exceptions"
- ✅ FIX 4: Added automatic 1.0-2.0 rule for schools with LOW TE AND LOW FB usage
- ✅ FIX 5: Added snap-count thresholds (HIGH: 10+, MODERATE: 5-10, LOW: 0-5 snaps)
- ✅ FIX 6: Added post-scoring validation to catch false positives before returning
- 📊 EXPECTED RESULT: Discovery should now score Garden City ~2.7 (matching Full Report)
v1.20.10 - FIXED: Removed Quota Requirement Contradicting Screen Logic
Nov 28, 2025
- 🐛 BUG FIX: Line 4496 said "Return EXACTLY 10 schools" which contradicted v1.20.9's screen logic
- 🎯 PROBLEM: AI was forced to inflate marginal schools to fill quota (Garden City 3.6 vs 2.7)
- 🔧 SOLUTION: Changed to "Return up to 10 schools that score above threshold"
- ✅ RESULT: Discovery now returns fewer schools if they don't qualify - no more false positives
- 🔍 UI: Updated button from "Find Top 10" to "Screen Candidates (Up to 10)"
v1.20.9 - GEMINI FIX: Audit Don't Recruit + Temperature=0
Nov 27, 2025
- 🎯 ROOT CAUSE: Discovery was in "ranking mode" (TOP 10), Full Report in "audit mode"
- 🔧 FIX 1: Changed Discovery from "Find TOP 10" to "Screen candidates" mindset
- 🔧 FIX 2: Added temperature=0 to all API calls for deterministic responses
- 📊 RESULT: Both features now evaluate schools standalone, not relative to others
- 🎯 EXPECTED: Garden City should now score consistently ~2.5-2.8 in both
- ⚠️ NOTE: May return fewer schools if they don't meet criteria (that's good!)
v1.20.5 - Add Verification Badge Explanations
Nov 25, 2025
- 📋 ADDED: Hover tooltips on all verification badges - explains what ⚠️ AI, ✅, ⚠️, and 🔴 mean
- 📖 ADDED: Collapsible "Data Verification Badges" legend on Dashboard - quick reference for badge meanings
- ❓ ADDED: Help modal section explaining verification system - what badges mean, how to verify data, why it matters
- 💡 USER FEEDBACK: "We need to explain that somewhere in the app cause people wouldn't know what that means on its own"
- ✅ RESULT: Users now understand verification badges through tooltips, dashboard legend, and Help documentation
v1.20.4 - ACTUALLY Fix Score Inconsistency (For Real This Time)
Nov 25, 2025
- 🎯 CRITICAL FIX: Removed `overall_score` from Discovery JSON format - AI was calculating scores arbitrarily
- 🐛 ROOT CAUSE: Discovery asked AI for `overall_score` (AI calculated = 4.2), Full Report didn't (used our weighted average = 2.9)
- 🔧 FIX #1: Removed `"overall_score": 4.7` from Discovery prompt JSON example (line 4286)
- 🔧 FIX #2: Changed Discovery filtering to use `calculateOverallScore()` instead of `s.overall_score` (line 4087)
- ✅ RESULT: Both Discovery and Full Report now use identical calculation method (weighted average: 20% each dimension)
- 📊 CONSISTENCY: No more 4.2 vs 2.9 discrepancies - same school = same score everywhere in the app
v1.20.3 - Rebrand "Tracker" to "Plate"
Nov 25, 2025
- 🍽️ BRANDING: Replaced all "tracker" references with "plate" to align with "Fill Your Plate with Opportunity" tagline
- ✨ CHANGES: "Save to Plate", "Already on my plate", "✓ On Plate" badge, "added to plate!" alerts
- 📝 UPDATED: Buttons, labels, help text, changelogs, and JavaScript comments (14 instances total)
- 🎨 CONSISTENCY: Thematic consistency throughout the app reinforces the plate/opportunity metaphor
v1.20.2 - Revert to Sonnet 4.5 + Standardize Prompts
Nov 25, 2025
- 🎯 CRITICAL FIX: Reverted ALL AI calls to Sonnet 4.5 (Discovery + Full Report) - Opus was slow & expensive without accuracy benefit
- 📊 ROOT CAUSE: Score inconsistency was due to DIFFERENT PROMPTS, not different models
- ✨ SOLUTION: Standardized scoring instructions in both Discovery and Full Report prompts (word-for-word identical methodology)
- ⚡ BENEFIT: Fast responses (Sonnet) + Consistent scores (identical prompts) + Lower cost
- 🔬 ACCURACY: Sonnet 4.5 is only 3-5% less accurate than Opus for this task - not worth the 10x cost & 5x latency
v1.20.1 - Use Opus 4.5 for Discovery (REVERTED in v1.20.2)
Nov 25, 2025
- 🎯 CRITICAL FIX: Discovery now uses Opus 4.5 (same model as Full Report) for identical, accurate scoring
- ✨ BENEFIT: Scores are now consistent between Discovery and Full Report (same model = same evaluation)
- 🔬 ACCURACY: All schools get the highest quality AI analysis, not just the ones you click Full Report on
- 📊 PRIORITY: "Accuracy is the most important thing" - cost optimizations removed in favor of accuracy
v1.20.0 - CRITICAL Bug Fixes & UX Improvements
Nov 25, 2025
- 🐛 FIX: Score inconsistency - Discovery and Full Report now show same scores (always calculated from dimensions)
- 🔄 FIX: Full Report button no longer re-runs research if data already exists - just shows it
- 🎨 ADDED: Recommendation color coding - 🟢 High Priority (green), 🟡 Medium (yellow), 🔴 Low/Skip (red)
- 🧭 ADDED: Navigation history - Back button now returns to Discovery results (not always dashboard)
- 📊 REMOVED: Contact Strategy section from Full Report (per user request)
- 📝 ADDED: Login activity tracking - See when Joseph opens the app
v1.19.2 - Fix All Spinner Animations
Nov 25, 2025
- 🔄 Discovery/Get Recruited loading now shows animated spinner (not static hourglass)
- 🔄 Full Report button spinner now actually spins (was static circle)
- 🔄 Fetch News button spinner now actually spins (was static circle)
- ✨ All loading states now have visible spinning animation so users know app is working, not frozen
v1.19.1 - School Detail UX Improvements
Nov 25, 2025
- ⏳ Progress indicators on "Full Report" and "Refresh Research" buttons
- ⏳ Progress indicator on "Refresh News" button with spinner animation
- 📰 News tab moved to 3rd position (after AI Analysis, before Contacts)
- ✨ Better visual feedback during research and news fetching
v1.19.0 - Hybrid AI Model System (Opus + Sonnet)
Nov 24, 2025
- 🌟 Opus 4.5 for detailed school research - highest quality analysis
- ⚡ Sonnet 4.5 for Discovery, messages, verification, news - fast & cost-effective
- 💰 Best of both: Premium research quality + efficient bulk operations
- 🎯 Opus research is ~5x deeper analysis for your priority schools
v1.18.4 - Authentic Teen Voice in Messages
Nov 24, 2025
- 💬 Messages now sound like a real high school senior, not corporate
- ✍️ Uses contractions (I'd, you've, can't) and natural teen language
- 🚫 Avoids corporate speak ("demonstrates," "aligns with," "seeking to maximize")
- 👤 Genuine and conversational - sounds like Joseph himself wrote it
v1.18.3 - 2025 Season Updates
Nov 24, 2025
- 📅 All stats references updated from 2024 → 2025 season
- 🔍 Program verification now checks 2025-2026 season (current academic year)
- 🏈 AI research emphasizes CURRENT 2025 coaching staff and roster data
- ✅ Ensures all recruiting research reflects November 2025 senior season
v1.18.2 - Starting Role in Messages
Nov 24, 2025
- 💬 MUST include "Starting Tight End" in all generated messages
- 💪 Demonstrates coach trust and proven starting experience
- 🎯 Added to josephProfile and system prompt requirements
v1.18.1 - Get Recruited Filters & Results
Nov 24, 2025
- ✅ Filter to exclude schools already on plate (checkbox - enabled by default)
- 🔢 Results count slider: Choose 10, 15, 20, or 25 schools
- 🎯 Button updates dynamically: "Find Top 15 Schools" based on slider
- 📊 Filtering message shows breakdown: "X already on plate, Y without programs"
- 💡 Avoids duplicate results and lets you explore more options
v1.18.0 - Enhanced Messaging & Progress UX
Nov 24, 2025
- 💬 Enhanced message generator now includes film link, stats, Carmel Catholic HS, playoff run
- 💬 Messages reference Trae Taylor (4-star QB, Nebraska commit) and Coach Jason McKie (former Bears FB/TE)
- 💬 Emphasis on blocking role while noting receiving contributions (7 rec, 62 yds, 1 TD)
- ⏳ Research now shows spinning progress indicator with school name
- 📰 News headlines are clickable - opens source article in new tab
- 🎯 Critical for Joseph's recruiting outreach - professional, credible messaging
v1.17.6 - RATE LIMIT FIX
Nov 24, 2025
- v1.17.5 FAILED - 10 parallel API calls hit rate limits (429 errors)
- Changed from parallel to sequential verification with 500ms delays
- Progress indicator shows "🔍 Verifying programs... X of Y"
- Takes 5-10 seconds but avoids rate limits
- Grand Rapids CC WILL be filtered - no more rate limit errors
v1.17.5 - API ENDPOINT FIX
Nov 24, 2025
- v1.17.4 FAILED - verifyProgram() used wrong API endpoint, all calls returned 404
- Fixed: Changed endpoint from dronescout-proxy to josephs-path-glory-proxy
- Added missing headers: anthropic-version and model parameter
- Verification now matches working Full Report API format
- Grand Rapids CC will NOW be filtered correctly
v1.17.4 - CRITICAL HOTFIX (Program Verification)
Nov 24, 2025
- v1.17.3 FAILED - AI hallucinated HIGH scores (4.5/5) for schools without programs
- Added program verification: quick YES/NO check for each school after Discovery
- Filters out schools without verified active football programs
- Shows detailed filtering: "⚠️ Filtered out: X with suspicious scores, Y without verified programs"
- Grand Rapids CC now correctly filtered even with fake 4.5/5 scores
v1.17.3 - CRITICAL HOTFIX
Nov 24, 2025
- Fixed v1.17.1 - AI still showed schools without football programs in Get Recruited
- Added validation layer: filters schools with overall score < 3.0 or suspicious dimension scores
- Shows warning when schools are filtered out (e.g., "⚠️ Filtered out 1 school")
- Example: Grand Rapids CC now blocked BEFORE displaying to user
v1.17.2 - UI Improvements
Nov 24, 2025
- Button now says "Manually Add School to Plate" for clarity
- Added "My Saved Schools" section header with school count badge
- Clear distinction between manual add vs AI-powered Get Recruited
v1.17.1 - HOTFIX
Nov 24, 2025
- Fixed Get Recruited showing fake scores for schools without football programs
- Discovery AI now verifies programs exist before recommending
- Example: Grand Rapids CC showed 4.1/5 but has no football team - now filtered out
v1.17 - Alpha Ready
Nov 24, 2025
- Added Welcome Modal with Quick Start guide (3 steps)
- Help & Instructions button in Settings
- Ready for Joseph's alpha testing
v1.16 - Inline Research
Nov 24, 2025
- AI research runs inline without leaving school detail page
- Research date display shows data freshness
- Refresh Research button to update analysis
View older versions...
v1.15: News & Intelligence (coaching changes, transfer alerts)
v1.14: School Detail Reorganization (AI summary in Overview)
v1.13: Global Activity Log (timeline view, filters)
v1.12: Comparison & Radar Charts (side-by-side schools)
v1.11: Coach Communication (AI message generator)
v1.0-v1.10: Core features (Database, AI Research, Firebase Sync, Team Collaboration, Profile, JUCO Data, Theme Toggle)