InvestIQ — M0 Baseline & Freeze
דוח M0 — Baseline & Freeze
מדידה וניתוח של InvestIQ במצבו הקיים, כנקודת ייחוס מחייבת לבדיקת איכות, כיסוי, ביצועים ועלות בשלבי M1–M3. שלב מדידה בלבד — ללא שינוי קוד, Workflows, SystemSettings, AgentConfig, AgentProfile, תדירויות, LLM, Risk Gate, CIO/Decision, Trade/Execution או כל התנהגות אחרת.
תאריך: 9 בספטמבר 2026, 14:31חלון: 2026-08-23 עד 2026-09-02תוצאה: PASS WITH LIMITATIONS
M0-01 — Freeze Verification
PASS WITH CAVEAT| SystemSettings | Hash mismatch — market_regime operational data updates by workflows. No config field changes. |
| AgentConfig | Hash mismatch — 0 records updated after PRE-M0 (updated_date). All 19 config_locked + identity_locked. |
| AgentProfile | Hash mismatch — 0 records updated after PRE-M0. No configuration changes. |
| Code changes | Added routes for PRE-M0/Supplemental/M0 report pages only. No system behavior change. |
| No changes to | Workflows, SystemSettings config, AgentConfig, AgentProfile, frequencies, LLM, Risk Gate, CIO/Decision, Trade/Execution, Local Engine (not created). |
הערה: Hash mismatch on current-state entities is due to operational data updates (market_regime, last_run, decisions_count) by running workflows — not configuration changes. All AgentConfig records are config_locked and identity_locked.
M0-02 — LLM / Base44 Cost & Activity Baseline
PASS — Call Count Evidence| AnalysisOutput count (10 days) | 4,412 — primary LLM call proxy |
| Decision count (CIO LLM) | 546 |
| Estimated total LLM calls | ~4,958 |
| Per day (min / max / avg) | 5 (Sep 1) / 1,098 (Aug 26) / 441 |
| By agent type | scanner (2,204) · technical (1,105) · news (1,103) |
| AgentLog total | 23,250 (all with confidence — includes deterministic checks) |
| base44_credit_units | null — NOT verifiable from DB. Not converted to 0. |
הערה: Verified Credit Evidence: NONE. Call Count Evidence: AnalysisOutput + Decision proxy. Estimated/Unverified: ~4,958 LLM calls. Estimate is NOT presented as verified accounting.
M0-03 — Discovery Baseline
PASS| Total candidates | 52,387 (full paging export) |
| By region | US: 33,489 · IL: 18,898 |
| By market | NYSE: 24,931 · TASE: 18,898 · NASDAQ: 8,558 |
| By discovered_by (top 5) | Correlation/Allocation (5,221) · Fundamental (4,313) · CIO (4,071) · US Scanner (3,975) · Macro/Sentiment (3,698) |
| Unique symbols | 2,591 |
| Duplicate symbols | 1,544 (max: NVDA × 1,449) |
| Dedup rate | 2.3% (51,168 unique dedup_keys) |
| All candidates status | expired (52,387) |
| Discovery Latency | MISSING EVIDENCE — no first market occurrence timestamp (MarketData = MISSING EVIDENCE) |
M0-04 — False Positive / False Negative Baseline
NOT MEASURABLE| Pipeline | 52,387 candidates → 546 decisions (1.04%) → 96 trades (0.18%) → 16 executed (0.03%) |
| Decision actions | BUY: 474 · SELL: 71 · WAIT: 1 |
| Decision status | EXPIRED: 543 · CANCELLED: 3 |
| Risk Gate | 1,302 verdicts — ALL PASS (no rejections) |
| Shadow cycles | 411 — ALL PENDING (no T+7 outcomes yet) · ALL NONE mismatches |
| Trade status | expired: 49 · rejected: 31 · executed: 16 |
| FP Rate | NOT MEASURABLE — cannot define FP just because candidate didn't become trade |
| FN Rate | NOT MEASURABLE — no Ground Truth to prove undiscovered event |
הערה: All candidates expired — linkage to decisions unclear. All decisions expired/cancelled. All shadow cycles PENDING (not matured for T+7).
M0-05 — Intraday Coverage
PARTIAL| Weekdays 08-25 to 08-28 | 24h activity, 12-15 min gaps — excellent coverage |
| 08-29 (Saturday) | 1,007 entries, reduced hours — expected (weekend) |
| 08-30 (Sunday, IL) | 2,228 entries — moderate |
| 08-31 (Monday) | 427 entries — declining, longest gap 442 min (7.4h) |
| 09-01 (Tuesday) | 24 entries — very low, only 3 hours active |
| 09-02 (Wednesday) | 32 entries — very low, 6 hours active |
| Longest gap | 442.4 min (7.4 hours) on 2026-08-31 |
הערה: 08-31 to 09-02: significant activity reduction on weekdays — unexpected. May indicate reduced frequency, system issues, or quiet market. Weekend/market-closed not classified as coverage failure.
M0-06 — Quiet-Market / Slow-Trend Coverage
PARTIAL-MISSING EVIDENCE| System-observed quiet periods | Yes — candidate drops: -77% (08-29 Sat), -81% (08-31), -95% (09-01) |
| Market-confirmed quiet periods | MISSING EVIDENCE — no MarketData to confirm market behavior |
| Scanner pre-filter impact | Cannot be confirmed without MarketData price history |
הערה: System-observed quiet periods separated from market-confirmed quiet periods per instruction. MarketData = MISSING EVIDENCE prevents market-confirmed quiet-market baseline.
M0-07 — Protection Analysis Baseline
PARTIAL| AgentLog protection entries | 2 in 10 days (evaluateProtectionStatus: 1) |
| Alert protection alerts | 34 (stop_loss: 18, risk_breach: 16) |
| Message risk_warnings | 217 |
| Risk Gate entries | 7,424 (frequent deterministic checks) |
| LLM involvement | Cannot be determined from evidence |
| Deterministic vs LLM | Risk Gate (7,424) likely deterministic; evaluateProtectionStatus (1) likely LLM |
| Longest gap (protection) | Cannot be reliably measured (only 1-2 entries in 10 days) |
הערה: Very low protection evaluation frequency — known issue from Quality Rehabilitation Map. No-alert ≠ no-problem unless evidence exists.
M0-08 — Operational Baseline
PASS| Routine runs | 126 total — 122 success, 4 failed (96.8% success) |
| Total tests | 2,735 — 2,728 pass, 7 fail (99.7% pass) |
| Messages | 609 (info: 391, risk_warning: 217, daily_summary: 1) |
| Alerts | 50,934 (buy_signal: 47,869, sell_signal: 1,640, news: 1,391, stop_loss: 18, risk_breach: 16) |
| Retries | NOT MEASURABLE — no retry field in entities |
| Latency | NOT MEASURABLE — no start/end timestamps for workflow runs |
| Known anomalies | Declining activity 08-31 to 09-02; Hebrew asset_type values in Candidate; config hash mismatch alerts on 08-23 |
M0-09 — Dependency Inventory
PASS| Base44 Database | Critical — M1/M2/M3 — 5000-record query cap |
| Base44 Workflows | Critical — M1/M3 — credit consumption, fixed schedule |
| Base44 Backend Functions | Critical — M1/M2/M3 — serverless, cold start |
| Base44 InvokeLLM | Critical — M4 — credit consumption |
| Yahoo Finance API | Critical — M2 — no TASE coverage, rate limits |
| FRED API | Medium — M2 — US-only indicators |
| Base44 Auth | Critical — M1 — platform-managed |
| Base44 Realtime | Medium — M1 — platform-managed |
| Stripe | Medium — None — IL region |
| System Time | Critical — M1/M3 — UTC, no NTP |
הערה: 12 dependencies inventoried. No migration or replacement at this stage. Full details in m0_raw_metrics.json.
M0 Known Baseline Limitation — Historical MarketData
MarketData נשאר MISSING EVIDENCE. לא בוצע backfill, לא נמשך Yahoo Finance/FRED לשחזור היסטורי, לא נעשה שימוש בנתוני צד שלישי חדשים ל"תיקון" PRE-M0.
מדדי Quality שאינם ניתנים למדידה מלאה בגלל חוסר MarketData:
- • Discovery Latency — זמן מ-first market occurrence ועד Candidate creation
- • Market-confirmed quiet periods — אישור שירידת candidates נבעה משוק שקט
- • Scanner pre-filter impact — אישור ש-scanner דילג על analysis בתקופות low movement
- • Price-based quality metrics — דיוק מחירים, drift, slippage
שכבות נתונים — Raw Metrics לעומת Interpretation
A. Raw Baseline Metrics
מספרים, timestamps, counts, distributions ו-gaps כפי שנמדדו. קובץ: snapshots/pre-m0/m0_raw_metrics.json. ללא inference או מסקנות.
B. Baseline Interpretation
מסקנות לגבי Quality, Coverage, Cost ו-limitations. מוצגות בכל סעיף M0 ובטבלת Quality Baseline Summary. Inference אינו מעורבב בתוך Raw Metrics.
Quality Baseline Summary
| Metric | Baseline Value | Evidence Source | Confidence | Limitation |
|---|---|---|---|---|
| Total LLM calls (10 days) | ~4,958 | AnalysisOutput + Decision | MEDIUM | Proxy, not direct count |
| LLM calls/day (avg) | ~441 | AnalysisOutput by day | MEDIUM | Proxy |
| LLM calls/day (min/max) | 5 / 1,098 | AnalysisOutput | HIGH | Direct count |
| Base44 credit units | null | N/A | NOT MEASURABLE | Not verifiable from DB |
| Total candidates | 52,387 | Candidate snapshot | HIGH | Full export |
| Candidates/day (avg) | ~5,239 | Candidate by day | HIGH | Direct count |
| Unique symbols | 2,591 | Candidate snapshot | HIGH | Direct count |
| Duplicate rate | 2.3% | dedup_key | HIGH | Direct count |
| Discovery latency | MISSING | N/A | NOT MEASURABLE | No MarketData |
| FP rate | NOT MEASURABLE | N/A | NOT MEASURABLE | No Ground Truth |
| FN rate | NOT MEASURABLE | N/A | NOT MEASURABLE | No Ground Truth |
| Intraday coverage (weekdays) | 24h, 12-15 min | AgentLog | HIGH | Direct observation |
| Longest gap | 442 min (7.4h) | AgentLog Aug 31 | HIGH | Direct observation |
| Protection evaluations | 1-2 in 10 days | AgentLog | HIGH | Very low frequency |
| Risk Gate checks | 7,424 | AgentLog | HIGH | Direct count |
| Routine success rate | 96.8% | RoutineRunLog | HIGH | 126 runs |
| Trades executed | 16 | Trade entity | HIGH | Direct count |
| Market-confirmed quiet | MISSING | N/A | NOT MEASURABLE | No MarketData |
כלל: לא נעשה שימוש ב-HIGH כאשר הנתון מבוסס על inference ולא על evidence ישיר.
M0 Technical Result
PASS WITH LIMITATIONS
Passing: M0-02 (Call Count), M0-03 (Discovery), M0-08 (Operational), M0-09 (Dependencies)
Caveat: M0-01 (Freeze — hash mismatch due to operational data, no config changes)
Partial: M0-05 (Intraday — excellent weekdays, declining 08-31 to 09-02), M0-07 (Protection — very low frequency)
Missing Evidence: M0-06 (Market-confirmed quiet — no MarketData)
Not Measurable: M0-04 (FP/FN — no Ground Truth)
MISSING EVIDENCE של MarketData לבדו אינו גורם אוטומטי ל-FAIL, אך השפעתו על Quality Baseline מפורשת: Discovery Latency, Market-confirmed quiet periods, Scanner pre-filter impact ו-Price-based quality metrics אינם ניתנים למדידה מלאה. ההחלטה הסופית על G-M0 תתקבל רק לאחר review של הדוח והראיות.
STOP CONDITION
לאחר הפקת דוח M0 — עצירה. לא התחיל M1. לא נוצר Local Engine Skeleton. לא הותקן Windows Service. לא נוצר Event Client. לא שונה Base44. לא הופעלו Workflows. לא שונו Agents או frequencies. לא בוצע optimization. לא התחיל M2/M3.
דוח M0 המלא והראיות מוחזרים לעיון ואישור. ההחלטה על G-M0 תתקבל רק לאחר review.
InvestIQ M0 Baseline & Freeze Report | 9 בספטמבר 2026 | חלון: 2026-08-23 עד 2026-09-02 | 9 סעיפי M0 | MarketData: MISSING EVIDENCE | תוצאה: PASS WITH LIMITATIONS.
Raw metrics: snapshots/pre-m0/m0_raw_metrics.json · PRE-M0 manifest: snapshots/pre-m0/manifest.json