# Strategy × Timeframe audit — first deliverable (2026-09-21, pre-market build 08:3x–09:0x ET)

Scope: James's brief "track today's names so we can answer WHAT strategy / WHAT timeframe / WHEN / WHAT direction /
DID I TRADE IT / WHAT HAPPENED NEXT". Universe today: HPE DELL SMCI TJX ALAB MARA TSLA TXN FLEX UMAC.
Live page: https://smb.a1prospect.com/strategy-matrix/ · source `Z:\trading\dev\strategy-matrix\`.

## 1. Which strategy / timeframe combinations we already support (engines that emit STATE)

| engine (file) | frames it is defined on | what it emits | status |
|---|---|---|---|
| `playbook-lens/engines.py::sigma_engine` (Adaptive Mean + σ, v5.3 port, parity-passed daily) | **timeframe-parameterized**: `<30m` → basis = session VWAP (execution frame); `30m..intraday` → basis = true 5-DMA (5×bars/day); `D` → 20-DMA of closes ±2σ | per-bar regime (COMPRESSION / BALANCE / EXPANSION UP-DN / OVEREXTENDED UP-DN), COIL/BASE, BW %ile; events SIGMA_COMPRESSION_START/END, VWAP_PULLIN_ARMED/VOID, LONG/SHORT_ARMED, PDH_BREAK, CONT_LONG/SHORT, EXPANSION_UP/DN, TRADE_ENTRY/EXIT/REFUSED (kinds VWAP L, CONT L/S, PULLBK S, BACKSIDE S, EXP L…), REGIME_CHANGE | production (Lens) |
| `playbook-lens/engines.py::am_atr_engine` (AM ATR v4.22-CL parity surface, rules = Pine v4.23-CL) | any intraday frame uses the **5-DMA** basis; daily uses **20-DMA** | ATR_RECLAIM_UP/DN/FILTERED, BRK_ARMED/CANCELLED, PLAY_OPEN/OPEN_BRK/FAILED_AT_OPEN, T1, T2_TRAIL, PLAY_EXIT/STOPPED/VOID, FADE_*, BAND_TAG/EXT/DTR/VWAP_EXT markers; per-bar play dir, tier, extState, trail | production (Lens) |
| TradingView `fl_engine_V2` (SMB_FASH_EVT_Office_Warriors__.25SD_1.3SD_Stop = user james; SMB_FASH_EVT_Cursor_ = user jim) | **2m** (`"tf":"2"`, session 0930-1600) | ENTRY (FL2_BandPullback, LONG/SHORT, entry, stop_1_atr, target_1_5_atr) / EXIT (FL2_TP1_Band3, Trend Break : 4 EMA, Stopped Out => Broke Lower/Upper Band) | LIVE — webhooks land in a2 PROD `alerts_tradingview.indicator_alerts` within minutes |
| MTF Compression→Expansion v2.1 (Pine, `mtf-compression-expansion`) + preview port in `live-lens-preview/build_preview.py` | canonical **1m** engine (wall-clock windows) | IDLE / COMPRESSION / RELEASE / DECELERATION / FLIP_WATCH … (the `state/dir/bw/bucket/traj` block in the preview payload) | RESEARCH — Tests A–D / Gate 1 still owner-run; NOT used as a state source here (roadmap rule) |
| `playbook-lens-staging/entry_review.py` library classifiers | evaluated **at a fill** on the 2m (exec) + 30m + D states | ON_BOOK / ENGINE / LAB_EXTENSION / OFF_BOOK per fill, families: Earnings D1, Backside, VWAP Pull-In, Trend Trade, ORB, Back-Through-Open, Gap-Give-and-Go, Day-2, σ engine, AM ATR | production (staging Lens) — no live state, only at-fill |

### 1b. Secondary engines on Z: (read-only sweep, 2026-09-21 09:2x) — what they are, which TF they pin, whether they run

| project | family (name in code) | canonical TF (pinning evidence) | states / codes | dir | output · schedule | status |
|---|---|---|---|---|---|---|
| `gapma` | GAPMA gap vs 5-DMA scanner (`★ Better trade` = retraced to 20-DMA within 10 sessions) | **Daily** bars + premarket open print (`gap_scan.pl`: `sma(closes,5)`, `gap_pct = (open − yest close)/yest close`) | list membership + flags, no lifecycle | both (longs.html / shorts.html) | `Z:\gt_dev\gapma\data\{longs,shorts}.json` → /gapma/; **schtasks** 19:45 / 04:05 / 08:00 / 09:35 | **production** — used in the matrix from v0.3 as HTF context (`gapma-gap-vs-5dma`, state SETUP) |
| `multi-tf-stack` (output only, producer not found on Z:) | market-environment stack grade (futures alignment, ETF prints, breadth, narrow leadership) | one snapshot/day frozen **09:30 ET** | `verdict` LEANING RISK ON/OFF, `grade` 0–100, `confidence` | market-wide | `Z:\gt_dev\multi-tf-stack\premarket_freeze\<date>.json`, daily unbroken to 09-18 | production — shown in the matrix header when today's file exists |
| `j3p` (JAMES THREE PLAYS v2.6.9, Pine) | `vwap_pull_in` (L), `vwap_pull_in_short` (S), `backside_scalp` (S), `backside_long` (L); `trend_trade` env-only | **intraday ≤15 min, RTH** (`max_tf_min` input, `tf_ok`), payload carries `"tf"`; tuned on 2m/5m | `evt`: entry / arm / no_chase / trend_break / pb_held / opp_exit | both | `alert()` JSON → webhook → **AMS ingest** (landing table being confirmed) | **production indicator** — candidate second primary 2m/5m feed once the AMS table is confirmed live |
| `alpha-compression-break` (ACB v1.6.7) + `acb_live.pine` (v1.5.5, superseded) | Alpha Compression Break → `vcet_compression_break` | **chart-TF driven — ambiguous** (Day/Swing/Invest presets, `comp_minutes` ≤390, RTH gate) | `evt` entry / partial / exit; labels ARMED, REV, HIT STOP, TARGET, FLIP ARM | both | `alert()` JSON → AMS ingest | production indicator (TF must be stored per alert from `"tf"`) |
| `swing-playbook-engine` | Python port of Adaptive Mean+σ v5.2; Play 1 = VWAP L | **30m RTH** (`resample(df1m, tf_min, session="RTH")`, `*_30m.csv`) | COMPRESSION/EXPANSION/BALANCE/COIL/BASE; exits STOP/TARGET/TIME/PREEMPT/REVERSE/WAVE5 | long-biased | CSV in `out/`; no schedule | research, **blocked** (Play-1 leg filter miss) — superseded by engines.py v5.3 port |
| `projected-5dma-environment` | Kaufman MA-turn anticipation over 5/20/50-DMA | **Daily** (`projected_ma.py`, `kti price daily --days 260`) | WATCH/PLAN/TRIGGERED/PARTICIPATE/REDUCE RISK/AVOID…; phases none→anticipation→trigger→participate | both | CSV/JSON `_out/`; no schedule | research, dormant (06-11) |
| `alpha-setup-playbook` (Chiphazzard) | 5DMA · MA1 · MA2 · 2D AVWAP sequence | **chart-TF — ambiguous** (`MAs: days × 390 / chart_minutes`); Pine lives in a1p_journal | S1–S5 / L1–L5 watch steps | both | spec markdown only | research / spec |
| `vwap-strategy-selection-grid` | permission grid over classify_v2 / regime_classifier_v2 / VCET | symbol × as-of, inputs entry-time intraday — **ambiguous** | STAND_DOWN / BLOCKED / WITH_GRID / … ; regimes TREND_UP…CHOP | both | `out/*.csv`; SQL views not executed | research (06-08) |
| `vwap-band-entry-surface` (`cmt-…` retired) | chase-extended reclassification study | needs 1-min backfill — **ambiguous by its own admission** | none | both | HTML/MD | research complete, awaiting approval |
| `tv_alerts_realtime` / `tv_alerts_analytics` | Live Terminal (Fashionably Late) over `alerts_tradingview.indicator_alerts` | **TF never persisted in the row** (mixes `atr_session`/`ma_50m` with daily slopes) — the matrix stores `timeframe` from the row (`'2'`) | ENTRY / EXIT; grades A++/A/B/C client-side | both | Postgres (a2 prod), browser long-poll | production web app |
| `readiness-study` | 8 long-side entry models incl. Fashionably Late V2 (SMCI vs MRVL) | **2m RTH** karb | R1–R8 | long | one-shot HTML/JSON | done (06-12) |
| `backside-classifier-audit` (`backside_lifecycle.py`) | backside-long setup/outcome classifier | intraday, session-anchored; **bar size not pinned in the module** | candidate → anticipation → triggered → partial_failed_before_vwap / completed_to_vwap | long | JSON/CSV/HTML; no schedule | research, promotion-ready |
| `live-lens-preview` | v2.1 lifecycle port + `e1` canonical slice | 1m → 1m/2m/5m (`default_tf` 2m) | v2.1 IDLE/COMPRESSION/RELEASE/DECELERATION…; e1 status | both | per-symbol JSON, **every 3 min RTH on the HOME box** | production, temporary (also today's only fills source) |
| `execution-hud` | presentation layer, no engine | n/a | Key Level → Compression → Signal → Early Trigger → Expansion → Management | n/a | none (design audit only) | no code |
| `ams` | platform (ingest / API / auth), not a strategy | n/a | n/a | n/a | PHP under /gt_dev/ams/, `*_dev` schemas | production, phased |

Implications for the matrix: (a) GAPMA and the multi-tf-stack freeze are the two scheduled HTF context producers that already run — wired in v0.3.
(b) J3P is James's own "three plays" engine on the execution frame; if its AMS landing table is live it becomes the second primary 2m/5m
feed next to FL2 (stored with the `"tf"` the alert carries — never assumed). (c) ACB's TF is chart-driven and must be taken from the
alert payload per row. (d) Nothing else emits live state; the Lens engine remains the only state source for D / 30m / 2m.

## 2. Canonical timeframe per strategy (what the matrix stores as `strategy_tf`)

| tier | strategy_id | strategy TF | evidence |
|---|---|---|---|
| HTF context | `sigma-daily-20dma` | **D** | `make_frame(daily=True)`: mode `daily20`, L=20, BW_LOOK=20 (engines.py L182-190) |
| HTF context | `am-atr-daily` | **D** | `L = 20 if F['daily'] else 5*BPD` (engines.py L871) |
| intermediate | `sigma-30m-5dma` ("5DMA compression / release") | **30m** | `mode = 'vwap' if tf_min < 30 else 'dma5'`, L = 5×13 bars (engines.py L191-194); STRAT_TF='30m' in entry_review.py L68 |
| intermediate | `am-atr-30m` | **30m** | same frame policy; `FRAME_30M` is the engine default |
| execution | `tv-fl2-fash-late` (Fashionably Late FL2) | **2m** | every payload `"tf":"2"`; TV alerts resolution 2 |
| execution | `sigma-2m-vwap` (VWAP pull-in / CONT / PULLBK family) | **2m** | EXEC_TF='2m' (entry_review.py L67); `<30m` → session-VWAP basis |
| research | `am-atr-2m` | 2m | AM ATR run on the exec frame; kept for research only |

Chart TF is stored separately (`chart_tf` column) and is never used to infer strategy TF.

## 3. Strategies whose timeframe definition is ambiguous / undefined

- **Range Break Day 1 / Day 2** — no canonical written definition on Z: (only `Range_Break_Strategy.pine` rb_engine_v2, no Day-1/2 concept). Not classifiable; reported, not guessed.
- **Earnings Day 1 / Day 2** — library rule is a 2m VWAP-reclaim trigger in the first hour; the "Day 1/2" part is a **daily catalyst calendar** attribute, not a timeframe. Needs a catalyst source reachable from DOMINO (TV scanner rate-limited today; InfoCenter unreachable).
- **"20DMA state" / "5DMA release"** as James phrases them are not separate engines: the 20-DMA state is the daily σ + AM ATR daily play; the 5DMA release is the 30m σ frame's COMPRESSION_END → EXPANSION. Both already emitted; the matrix names them by their engine + frame.
- **SIGMA_LONG_ARMED / SHORT_ARMED** (strength-machine arms) have no disarm record in the engine; the matrix shows them with their age (e.g. SMCI D SHORT armed since 09-16). Expiry rule = James's call.
- **50-minute MA** definition (50×1m vs 25×2m vs platform) remains an open question from the SIGNAL→MANAGE workstream; not used here.
- **Library families without an engine** (Hitchhiker / Trend Pullback, Backside / Rubber Band, ORB, Back-Through-Open, Gap Give and Go, Day-2 meta) have a documented *window* (e.g. ORB 09:45–11:45) but no state machine; they can only be scored at a fill → RESEARCH MATCHES.

## 4. Today's actual executions (pulled 08:33–09:05 ET)

- **No fills for 2026-09-21 exist in any reachable source** (pre-market): a2 `accountability.v_recent_fills` / `trades_closed` (stale since 2026-05-11 anyway), prod `smb.fills` (last 2026-04-10), `alerts_tradingview.fills` (last 2026-04-30), live-lens-preview payloads (dated 2026-09-18), v3 fixtures (latest 2026-09-18).
- kti `trades date` cannot be called from DOMINO (InfoCenter socket blocked). Today's fills will arrive via the HOME box: live-lens-preview refresh every 3 min from 09:25 (`Z:\gt_dev\live-lens-preview\data\<SYM>.json`, only for its seed set + the symbols James actually orders), and the v3 EOD job at ~12:04 (`a2:~/trading/dev/execution-scorecard-v3/fixtures/`). `build_matrix.load_fills` reads both; every symbol is **NOT TRADED** until one of them lands, and the row says which source was checked.

## 5. Today's matrix — see the live page (states as of the 09-18 close + newDay-decided records for today)

Rows with live state at build time (D / 30m; 2m resets at the open):
- HPE — 30m AM-ATR LONG in play (09-18 09:30 @ 61.25 stop 55.02). Fri FL2 LONG 09:44 → exit 10:02 (Trend Break).
- DELL — 30m AM-ATR LONG T2-trail; 30m 5DMA σ LONG armed; 30m σ PULLBK **SHORT** trade open (09-18 14:00 @ 574.20 stop 581.08) → conflicting lanes shown, not collapsed.
- SMCI — D AM-ATR LONG in play (09-18 @ 39.09 stop 35.36), D σ SETUP after compression, D σ LONG + SHORT armed; 30m AM-ATR LONG in play, 30m σ LONG armed → **2-TF LONG alignment (D, 30m)**.
- TJX — 30m AM-ATR LONG T1 hit; 30m σ BACKSIDE **SHORT** trade open (09-18 15:00 @ 128.16 stop 129.47); Fri FL2 SHORT 15:32 / 15:54 left open at the close.
- ALAB — D σ EXPANSION UP + D AM-ATR LONG in play (@ 303.25 stop 254.27); 30m AM-ATR LONG in play → **2-TF LONG (D, 30m)**; D σ SHORT armed (09-17) still on the books.
- MARA — D σ COMPRESSED (COIL since 09-11, BW %ile 0 → 10); D AM-ATR LONG in play (@ 13.24 stop 11.41) AND SHORT break-arm @ 10.77; 30m AM-ATR LONG in play, 30m σ VWAP L trade open (09-18 11:00 @ 12.71 stop 12.28), 30m σ SHORT armed → 2-TF LONG and 2-TF SHORT both flagged (descriptive).
- TSLA — 30m AM-ATR LONG in play (@ 365.64 stop 354.05); D σ COMPRESSED (BW %ile 0). Fri FL2 (jim/Cursor) SHORT 09:56 → TP1 10:20, exit 10:24.
- TXN — **only name with records dated today**: D + 30m + 2m AM-ATR reclaim-filtered → break-confirm ARMED LONG @ 268.28 (D) / 268.41 (30m), decided from Friday's close; D σ EXP L trade open (@ 266.64 stop 257.10); D/30m AM-ATR SHORT plays still in play (@ 258.14 / 260.71) → **3-TF LONG alignment + 2-TF SHORT**: the cleanest "watch the 268.3–268.4 break" setup on the list.
- FLEX — D + 30m σ COMPRESSED (COIL, BW %ile 0–19); D AM-ATR LONG armed @ 117.68; 30m σ SHORT armed. Nothing triggered.
- UMAC — D + 30m σ COMPRESSED (COIL, BW %ile 0); 30m AM-ATR LONG armed @ 24.44; 30m σ SHORT armed. Nothing triggered.

## 6. Which signals / states can already be reconstructed from existing data

- Every σ and AM ATR state on D / 30m / 2m for any symbol with karb bars (daily 90+ days; minutes ~10 trading days — enough for the 30m 5-DMA basis and the daily 20-DMA; the 30m BW %ile window is 260 bars so ranks are computed on the available ~117).
- FL2 entries/exits (2m) for both TV users since 2025-11 (85k rows in the prod table).
- Causal `known_at` for every record (newDay-decided at 00:00, close-decided at bar close) — the same rule as `lifecycle_sides.py`.
- Today's newDay-decided AM ATR records (reclaim / filtered / cancel / filtered-arm) BEFORE the first bar prints, from the prior close (flagged `provisional`).

## 7. What is missing

| gap | effect | fix |
|---|---|---|
| fills for today (InfoCenter blocked on DOMINO; HUD feeders stale since 05-11) | TRADED? column empty until the HOME box refresh / v3 EOD lands | run `live-lens-preview` (or `v3_live.sh`) on the HOME box today; add today's symbols to its seed set so payloads exist for all 10 |
| earnings / catalyst dates | HTF "Earnings D1/D2" row absent | TV scanner 429 today; retry, or read `pb-staging/fixtures/catalysts.json` convention |
| a1prospect accountability DB stale | no DB-backed fills/positions since May | separate ops item (feeder chain on DESKTOP-AKU8I62) |
| σ ARMED disarm rule | some ARMED lanes show multi-day age | James's call (expire after N sessions? on opposite expansion?) |
| MTF v2.1 (1m) states | not shown | by design until Gate 1 |
| research families | only at-fill | unchanged; entry_review.py runs in the Lens build |
| MFE/MAE per strategy×TF (review Q8) | needs fills + forward bars | computed by the Lens/v3 pipeline once fills exist; matrix will link |

## 8. Compact dashboard layout (as shipped)

`SYMBOL · DIR · [TF · strategy · state · when] lanes (D → 30m → 2m, live states only) · MULTI-TF ALIGNMENT · TRIGGERED TODAY · TRADED?`
Click a row → per-frame readout (regime, BW %ile, distance in σ, basis, AM ATR basis/dATR/play/tier, 5/20/50 DMA, VWAP, PDH/PDL),
the full lane table (first detected / armed / trigger / last event), fills + same-side / opposing overlay, research matches,
and the last 4 sessions of raw engine events. "Lens" opens `playbooks-dev/lens.html?ticker=SYM&date=…` (the trade object
must exist there; `build_lens_daily.py --specimen SYM DATE` builds it).

## 9. Event-ledger schema (as shipped, `out/ledger_<date>.csv`)

`date · timestamp · known_at · ticker · strategy_id · strategy_family · strategy_tf · chart_tf · direction · prior_state ·
new_state · event_code · kind · trigger_level · price · stop · bw_rank · session (RTH/ETH) · provisional · source_version · note`
One row per normalized state change; raw engine payload retained in `matrix_<date>.json`. First run: 3,120 rows over
09-11..09-21 for the 10 names (2,649 of them on the 2m σ frame — the noisy one, as expected).

## 10. No new engine

`build_matrix.py` imports `engines.py` by path and re-shapes its records; `statewalk.py` only maps engine codes to the
normalized vocabulary and applies expiry rules (documented in its docstring). The FL2 feed is James's own TV engine.
