Strategy Rules · what the engine ACTUALLY runs · ruleset-3.1.0 · generated from live code + your own campaigns

How to use this page: read each rule. If a rule is wrong or you want it different, tell Claude the rule ID — e.g. "change C-2 to 20 minutes" or "VPI-1's zone should be −1σ to 0" — and it changes in one place, versioned. PROXY means my stand-in math until your Pine exports arrive. NOT WIRED means the check exists on paper but can never fire yet — those are the honest blind spots.
Version history: the previous ruleset is frozen at ruleset-3.0.0 (archived copy), and every grade it produced is preserved in grades_snapshot_ruleset-3.0.0.csv. The grade-change diff between 3.0.0 and this version is at ruleset_3.1.0_diff.
How your fills become campaigns · Playbook detection — SHORT side · Playbook detection — LONG side · Playbooks with NO detector yet · How the winner is picked · Off-book checks · Daily state flags · Intraday cycle phase at entry · Grading · Glossary — phase definitions · Glossary — look-alike pairs

How your fills become campaigns

IDRule in plain wordsExact thresholdStatusReal example
G-1Every fill goes into a running position count. While the count isn't zero, fills belong to the same episode. Flat = episode over.net-position ledgerACTIVE
G-2Same symbol + same direction + re-entry within the window = same campaign (another attempt at the same idea), unless direction changed in between.window = 45 minACTIVE
G-3A cover bigger than your position is SPLIT: closing part + new opposite position flagged as a flip.over-cover splitACTIVE
G-4The whole day must match the broker's own P&L to the penny or the engine refuses to grade anything.reconcile tolerance $0.01ACTIVE

Playbook detection — SHORT side (classify_v3.py:66-101)

IDRule in plain wordsExact thresholdStatusReal example
BMR-1Backside / MA-Rollover short: MA cloud falling+0.35 scorePROXY — replace when your Pine arrivesNVDA-2026-07-01-C4
BMR-2…fast MA rolling over / intermediate MA rolling over+0.15 / +0.10PROXY — replace when your Pine arrives
BMR-3…prior 30-min low broken+0.20ACTIVE
BMR-4…entry came from an extension above VWAPz ≥ +1.0σ → +0.10ACTIVE
BMR-5Tier: fast-MA down alone = AGGRESSIVE; + low break = MODERATE. (CONSERVATIVE tier = 20-MA rollover + first-band break — not detected yet.)band1/band2 proxyPROXY — replace when your Pine arrives
VPS-1VWAP reject short: entry in the value zone−1.0σ ≤ z ≤ +1.5σ → +0.35ACTIVEGOOGL-2026-05-14-C2
VPS-2…price below the MA cloud+0.20PROXY — replace when your Pine arrives
VPS-3MISSING: the actual cross-DOWN trigger (close loses VWAP from above) and the 2-day AVWAP down-context from the July-2 pilot.not computedNOT WIRED YET — never fires

Playbook detection — LONG side (classify_v3.py:102-121)

IDRule in plain wordsExact thresholdStatusReal example
VPI-1VWAP Pull-In long: entry in the reclaim zone−1.5σ ≤ z ≤ +0.5σ → +0.30ACTIVE
VPI-2…MA cloud rising+0.25PROXY — replace when your Pine arrives
VPI-3…fast MA rising+0.15PROXY — replace when your Pine arrives
BSL-1Backside Long: prior flush to −2σ or deeperz ≤ −2.0 → +0.45ACTIVENVDA-2026-05-13-C1
BSL-2…reclaim high break+0.15ACTIVE
BSL-3MISSING: exit-ENTIRE-at-VWAP contract is never checked on your actual exits.NOT WIRED YET — never fires

Playbooks with NO detector yet

IDRule in plain wordsExact thresholdStatusReal example
RB-1Rubber Band / Back$ide Scalp (+2σ extension short, structural stop, exit entire at VWAP)NOT WIRED YET — never fires
VCET-1VCET Compression Break (COMPRESSED/ARMED/EXPANSION/CONFIRMED/FAILED, frozen rails)needs v3.2 PinePENDING — blocked on TV alert / Pine export
FL-1Fashionably Late V1/V2 — reachable ONLY via the webhook joinneeds TV alert re-pointPENDING — blocked on TV alert / Pine export
GAP-1Gap Setup — gap%% isn't computed yetNOT WIRED YET — never fires

How the winner is picked

IDRule in plain wordsExact thresholdStatusReal example
P-1Order of trust: indicator event (trade_id) → your manual tag → rule engine → best guess → Unclassified.5 levelsACTIVE
P-2Rule-engine match needs a score ≥ 0.60; between 0.35–0.60 it's labeled a GUESS (INFERRED); below 0.35 = Unclassified.0.60 / 0.35ACTIVE
P-3P&L is NEVER an input to classification.ACTIVE

Off-book checks (what can flag a campaign)

IDRule in plain wordsExact thresholdStatusReal example
C-1Entered before 09:30 ETOUTSIDE_TIME_WINDOWACTIVEQQQ-2026-07-07-C1
C-2Re-entered within minutes of a LOSING attempt, no new trigger on file≤ 15 min → REENTRY_WITHOUT_NEW_TRIGGERACTIVESOXL-2026-07-20-C2
C-3Long while the MA cloud is falling (guilty until innocent)COUNTERTREND_ENTRYPROXY — replace when your Pine arrivesSPCX-2026-07-01-C1
C-4Entered WITH direction into an extreme stretch|z| > 2.6σ → CHASED_EXTENSIONACTIVEQQQ-2026-07-07-C3
C-5Opposite-side fill ≤60s after going flat, no signal (orphan/hotkey)REVERSAL_WITHOUT_SIGNALACTIVEAMZN-2026-07-22-C3
C-6Missing market data at entry → UNDETERMINED, never guessedRULESET_DATA_UNAVAILABLEACTIVE
C-7Stop not defined / stop violated / oversized / invalid add / low RVOL / pre-confirmation entry — defined but the engine has no data source yet, so they NEVER fire. Your book status is currently blind to these.6 dormant checksNOT WIRED YET — never fires
C-8P-2 (new in ruleset-3.1.0): a CONTINUATION play (pull-in, rollover, compression break, FL) entered while the 2-min cycle read ROTATIONCONTINUATION_IN_ROTATIONACTIVE
C-9P-5 (new in ruleset-3.1.0) three-strikes: attempt 3+ on a name after two losing attempts needs a fresh signal — spacing re-entries out no longer dodges the 15-minute ruleflags as REENTRY_WITHOUT_NEW_TRIGGERACTIVE
C-10Soft P-1 (new): entries in the 09:30–10:10 opening period (or premarket) get a WARNING chip — half size, A-setups only. Not an off-book flag.chip only, no grade changeACTIVE

Daily state flags (htf-state-1.1.0) — computed from the PRIOR close only

IDRule in plain wordsExact thresholdStatusReal example
H-15-DMA expanding up / down3-day slope beyond ±0.15%/dayACTIVE
H-220-DMA expanding up / down5-day slope beyond ±0.08%/dayACTIVE
H-3Backside of 5-DMA / 20-DMA = close BELOW it while it's fallingACTIVE
H-45/20 compressed (coiled)gap < 0.5% of priceACTIVE
H-5Extended above 5-DMA> 2× avg daily rangeACTIVE
HC-1Combos: aligned expansion up/down (5 AND 20 agree) · true backside (below both, both falling) · pullback-in-uptrend (5 down inside rising 20) · pop-in-downtrend · coiled frontsideAND of the flags aboveACTIVE

Intraday cycle phase at entry (intraday-phase-0.9.1-proxy) — 2-min bars, entry-time only

IDRule in plain wordsExact thresholdStatusReal example
IP-1COMPRESSION: the last 30 minutes' range is tight vs the sessionrange < 60% of session median (15×2-min window)PROXY — replace when your Pine arrivesSPCX-2026-07-01-C1
IP-2ROTATION: normal range, inside the bands — the defaultPROXY — replace when your Pine arrivesGOOGL-2026-05-14-C2
IP-3EARLY_EXPANSION: range just blew out OR a fresh break of the 30-min high/lowrange > 150% of median, or fresh breakPROXY — replace when your Pine arrivesNVDA-2026-07-01-C4
IP-4EXPANSION_CONFIRMED: the break HELD3 closes beyond the pre-break edgePROXY — replace when your Pine arrivesSPCX-2026-07-17-C9
IP-5TREND_UP / TREND_DOWN: extended past the band with both MAs agreeing|z| ≥ 1.0σ + fast&mid slopes same signPROXY — replace when your Pine arrivesAMZN-2026-07-01-C2
IP-6MEAN_REVERSION: stretched far and turning back toward VWAP|z| ≥ 2.0σ + 3 closes back toward VWAPPROXY — replace when your Pine arrivesNVDA-2026-07-07-C4

Grading (score_v3.py) — process only, P&L never counts

IDRule in plain wordsExact thresholdStatusReal example
S-1Setup 25 · Entry 20 · Risk 20 · Management 25 · Exit 10 — every point movement is written to a ledger you can read on any campaign.100 ptsACTIVE
S-2Each off-book reason costs 4 points.−4 eachACTIVE
S-3Ceilings: fully OFF-BOOK can't beat 68 (D range) · MIXED can't beat 82 (B−) · data-limited can't beat 79.caps 68/82/79ACTIVE
S-4Risk points are currently GIVEN unless there's contrary evidence, because no stop/size data source is wired — the grade says so via the data-limited cap.NOT WIRED YET — never fires
S-5P-6 exit contracts (partially wired in score-engine-3.1.0): X-1 trend rides — exiting while price is still on the right side of a still-sloping 20-EMA costs 3 pts ('left the trend early'). X-2 reversion plays (Backside Long / Rubber Band) — final exit past ±0.5σ beyond VWAP costs 3 pts ('contract is exit ENTIRE at the mean'). Still not wired: 30/50/20 ladder adherence, wave-5 cover.-3 pts each, exit-time state (management review only)ACTIVE

Glossary — phase definitions

IDRule in plain wordsExact thresholdStatusReal example
D-1COMPRESSION — volatility drying up: range abnormally tight AND shrinking; energy being stored; a break is coming, direction unknown. Play: position near the coil edge, tight stop.the springACTIVE
D-2EARLY EXPANSION — the first poke through the range; NOT yet proven; highest fake-break risk of any phase. Play: starter size or wait.the first pierceACTIVE
D-3EXPANSION CONFIRMED — the break HELD (closes stay beyond the old edge). Play: full-size with the move.held breakACTIVE
D-4TREND — the move became the state: extended with fast+intermediate MAs stacked and agreeing; pullbacks hold. Play: add at the EMAs, ride, exit on the objective line only.MAs agreeACTIVE
D-5ROTATION — normal-energy two-way chop around VWAP; no stored energy, no release coming. Play: quick scalps TOWARD VWAP — or nothing. Never continuation ideas.the washing machineACTIVE
D-6MEAN REVERSION — a stretch snapping back: beyond ±2σ and the last closes walking toward VWAP. Play: rubber-band fade, exit at VWAP.the snap-backACTIVE
D-7UNKNOWN — first ~40 min / premarket: not enough bars to read the cycle. Your 6-day tape: −$696 lives here.unreadable openACTIVE

Glossary — look-alike pairs (the tells)

IDRule in plain wordsExact thresholdStatusReal example
L-1Compression vs Rotation: both directionless — compression is abnormally tight and TIGHTENING (break coming); rotation is normal width going nowhere. Tell: is the range shrinking below normal, or just normal?ACTIVE
L-2Early vs Confirmed expansion: first pierce vs held break. Tell: 3 closes held beyond the edge? (Alpha Compression Break's no-first-pierce rule.)ACTIVE
L-3Expansion vs Trend: the event vs the state after. Tell: have the MAs caught up and stacked with the move?ACTIVE
L-4Trend vs Mean Reversion: both far from VWAP. Tell: last 3 closes — away from VWAP (trend) or toward it (reversion)?ACTIVE
L-5VWAP Pull-In vs Backside Long: buying a PAUSE in a healthy trend (dip to value zone, cloud rising) vs buying a RECOVERY after a −2σ flush (reclaim; exit ENTIRE at VWAP, no runner).ACTIVE
L-6MA-Rollover short vs VWAP reject short: the BIRTH of a downtrend (ride it) vs a rally back to VWAP getting REFUSED in a down context (fade it, done at the next band).ACTIVE
L-7Rubber Band vs MA-Rollover: shorting a +2σ STRETCH for the snap-back (exit entire at VWAP) vs shorting a BREAKDOWN for follow-through (trail it). Confusing them = giving back the snap, or cutting off your trend day (MU 7/24 AM).ACTIVE
L-8Continuation entry vs Chase: same trend, same side — pullback to the fast MA/band vs entry INTO the extension (flagged beyond 2.6σ). Winning a chase doesn't make it a continuation.ACTIVE

Sources: campaigns.py · classify_v3.py · htf_state.py · score_v3.py · taxonomy.py · thresholds imported live where possible. Examples link into your campaign review pages. Companion pages: Edge Lab · calendar.