MIDAS 2.0

It is the exhaustion-aware sibling of anchored_vwap: where Anchored VWAP is a pure Location sensor (distance-in-σ, no triggers), MIDAS 2.0 adds the launch-anchored S/R curve and the accelerated topfinder/bottomfinder that the classic MIDAS methodology is actually known for.

Features

• MIDAS curve anchored at a manual time, daily open, or auto swing (last / low / high), re-seeded from the real pivot bar, not the lagging confirmation bar
• Volume-weighted σ bands, ATR bands, or Hybrid (average of the two) at two configurable multiples, shown as quiet value/stretch zones or classic lines
• Topfinder / bottomfinder accelerated curve with maturity-filtered, candidate-confirmed closed-form fits — live-honest From Fit history by default
• Qualified exhaustion read (no fit / developing / mature / late / passed) from cumulative-volume progress d/D, followed by a separate 3-part price-reaction check
• Bias-independent stretched-from-value and reclaim-after-stretch context markers, classified as trend-aligned or counter-bias; persistent markers are bar-close confirmed by default
• Backpainted / Confirmed anchor display mode — trade off a clean historical look against a live-honest anchor
• Manual Time remains inactive before the selected timestamp; if it predates loaded history, the first available bar becomes the visible start
• Episode-wide volume quality with an explicit Per-bar fallback / Equal weight policy instead of a current-bar-only status
• Auto EMA bias with a configurable hold-bars hysteresis to reduce whipsaw flips in choppy phases
• Compact / Detailed / Tooltip-only live badge with the complete measurement context retained on hover
• Optional Compact / Detailed dashboard — a 5-row decision view plus the full diagnostic Element | Value | Interpretation table
• Active / Full visual modes — the complete current anchor episode by default, or every historical episode for diagnostic review
• Light-theme dashboard, alerts (de-directionalised wording), optional debug log

Instrumente

Data Contract verdict: Exchange-only. The MIDAS curve is a cumulative volume-weighted average price — volume is not an add-on filter here, it is the weighting term in the curve’s own formula, and the same applies to the topfinder/bottomfinder fit and the Reclaim Quality Score’s volume component. On an instrument without real trade volume there is no valid MIDAS curve to compute; a script that produced one anyway would just be publishing a mislabeled close-price average.

• Valid: any symbol whose syminfo.volumetype reports base or quote — as a rule futures, exchange-listed stocks and crypto exchanges, plus broker feeds that pass real traded quantity through.
• Proxy: where that volume is a broker’s own share of the market (a CFD feed reporting base), the resulting levels sit where that broker’s volume traded, which need not be where the exchange’s did. Read them as approximate, not as the levels other participants watch.
• Invalid: any symbol reporting tick (a count of price updates) or n/a — there is nothing to weight.
• The script checks syminfo.volumetype at runtime (volumeIsReal). When it is not "base" or "quote", every volume-weighted computation is forced to na — no curve, no bands, no topfinder/bottomfinder, no context markers, no dashboard, no live badge. A single warning label (“benötigt echtes Handelsvolumen — aktuell: …”) appears on the last bar instead. There is no equal-weight or fallback substitute in this state — that would just recreate the same problem under a different name.

On a reference-market variant (documented only, not implemented): MIDAS is a price-level indicator — the curve is plotted directly on the chart as an S/R level, not a normalized oscillator. Borrowing a reference market’s volume (e.g. NYMEX:NG1! for a NatGas CFD) while still anchoring to the CFD’s own price would weight one feed’s price by another feed’s volume — two different markets contributing to a single number, which is not what volume-weighting is supposed to represent. Using the reference market’s own price instead would fix that mismatch, but at that point the chart would be showing NYMEX:NG1!’s price level on a CAPITALCOM:NATURALGAS chart, which runs into the settlement/back-adjustment mismatch documented in DATA_VALIDITY.md §9.1. A naive reference-market implementation is therefore not recommended for this indicator; the plain Exchange-only gate above is the safer choice.

Roles (design skill §1)

Role	What provides it
Location	normalized distance from the MIDAS curve (σ / ATR / hybrid deviation), band zone
Exhaustion (Momentum)	topfinder/bottomfinder d/D phase + fit stability
Trend	price side + MIDAS slope in the live read; bias colours markers and picks top vs bottom only
Trigger / Setup	stretched-from-value, MIDAS reclaim, and EXH n/3 price-reaction markers (context, not entries)
Quality / Debug	dashboard + log.info events

The bias is not a hard gate on the curve or the exhaustion read — it only colours markers and (for Manual/Daily anchors) picks topfinder vs bottomfinder. For Auto modes the TBF direction follows the anchor side: a swing-low anchor fits a topfinder (uptrend, fitted to higher lows), a swing-high anchor fits a bottomfinder (downtrend, fitted to lower highs). Auto Last Swing (default) anchors on whichever significant swing came last and takes its side automatically.

Visual language

• Orange line: launch-anchored MIDAS value; it is the dominant chart object
• Neutral inner fill: price is inside the normal value area (L1–U1)
• Amber outer fills: price is stretched from value (U1–U2 / L1–L2); amber encodes distance, not direction
• Purple line: the current topfinder/bottomfinder curve; stronger purple means the fitted move is further advanced. Active mode defaults to From Fit, so the curve begins only where the current fit became known; Retrospective is explicitly analysis-only
• Right-edge live context: Compact (default), Detailed, or Tooltip-only; the detailed view keeps three explicit state rows (TREND, LOCATION, EXHAUSTION) plus NEXT, which names the next observable trigger instead of repeating the state
• Amber dot: a stretch rejection at the outer band; context only
• Green/red R n/5: directional MIDAS reclaim after a prior stretch, with annotated quality
• Purple EXH n/3: an advanced projection plus at least two separate price-reaction clues; exhaustion risk is confirmed, but this is still not a reversal entry
• Orange A: the active episode’s anchor

The MIDAS curve

From the launch (anchor), the curve is the cumulative volume-weighted average of the source:

MIDAS(j) = Σ pᵢ·Vᵢ / Σ Vᵢ        (i from launch to j)

In an uptrend price tends to ride above a curve launched from a swing low (dynamic support); in a downtrend below one launched from a swing high (dynamic resistance). The σ bands use the cumulative volume-weighted variance Σp²V / ΣV − MIDAS²; ATR bands use a rolling ATR instead.

Topfinder / bottomfinder

When a trend accelerates away from its MIDAS curve, Levine’s topfinder/bottomfinder fits a steeper, decelerating curve that catches back up to price and estimates where the move ends — an accelerated exhaustion estimate, not a forecast. It re-weights the accumulation by wᵢ = 1 − dᵢ/D, where dᵢ is cumulative volume since launch and D is the projected total volume at which the move terminates:

TBF(j) = Σ pᵢ·Vᵢ·(1 − dᵢ/D) / Σ Vᵢ·(1 − dᵢ/D) = (A·D − B) / (C·D − E)

D is solved in closed form so the topfinder passes through the most recent same-direction swing after the launch (a pullback low for a topfinder, a pullback high for a bottomfinder):

D = (B − T·E) / (A − T·C)      with  A=ΣpV, B=ΣpVd, C=ΣV, E=ΣVd, T = fit price

The fit is accepted only when it still projects forward from the current bar (D greater than the volume accumulated so far) and the fit pivot occurred after the configured Minimum Fit Maturity (default 20% of D). Repeated fits compare their projected D values:

• provisional: first accepted fit
• stable: a repeated fit changed D by no more than the configured tolerance (default 25%)
• candidate 1/2: a strongly shifted candidate is held without replacing the active fit
• a second candidate in the same new D region confirms the change and replaces the active fit

In Backpainted mode, fit pivots between the real pivot origin and its later confirmation bar read from the reconstructed canonical episode statistics. They never read immutable Series values from the previous anchor episode, but no longer need to discard an otherwise valid early fit.

The live read reports progress d/D in these phases:

State	Meaning
no fit	no valid forward projection (range / non-accelerating leg)
developing	fitted, but below 50% of projected cumulative volume
mature	d/D ≥ 50% — projection is decision-relevant, but not late
late	d/D ≥ warn threshold (default 85 %) — advanced move, not a confirmed turn (LATE marker)
passed	d/D ≥ 1 — reported as recently passed, beyond fit, or stale depending on overrun and reaction

Re-fitting on each new same-direction pullback keeps the projection current. Stable small refits retain the current reaction; materially new confirmed fits reset only their own reaction while the episode remembers earlier reactions. The reaction score checks rejection candle, ATR-sized displacement, and return through the inner band. At least 2/3 creates EXH n/3. A passed fit without a current-fit reaction becomes stale at the configured threshold (default 150%).

Dashboard

When enabled, the default Compact dashboard mirrors the live context with Trend, Location, Exhaustion, and Next. Detailed restores the full diagnostic table below. Both modes are timeframe-aware — the header shows the bucket (Intraday < 4h / Swing 4h–<1D / Position ≥ 1D), and intraday uses a wider stretch bar with softer reversion language.

Row	Value	Interpretation
Location	normalized distance in σ / ATR / hybrid deviation (hdev)	at value / stretched up·down / overextended up·down
Band zone	inside / U1–U2 / above U2 …	at value / stretched / extreme
Topfinder/Bottomfinder	phase + d/D	fit quality, projection maturity, and whether price has reacted 2/3
Trend	above/below value + slope arrow	buyers / sellers since launch
Anchor	bars since anchor (+ drawn Xb back if Backpainted)	fresh (thin) / established
Volume	percentage of episode bars with reported volume	volume-weighted / mostly volume-weighted / mixed fallback / equal-weight fallback
Reclaim Q	last reclaim’s score, 0–5 + direction	weak / medium / high quality
Next	—	next observable condition, e.g. inner-band test/return, TBF EXH 2/3 reaction, or MIDAS reclaim/loss

NEXT is not a new signal. It translates the current state into the next condition worth observing; no condition means no setup yet.

Modes & inputs

Group	Key inputs
Anchor	mode (Manual Time / Daily Open / Auto Last Swing / Auto Swing Low / Auto Swing High), anchor display (Backpainted / Confirmed), manual time, anchor swing left/right (significant degree), source, volume weighting
Bands	show, display (Zones / Lines / Zones + Lines), mode (VW StDev / ATR / Hybrid), mult 1 & 2, ATR length
Topfinder/Bottomfinder	enable, exhaustion warn d/D, minimum fit maturity, stable-fit tolerance, TBF history (From Fit / Retrospective), stale threshold, fit swing left/right
Bias	Auto EMA / Up / Down, EMA length, Auto EMA min hold bars before flip
Display	Active/Full visual mode, bar-close-confirmed context markers, reclaim settings, Compact/Detailed/Tooltip-only live context, compact/detailed dashboard + position, active anchor, debug
Colors	MIDAS, TBF, stretch zone, bullish, bearish

Notes

• Auto-anchor modes re-anchor on the most recent significant swing (the Anchor Swing degree, default 20) — the origin of the current leg. A larger anchor degree keeps the curve on the current move rather than resetting on every minor wiggle (and avoids pinning it to the all-time extreme). The topfinder fits to the smaller Fit Swing pullbacks within that leg.
• The running sums are re-seeded from the actual pivot bar (Anchor Swing Right back) only in Backpainted anchor display mode; the pivot only sets the anchor origin — it is not a trigger, grade gate, or dedup key (design skill §8). In Confirmed mode the curve instead starts flat at the bar the pivot is actually confirmed, so the anchor is never drawn earlier than it was live-knowable — at the cost of a less clean historical look.
• Volume missing/zero on an occasional bar of an otherwise valid exchange feed (na or ≤ 0) uses weight 1.0 only on the affected bar in the default Per-bar fallback mode. The hover and detailed dashboard report availability across the complete anchor episode and name mixed weighting explicitly. Equal weight deliberately uses weight 1.0 for every bar when reported volume is not meaningful. Neither mode is a substitute for real volume at the instrument level — see Instrumente: when syminfo.volumetype itself is tick/n/a (CFDs, Forex, most indices), the whole curve is forced to na instead of falling back to either weighting mode.
• Reclaim Quality Score (0–5): counts prior stretch, close location within the full candle range, volume, structure break, and only a direction-coherent TBF bonus (Bottomfinder for bullish reclaim, Topfinder for bearish reclaim). EMA bias classifies the reclaim but no longer blocks it.
• Auto EMA bias only flips once price has closed on the new side of the EMA for Auto EMA: min hold bars consecutive bars (default 3), which reduces marker whipsaw in choppy phases; set it to 1 to restore the old immediate-flip behaviour.
• MIDAS, bands, and TBF use a one-bar visual break at each re-anchor so independent launch episodes are never joined by a misleading diagonal connector.
• Active mode retains a consistent 2,500 bars for MIDAS, zones, and canonical TBF statistics. The bounded active episode is redrawn on the last bar so Pine’s realtime rollback cannot reduce it to a one-bar fragment. Previous clouds and marker IDs are bounded; at most 300 active marker labels are retained. Full restores every historical episode and LATE/END events.
• The topfinder is hidden when no valid forward fit exists (no fit in the dashboard). This is expected in range-bound or non-accelerating markets: the topfinder/bottomfinder only projects accelerated (parabolic) legs, so it typically fits on higher timeframes / strong trends and correctly stays no fit in intraday ranges (verified on NatGas — daily legs fit and the exhaustion warnings land near the major 2021/2022/2025 tops, while 1h/15m ranges show no fit). Turn on Debug log to see why a fit was rejected (D<=d = projected end already passed / no acceleration).
• MIDAS (orange) crossing the TBF (purple) is not a signal. They are different objects — a cumulative-VWAP level vs an accelerated projection — and they coincide at the launch by construction. What carries meaning is the gap between them: a TBF pulling far from MIDAS = strong acceleration (the move is outrunning its fair value); the two staying close = little acceleration. A literal crossover is just the projection passing through the level, not a turn.
• Context markers remain deliberately secondary: stretch uses a small amber dot, directional reclaim labels use R n/5, and the rarer purple EXH n/3 appears only after an advanced TBF phase and a qualifying price reaction. Persistent marker creation defaults to confirmed bars; alerts keep their separate bar-close setting.