Mine Shaft + Drift + Ore Pocket Detector (Gap+Touch)Mine Shaft + Drift + Ore Pocket Detector (Gap+Touch) — Full Description (v1.6.1, Pine v6)
*Experimental - *Test Phase*
1) What this indicator is intended to do
This indicator attempts to algorithmically discover “mine shaft” price structure on a chart by:
Collecting structural anchor points (gaps and optionally pivots),
Generating candidate trend “rails” (centerline + parallel upper/lower borders) from pairs of anchors,
Fitting an optimal channel width around each candidate centerline,
Scoring candidates based on how well price action conforms to the channel (touches + containment),
Selecting and rendering:
the main shaft channel (primary),
additional drifts (secondary shafts per direction),
And then detecting Ore Pockets: time locations where multiple selected lines intersect (time confluence / intersection clustering).
The conceptual model is:
A shaft = a best-fit channel that price respects over time (the “main tunnel”).
Drifts = alternate channels close in quality to the main shaft (secondary tunnels).
Ore pockets = future/past time coordinates where multiple channels’ centerlines intersect densely (confluence in time, not necessarily in price).
2) What it is doing right now (current behavior)
In its current form, the script does a bounded, performance-limited scan:
It stores a limited number of anchor points in arrays.
It only considers a bounded number of recent anchors per direction.
It constructs candidate lines from anchor pairs and evaluates channel fitness using sampled bars.
On the last bar, it selects top candidates per direction and draws:
a “main” channel per mode (single best overall, or separate up/down),
plus optional drift channels,
plus ore pocket markers.
It is producing meaningful channels and drifts, but it is currently more likely to lock onto a strong “local” shaft than the one macro shaft spanning the entire market structure.
3) Core mechanics (how the script finds shafts)
3.1 Anchor generation (what points it uses)
Anchors are the “support points” used to build candidate shaft centerlines.
Two anchor families are supported:
A) Gap anchors (from your selected gap mode)
These attempt to capture “displacement events” and their boundaries/mids.
B) Pivot anchors (optional structural anchors)
These use pivots to inject macro structure points that are not strictly gap-based.
All anchors are stored as:
anchorX: bar_index of anchor
anchorY: price of anchor
anchorD: direction flag (+1 for up, -1 for down)
Anchors are capped by maxAnchors with FIFO trimming.
3.2 Candidate generation (how it produces centerlines)
For each direction (+1 and -1):
Collect “recent” anchors of that direction within lookbackBars (bounded to maxDirAnchors).
For each pair of anchors (x1,y1) and (x2,y2) that satisfy:
spacing within ,
slope sign consistent with direction,
Construct the line equation:
slope m and intercept b
Fit a channel width w around that line (via width mode).
Score it (touches + inside count minus width penalty).
Keep the top K rails (K = driftCount+1 typically).
3.3 Scoring model (what “best” means right now)
For a candidate centerline:
At sampled bars (stride sampling), compute:
channel top = y(x) + w
channel bot = y(x) - w
Evaluate:
Inside: candle range fits within the channel ± tolerance
Touches: high near top border, low near bottom border (within tolerance)
Score formula:
score = insideCount * insideWeight
+ touchCount * touchWeight
- (w / ATR) * widthPenalty
So:
Higher inside and touch counts increase score
Wider channels are penalized (in ATR units) to avoid “cheating” via enormous width
3.4 Width fitting (how the channel thickness is chosen)
Width is either:
Fit (scan widths): scans widths between a min width and a max deviation cap and selects the best scoring width.
Fixed ATR Envelope: uses a fixed width derived from ATR (currently hard-coded to a 2.0 ATR envelope in your present draft).
Fixed Max Deviation: width is max observed deviation from line in sampled window.
This matters because “macro shaft” detection is strongly influenced by whether the width-fitting is allowed to expand enough to contain large historical moves, without being penalized into losing to a smaller local shaft.
3.5 Rendering (what gets drawn)
For any selected rail, it draws:
Upper border line (top rail)
Lower border line (bottom rail)
Optional centerline (main only)
Optional fill between borders (main only)
Label at current bar with touches and inside count
Drifts render similarly but without main-only features (depending on flags).
3.6 Ore Pocket detection (time confluence)
Ore pockets are not “price zones” directly.
They are computed as follows:
Collect selected centerlines (m,b) for:
the main selected shaft(s),
and all drift centerlines (both directions if present)
For each pair of selected lines, compute intersection x-coordinate:
x* = (b2 - b1) / (m1 - m2)
Only keep intersections within:
Cluster intersections by time proximity (clusterBars)
Mark the strongest clusters (highest counts) as “Ore Pocket” vertical dotted lines with labels.
Interpretation:
A dense cluster indicates many selected rails converge around a similar time coordinate.
It is a “time confluence” hypothesis point.
4) Full settings reference (what each setting is for)
01) Gap Anchors
Gap Mode
FVG (3-candle)
Uses a classic 3-candle fair value gap pattern:
Up gap if low > high
Down gap if high < low
Anchors are derived from the gap boundaries.
Candle Gap (open-close)
Gap based on open vs close of the same bar with a tick threshold.
Candle Gap (open-prev close)
Gap based on open vs close with a tick threshold.
Gap Threshold (ticks)
Only used for the candle gap modes.
Controls the minimum gap size required to register an anchor.
Anchor Price
Boundary: anchors at one gap boundary (more “structural edge”)
Mid: anchors at midpoint of the gap (more “center of displacement”)
Include Pivot Anchors (structure)
When enabled, adds pivots as additional anchors to stabilize macro detection.
Pivot Length
Pivot sensitivity (how many bars left/right define a pivot).
Larger values = fewer, more structural pivots.
02) Channel Fit + Touch Scoring
Lookback Bars
The historical window used to:
filter which anchors are considered “recent enough”
evaluate channel fitness (sampled evaluation)
Larger lookback tends to favor macro shafts, but also increases computational risk (mitigated by evalBars and stride).
ATR Length
ATR period used for tolerance and width penalty scaling.
Tolerance (ATR mult)
Defines how close price must be to a rail to count as “touch” and how strict the “inside channel” containment is.
Higher tolerance = easier to score high on touch/inside.
Min Border Touches (keep rail)
Minimum number of border touches required before a candidate is even eligible.
Score: Inside Weight
Weight of inside count in score.
Score: Border Touch Weight
Weight of border touches in score.
This is a strong driver of “shaft-like” behavior.
Score: Width Penalty (in ATRs)
Penalizes wide channels relative to ATR.
Higher penalty biases toward narrow/local shafts.
03) Performance Controls
Max Stored Anchors (global)
Maximum anchor points kept in memory arrays.
Too low can cause loss of macro structure; too high increases candidate noise.
Max Anchors / Direction (scan)
Hard cap on how many anchors are used in candidate generation per direction.
Critical: this strongly influences whether macro shaft can be found, because if you only keep the most recent anchors, you lose the early-structure anchor points.
Eval Bars (max)
Maximum historical bars actually evaluated for scoring.
Even if lookbackBars is large, evaluation is capped here.
Eval Stride (sample every N bars)
Sampling step for evaluation.
Larger stride = faster but less accurate scoring.
04) Candidate Generation
Min Anchor Spacing (bars)
Minimum distance between the two anchors used to define a candidate line.
Prevents micro-noise lines from being evaluated.
Max Anchor Spacing (bars)
Maximum distance between the two anchors used to define a candidate line.
If this is too low, you cannot generate truly macro candidate lines.
05) Shaft + Drift Display
Main Shaft Mode
Best Overall (Single Shaft): chooses one best rail among Up/Down and draws it as main.
Up Only: show only the best upward rail.
Down Only: show only the best downward rail.
Up + Down: show both main up rail and main down rail simultaneously.
Show Ascending Shaft
Toggles rendering for the “up” main shaft (when mode allows it).
Show Descending Shaft
Toggles rendering for the “down” main shaft (when mode allows it).
Drifts per Direction
Number of additional top-ranked rails to draw per direction (after the best one).
Extend Lines
Right: extend lines to the right only.
Both: extend both left and right.
Fill Main Shaft Channel
Fill between upper and lower borders for main shaft.
Main Shaft Fill Transparency
Transparency level for main fill.
Show Main Shaft Centerline
Draw the dashed centerline for the main shaft.
06) Ore Pocket (Intersection-Time Confluence)
Show Ore Pockets (Time Confluence)
Enables ore pocket discovery and rendering.
Intersection Window Forward (bars)
How far into the future intersections are considered.
Intersection Window Backward (bars)
How far into the past intersections are considered.
Cluster Radius (bars)
How close in time intersections must be to merge into a cluster.
Min Intersections per Cluster
Minimum cluster count required before a pocket is shown.
Max Pocket Markers
Limit how many pocket clusters are drawn.
07) Visual Controls
Show Gap Anchors
Displays the gap anchor dots for debugging.
Show Pivot Anchors
Displays pivot anchor dots for debugging.
5) How to use it (practical workflow)
Step A — Confirm anchor behavior
Turn on Show Gap Anchors.
Choose your Gap Mode.
Verify you are seeing anchors where you expect (displacement boundaries).
If anchors are sparse:
Reduce gap threshold (ticks) for candle-gap modes
Enable pivots to inject structure
Increase lookbackBars and maxAnchors so early anchors are not dropped
Step B — Get stable main shaft candidate discovery
Enable Include Pivot Anchors with a medium pivotLen.
Use Fit (scan widths) initially.
Increase Max Anchors / Direction (scan) so you’re not only using recent anchors.
Increase Max Anchor Spacing so macro pairs are eligible.
If you keep getting only local shafts:
That is usually because the candidate pool does not include enough old anchors, or the maxSpacing prevents long-span lines.
Step C — Tune scoring so the “whole-structure” shaft wins
If the script picks a small local channel instead of the macro channel:
Increase insideWeight relative to touchWeight (macro channels tend to contain longer structure even with fewer perfect “touches”)
Reduce widthPenalty, because macro channels may need to be wider to accommodate historical volatility
Increase lookbackBars and evalBars to make “whole-structure fit” matter
Step D — Drifts as secondary shafts
Once main shaft is good:
Increase Drifts per Direction
Validate that drifts represent meaningful alternate sub-shafts rather than noisy duplicates.
If drifts look too similar:
This is expected if many candidates differ only slightly; future refinements should diversify drift selection (see “what still needs done”).
Step E — Ore pockets interpretation
Ore pockets indicate time confluence of multiple rails.
Use them as:
“Time windows to watch”
Not as deterministic price levels
Tune:
clusterBars (cluster tightness)
minClusterSize (signal strength)
6) What still needs done (explicit backlog)
The macro “main mining shaft channel” spanning the entire market structure, and
Smaller shafts/drifts nested inside the macro structure.
To accomplish that, the current algorithm needs additional architecture. Concretely:
A) True multi-scale / hierarchical discovery (primary missing feature)
Right now: one pass, one lookback, one score objective.
Still Needed:
Macro pass: discover a primary shaft using a very long evaluation window and anchor set.
Micro pass(es): discover drifts/secondary shafts using:
residuals (distance from macro centerline),
or segmented time windows (regime partitions),
or anchor subsets constrained to local regions.
This is the single biggest reason we are not consistently getting the full-structure shaft.
B) Anchor retention strategy for macro detection
Right now:
anchors are FIFO capped and direction scanning uses “recent anchors only.”
To reliably find 10-year shafts we need:
an option to store/retain representative anchors across the entire history, not only the most recent ones.
Examples of necessary improvements:
“Stratified anchor sampling” across time (keep some old anchors even when maxAnchors is hit)
“Macro anchor bank” (separate storage for pivots or major gaps)
C) Candidate generation constraints must support macro lines
If we want a shaft spanning the whole structure:
maxSpacing must allow it
the candidate pool must contain anchors far apart in time
So the algorithm needs:
better selection of anchor pairs for long-span candidates (e.g., include earliest/oldest anchors + newest anchors deliberately, not accidentally)
D) Drift diversification
Right now drifts are “next best by score,” which often yields near-duplicates.
We want:
“diverse” secondary shafts:
enforce minimum angular difference,
enforce minimum offset difference,
or penalize candidates too similar to the already-selected shaft.
E) Width fitting logic for macro channels
Macro channels often require:
either a higher width cap,
or a different penalty profile.
Current width penalty is simple and can bias against macro channels.
Needed:
width penalty that scales by timescale or by total evaluated bars,
or separate macro/micro scoring.
F) Ore pocket semantics enhancement (optional but aligned)
Currently pockets are time intersections only.
If you want “pocket zones,” improvements could include:
projecting intersection price and drawing a zone box,
clustering in (time, price) space instead of only time,
adding “importance” weighting based on which lines intersect (macro line intersections weighted higher).
7) Known limitations (current version)
Heavy compute only runs on last bar (good for performance), but means:
changes in anchors/parameters can reselect rails abruptly
Candidate set is bounded; macro shaft can be missed if not in pool
Drift selection can be redundant
Ore pockets are time clusters, not price clusters
Pine Script® 인디케이터
