Polarity as Structure — the Two Poles, the Tension, and the Middle

The thirteenth companion — polarity not as a flip but as a standing structure: two opposing attractors, a productive tension between them, and a reconciling middle — mined across magnetism, bistability, yin-yang, and the dialectic for one structural invariant, then hammered into strict-causal, falsifiable "polar edge" query lenses.

This is the thirteenth entry in the deep-dive series and the direct successor to 12_path_of_least_resistance. Where 12 followed one thread (the gradient down which a flow descends), this one follows a different thread the operator keeps pointing at: polarity. Not the polarity_shift family's job — detecting the instant a sign flips — but polarity as a standing structure: at any moment the market is held between an upper attractor and a lower attractor, with a tension between them and a middle that either reconciles or repels. The discipline of this document is the opposite of 12's lyricism. 12 could afford to sing because it was naming a law the operator already trades. This one is a skeptic's ledger: every lens is dragged to the same three gates — is it STRICT-CAUSAL (feature ≤ t, signal t → entry t+1, forward returns/geometry are LABELS only), is it FALSIFIABLE (state the test that kills it), and does it plausibly move the operator's utility (WR / R:R-as-ratio / frequency / survivability)? If a "polar" lens just re-describes a family the Atlas already has (polarity_shift, value_area, statistical_regime, oscillation_state), it is killed as SKIP-redundant and said so plainly. Brand-only ideas are kept, but quarantined and labeled BRAND-ONLY so they never contaminate the edge ledger. The rule from the rest of the series still holds — no idea is worthless if it makes sense of something — but here the standard for "makes sense" is a forward-return test that could come back negative.

0. The distinction that starts the whole document

The Atlas already has a family called polarity_shift. It is good and it is memoryless and instantaneous: it watches a sign (price-vs-EMA, up-count vs down-count, multi-TF coherence) and reports the moment it flips and the present net. That is polarity-as-event.

The operator's idea is different, and the difference is the entire seed of this document. Polarity-as-structure says: forget the flip for a second. At every bar the market is suspended between two opposites — a pull up toward something and a pull down toward something — and the interesting, possibly-tradeable facts live in the geometry of that suspension:

• How far is each pole, and how strong? (the dipole field)
• Is the pull balanced (dead, choppy equilibrium) or imbalanced (loaded, directional)?
• Is the middle a resting place (attractive, mean-reverting) or an unstable saddle (repulsive, ejecting)?
• Does the present moment carry memory of which pole it last committed to? (hysteresis)
• Is the regime bistable (two crowded poles, hollow middle, fast snaps) or monostable (one central hump)?

None of those are "did the sign flip." They are the shape of the field the sign lives in. polarity_shift measures the needle; this document measures the magnet. That is the gap we are mining, and the test for every candidate below is whether it actually falls in that gap or whether it secretly collapses back onto the needle.

I. The cross-domain mining — and the one structural invariant

The operator handed a menu of lenses: magnetism, electric poles, bistability, bimodal distributions, tipping points, yin-yang, Taoist wu-wei middle, the Hegelian dialectic, the Masonic Jachin & Boaz, the Kabbalistic three pillars, masculine/feminine/androgyne. The skeptic's job is not to be charmed by them but to strip each to its mechanism and ask what testable structure it implies. Here is each, demystified to its load-bearing core.

Read down the right-hand column and the menu collapses into one structural invariant, repeated in every domain:

Two opposed attractors, a productive tension between them, and a middle that is either a reconciling rest or an unstable seam — with the system either dwelling at a pole or traversing between them, in a state that may carry memory of where it has been.

Everything tradeable in this document is some measurement of that one skeleton:

• the joint of the two pulls (dipole) — not one distance,
• the nonlinearity of a single pull near a pole (capture radius),
• the stability of the middle (rest vs saddle),
• the memory of the last commitment (hysteresis),
• the type of the regime (bistable vs monostable),
• the balance of time/force between the poles (occupancy, force-imbalance, co-presence),
• the early-warning that the balance is about to tip (critical slowing).

The mystic stops at "two poles and a middle, how beautiful." The skeptic asks: of those measurements, which one (a) the Atlas does not already compute, and (b) predicts a forward return in a way single-pole distance does not? That question is the rest of the document.

II. The skeptic's three gates (applied to every candidate)

1. Strict-causal. Every feature uses only bars ≤ t. Poles, strengths, distances, memory, regime reads — all computed from the closed window ending at the signal bar. The signal fires at t; entry is t+1. Forward returns, MFE/MAE, and the (SL,TP) geometry grid are labels (hindsight), never inputs. A candidate that needs to know where price went to compute a feature is dead on arrival.
2. Falsifiable. Each candidate ships with the exact test that would kill it — almost always: bin bars by the candidate's score, at matched location and matched volatility, and check whether the forward-return label actually separates. If it doesn't separate (or doesn't separate beyond what an existing column already explains), the lens adds nothing.
3. Utility-relevant and non-redundant. It must plausibly move WR, R:R, frequency, or survivability — and it must not be a recombination of columns the Atlas already emits. The harshest gate: a beautiful idea that turns out to equal oscillation_state.range_state_score or vwap_deviation is SKIP-redundant, no matter how good the story.

III. The ranked shortlist of polar-edge query proposals

Ranked by edge rating (1–5) then by novelty. Each carries mechanism, exact measurement over Atlas columns, why it is strict-causal, the kill test, rating, and verdict. The honest count up front: 9 BUILD, 3 SKIP-redundant, 1 BRAND-ONLY. The three skips are the empty-but-pretty ones, killed on purpose.

1. Hysteresis / remanence / coercive-threshold path-dependence — BUILD — edge 5/5 — the most novel thing in the whole document

Mechanism. A ferromagnet's next state depends on its history, not just the present field: drive it to a pole and relax, and it stays magnetized (remanence); flipping it requires overcoming a coercive threshold larger than the move that set the state. The structural invariant: identical present conditions yield different next-state odds depending on which pole the market last committed to and how hard it was driven there.

Why it is the most novel. The entire 39-family Atlas is effectively memoryless — every column is a present-window function. regime_transition finds the change-point but forgets the prior commitment; polarity_shift is an instantaneous cross; state_machine labels the instantaneous state; level_interaction tracks proximity/bounce but not an asymmetric switch cost. Nothing measures that the same momentum reading means different things depending on inherited polarity. This is a genuine blank spot.

Exact measurement (features ≤ t). Two framings of one mechanism, built as a single path-dependence module:

• Drive-state remanence. Track last_committed_pole = sign of the most recent regime that exceeded a high drive threshold (e.g. statistical_regime trend-quality or net-drift z past +H); remanence_age = bars since; remanent_bias = last_committed_pole · exp(−remanence_age/τ) (decaying inherited polarity); coercive_gap = how far current opposite-drive sits from the flip threshold.
• Level remanence. For each causal level L (naked VPOC from value_area, confirmed pivot from polarity_apex): record the set-point (how far past L price ran on the last cross, in ATR) and the current re-penetration depth (how far back into/through L price has come). hysteresis_state ∈ [−1,+1] = which side price is magnetized to; coercive_margin = re_penetration_depth − historical-median-reject-depth for that level type.

Two concrete tradeable claims. (1) Remanence/persistence edge — in the committed zone, fade small counter-moves; identical neutral-momentum readings carry a directional bias from the last pole, raising WR by not fading what isn't really a reversal. (2) Coercive filter — only respect a flip once it clears the measurable coercive width; a retest penetrating less than that width is a high-WR continuation entry, more = a genuine flip (don't fade it). Both serve WR + survivability by killing chop-fade losses.

Strict-causal. last_committed_pole, remanence_age, set-point, re-penetration depth are all functions of bars ≤ t. Continuation/reversal is the label.

Falsification. At matched current-momentum buckets, split bars by inherited last_committed_pole; if forward returns do not differ by inherited pole, there is no memory effect → kill. For coercivity: bin retests by re-penetration depth; if continuation WR does not fall monotonically across an estimable coercive width (recross odds symmetric in penetration), the loop is false → kill.

2. Dipole field tension / two-pole imbalance — BUILD — edge 4/5 — the literal "polarity as structure"

Mechanism. In a real dipole the object is not a single charge but the field stretched between two poles. At any bar the market is held between an upper attractor and a lower attractor simultaneously; the net force is the vector sum. The edge lives in the imbalance, not the position.

Maps to / gap. value_area (dist to VPOC/VAH/VAL) and level_interaction (nearest-level distance) are all single-level, signed-distance reads. None compute the joint two-pole net force. This is the operator's idea made literal and it is genuinely missing.

Exact measurement (features ≤ t). Pick p_up = nearest causal resistance attractor above close (swing high from structural_levels, HVN/VAH from value_area, prior-session high), p_dn = nearest below. With ATR-normalised distances d_up, d_dn and pole "charge" strength = volume-at-node or touch-count:

• net_force = strength_up/d_up² − strength_dn/d_dn²
• field_tension = strength_up/d_up² + strength_dn/d_dn² (total grip)
• field_imbalance = net_force / field_tension ∈ [−1,+1]

Tradeable claim. Wide, symmetric, balanced field (both poles far, pull balanced) = low-conviction equilibrium → t+1 is chop/mean-reverting, low edge. Imbalanced field (one pole much nearer/stronger, large one-sided net force) = price actively pulled → t+1 directional follow-through in the net-force direction. The polar edge is in the imbalance.

Strict-causal. Pole prices and strengths are features ≤ t; label is t+1 forward return / geometry. Cheap — derived from columns the Atlas already emits.

Falsification (the honest risk). field_imbalance may end up ~monotone in plain distance-to-the-nearer-pole. Bin bars by field_imbalance at matched ATR-vol AND matched distance-to-nearer-pole; if signed t+1 return shows no monotonic relationship with field_imbalance beyond what single-level distance already explains → the dipole framing adds nothing → SKIP.

3. Capture radius / near-field magnet — BUILD — edge 4/5 — defined-geometry, frequency-friendly

Mechanism. A pole's pull is nonlinear in distance: negligible far away, dominant inside a capture radius where it snaps the object in. A level is a magnet only within its near-field.

Maps to / gap. value_area emits naked-POC distance linearly; level_interaction has an ATR-kernel proximity but not a force-law / fill-rate model with a near-field plateau. The novel test is whether fill-rate is a step/saturating function of distance, not linear.

Exact measurement (features ≤ t). Take the strongest untested causal magnet within scan range (naked POC, untested HVN, unfilled gap edge from value_area/structural_levels). Emit two force laws pull_1r = strength/d_atr, pull_inv_sq = strength/d_atr²; a soft saturating gate capture_score = sigmoid((r_capture − d_atr)/w) ∈ [0,1] (high inside the near-field); magnet_dir = sign(magnet_price − close).

Tradeable claim (defined geometry). "Price has entered the capture radius of a strong unfilled magnet pointing toward it" → fixed target = magnet price, stop = beyond the opposite pole. High-WR, fixed-R:R, frequency-friendly — exactly the operator's utility shape.

Strict-causal. Magnet identity + distance ≤ t; "reached magnet within H bars" is the hindsight label. Signal t → entry t+1.

Falsification. Fit forward-fill-rate vs d_atr to the dominant untested magnet; if the curve is flat or linear with no near-field plateau/knee, there is no capture radius → it collapses to plain distance → SKIP. (And validate as a new lens vs level_interaction proximity, not a reuse.)

4. Bimodality / bistability regime gate — BUILD — edge 4/5 — consolidates four lens entries

Mechanism. A bistable system has two stable attractors and an unstable barrier between; it dwells at a pole and snaps fast between them, spending little time in the hollow middle. The signature is a bimodal state distribution + long dwell + fast snap.

Maps to / gap. statistical_regime emits skew/kurtosis (unimodal-assumed moments); value_area emits bimodality of volume-at-price (a magnet read); oscillation_state.amplitude_percentile is an amplitude read. None detects two-modedness of the trailing return/price distribution. This entry consolidates four overlapping lenses (bimodal-detection, bistability/snap, two-attractor-structure, pole-dwell-vs-traversal) into one object.

Exact measurement (features ≤ t, windows ending at T). On trailing returns/prices over W ∈ {60,120,240} (1m + causal HTF):

• Sarle's bimodality_coeff = (skew² + 1)/kurtosis (>~0.555 suggests bimodal), continuous;
• hollow_center = density at the central bin / mean density of the two modal bins (low = hollow);
• 1D-kmeans(k=2): mode_separation = |c1−c2|/ATR, mass_balance = min(n1,n2)/max(n1,n2);
• middle_dwell = current close's local kernel-density rank in the window (low = in the depleted trough). Optionally pole_dwell (mean run-length where |state|>θ) and snap_speed (bars to cross +θ→−θ). Combine into bistability_score ∈ [0,1].

Tradeable claim (a regime gate, not a standalone edge). High bimodality + middle-dwell flips a mid-range pullback from "fade" to "go-with" and supports breakout geometry (snap pole-to-pole = clean R:R). Monostable/centered → mean-reversion wins, breakouts fail. Improves the frequency-quality tradeoff — the operator's lever — by firing each playbook only in the regime that supports it.

Strict-causal. Window ends at T; all moments/clusters from bars ≤ t. Forward breakout-vs-fade outcome is the label.

Falsification. At matched location/vol, if forward |return|, traversal length, and realized R:R at an opposite-mode target do not separate between high- and low-bimodality bars (bootstrap on April+May OOS) → kill. Must also beat oscillation_state.range_state_score head-to-head as a fade-vs-breakout gate, or → SKIP-redundant.

5. Force-imbalance toward the reconciling middle — BUILD — edge 4/5 — the dialectic, made into a trap-filter

Mechanism. Hegelian thesis/antithesis → synthesis; Kabbalah's Mercy/Severity resolved on the Middle Pillar of equilibrium. Two opposing flows (buy-pressure vs sell-pressure) resolve toward a synthesizing fair-value middle. Reversion should scale with the force imbalance pointing back toward the middle — not merely with displacement.

Maps to / gap. value_area gives where the middle is (VPOC); vwap_deviation gives stretch + a reversion score; orderflow_pressure gives signed net force. The interaction — force pointing back toward the middle — is computed nowhere. Existing reversion reads are distance-driven; this adds the engine term.

Exact measurement (features ≤ t). Equilibrium E(t) = causal VWAP or rolling VPOC; displacement d = (close − E)/ATR; net force F(t) = normalized signed pressure from orderflow_pressure + volume-delta + body-direction·volume. Then:

• restoring_imbalance = −sign(d) · F (positive = force points back toward the middle; the antithesis is winning),
• tension_balance = 1 − |F|_norm (1 = balanced/dead, 0 = one-sided),
• synthesis_pull = restoring_imbalance · |d| (loaded-spring magnitude).

Tradeable claim (a trap-filter that lifts fade WR). Far-from-middle + balanced force = weak reversion, a trap (no engine). Far-from-middle + force loading against the displacement = strong, fast reversion → high-WR fade, middle as TP, recent pole as stop. The field, not the distance, is the predictor.

Strict-causal. E from value_area/VWAP ≤ t, F from orderflow_pressure ≤ t. Forward reversion distance is the label.

Falsification. Regress forward reversion-to-E on displacement alone vs displacement + restoring_imbalance; if the force term adds no OOS predictive power, it collapses to plain distance-driven mean-reversion → SKIP-redundant. If it adds power, it is a genuine new gate.

6. Unstable null-point ejection — BUILD — edge 3/5 — the complement of VWAP reversion

Mechanism. Between two poles is a null point (net field = 0); for a saddle it is unstable — tiny displacement is amplified away. Sitting on the null is indecision; departing it is self-reinforcing toward whichever pole price heads to.

Maps to / gap. Null = midpoint of the dipole pole-pair (Mechanism 2) or session VWAP. Adjacent to vwap_deviation.band_reversion_score, but the ejection direction (continuation away from center once displacement starts increasing) is the opposite regime to VWAP reversion and is not emitted.

Exact measurement (features ≤ t). null = (p_up + p_dn)/2 or session VWAP; disp = (close − null)/ATR; ejection = sign(disp)·(|disp_t| − |disp_{t−1}|) (departing if >0); dwell_at_null = consecutive bars with |disp| < ε.

Tradeable claim. Dwelling at the null = low-edge chop. Small |disp| AND ejection > 0 AND directional = a t+1 launch away from center (unstable-equilibrium ejection) — a continuation signal complementary to fade-to-VWAP.

Strict-causal. disp_t, disp_{t−1}, dwell all ≤ t. Forward continuation is the label.

Falsification. Condition forward continuation on (small |disp| AND positive ejection AND directional); if it does not beat the unconditional base rate → kill. Redundancy check: if ejection correlates ~1 with the inverse of vwap_deviation.band_reversion_score it is just an existing column negated → SKIP-redundant. (Edge 3: launch-from-balance is lower-WR / higher-variance than magnet-fill; it earns its place as a regime axis, when to fade vs ride from center.)

7. Critical slowing down — tipping-point early-warning — BUILD — edge 3/5 — a true leading indicator

Mechanism. Just before a bistable system tips, perturbations decay slower (lag-1 autocorrelation → 1) and variance inflates — the canonical complex-systems early-warning signal. When both are rising together, a flip/breakout is imminent.

Maps to / gap. statistical_regime emits autocorr and vol levels; regime_transition detects the shift as/after it happens (CUSUM/novelty with confirmation lag). Neither emits the co-rising slope of AR(1) and variance as a pre-tip leading score. This also absorbs the useful piece of the polarization "susceptibility" idea (Mechanism 13).

Exact measurement (features ≤ t). Trailing AR(1) of returns and trailing variance over a short sub-window; take their short-horizon slope (linear-fit over the last k estimates, all ≤ t). critical_slowing_score = tanh(ar1_slope_mag · var_slope_mag · [both>0]) ∈ [0,1]. Emit ar1_slope and var_slope separately.

Tradeable claim. Fired the bar before the break → entry t+1 catches the move instead of chasing it. Improves frequency-quality (fewer entries, each near a real tip) and survivability (avoids mid-stable-regime noise).

Strict-causal. All estimates end at T. Has a clean label: regime_transition's confirmed change-point marker.

Falsification. If the rate of regime_transition confirmed change-points is not elevated following high critical_slowing_score (vs base rate), the early-warning is absent here → kill. (Edge 3: market early-warnings are false-positive-prone; the kill test does real work — but the label is clean.)

8. Polarity balance / time-occupancy imbalance — BUILD — edge 3/5 — the yin-yang clock

Mechanism. Yin-yang / wu-wei middle: a healthy oscillation spends balanced time on both poles; persistent imbalance (window spent overwhelmingly above or below center) = one pole dominating, an accumulating directional pressure that either sustains (trend) or snaps back (the suppressed pole reasserts).

Maps to / gap. Distinct from value_area.balance (symmetry of the volume profile, a static auction shape) and oscillation_state.range_state_score (efficiency). This is a time-occupancy count of price vs a central tendency — emitted nowhere.

Exact measurement (features ≤ t). With centered oscillator o(t) or sign(close − rolling_median) over bars ≤ T: polarity_balance = (#above − #below)/N ∈ [−1,+1]; occupancy_extremity = |polarity_balance|; balance_shift = polarity_balance(recent) − polarity_balance(prior) (rebalancing vs still building). Optionally amplitude-weight each bar by |o(t)|.

Tradeable claim. Extreme one-sided occupancy that has just begun to roll back (high extremity + opposing balance_shift) = high-quality reversion-to-middle entry; still-building one-sidedness = trend continuation.

Strict-causal. All counts over bars ≤ T. Forward return is the label.

Falsification. If forward returns are independent of occupancy_extremity and its first difference (balance_shift) at matched location/vol → kill. (Edge 3: likely correlated with longer-window trend reads, so the independence test matters — build alongside the bistability gate, shared machinery.)

9. Polarity flip-flop frequency — BUILD — edge 3/5 — the oscillation-frequency axis

Mechanism. Bipolar systems have a characteristic oscillation frequency: chop = high flip rate (many fast pole-to-pole reversals); trend/coil-then-break = low flip rate. Flip-rate is a regime conditioner orthogonal to amplitude.

Maps to / gap. Partial overlap with fractal_texture.zigzag_legs_count (coarse swing count) and macd_derived zero-cross events/bars-since. Neither emits a normalized windowed flip-RATE of a centered polarity oscillator as a continuous frequency axis with a recent-vs-prior shift.

Exact measurement (features ≤ t). Centered oscillator o(t) = sign(close − rolling_median) or sign(fast_EMA − slow_EMA) (≤ t). flip_rate = (sign-changes of o over last N)/N ∈ [0,1]; dominant_period = N/flip_count; flip_rate_shift = flip_rate(recent) − flip_rate(prior) so a slowing oscillation (locking into a leg) is detectable.

Tradeable claim. High-amplitude low-flip swing = clean tradeable traverse; high-amplitude high-flip = whipsaw that destroys WR. Gating to a falling flip-rate (oscillation locking into a directional leg) plausibly lifts WR and cuts death-by-a-thousand-stops.

Strict-causal. All sign changes over bars ≤ N. Forward WR/stop-out is the label.

Falsification. Bucket a fixed entry rule by flip_rate; if WR and stop-out frequency are invariant across buckets → kill. If flip_rate correlates ~1 with normalized zigzag_legs_count → SKIP-redundant (demote).

10. Yin-in-yang co-presence — BUILD — edge 3/5 — the seed-dot, kept un-netted

Mechanism. Yin-yang's seed-dot: the germ of each pole living inside its opposite. The object is not net polarity but the simultaneous strength of BOTH the bull-force and the bear-force at the same instant — a charged, ambivalent bar.

Maps to / gap. polarity_shift emits up_count minus down_count (a net); divergence is one indicator vs price. Neither emits co_presence = min(B,S)/max(B,S) or tension = B·S — the "both poles strong at once" axis is orthogonal to net polarity and emitted nowhere.

Exact measurement (features ≤ t). Build bull-force B(t) and bear-force S(t) each as a normalized composite (≥0) of same-instant evidence — B from {close in upper bar range, lower-wick rejection, RSI rising, MACD>signal, up-volume pressure}, S the mirror — all from existing atomic indicators ≤ t. Do NOT emit the net (that is polarity_shift). New columns only: co_presence = min(B,S)/max(B,S) ∈ [0,1], tension = B·S, ambivalence_at_extreme = co_presence · |dist from recent range mid|.

Tradeable claim. High co-presence at a pole (strong up-thrust that also prints a large rejection wick + bearish microstructure) marks the exhaustion seam and reverses more often; low co-presence = clean one-sided continuation; rising co-presence inside a winner = exit cue.

Strict-causal. B, S from indicators ≤ t. Forward reversal/continuation, MFE-before-adverse, are labels.

Falsification. At matched net_polarity AND matched location, if high-co_presence bars show no elevated reversal rate / no shorter favorable excursion vs low-co_presence bars → the seed-dot has no market content → kill.

11. Dwell-time at pole vs fast traversal (occupancy-fraction) — SKIP-redundant — edge 2/5

Why it dies. The dwell-fraction-near-pole vs time-in-transit idea is the efficiency ratio already emitted as oscillation_state.range_state_score (net move / path length is precisely "how much of the window was spent parked vs in motion"). The occupancy-fraction reformulation adds little beyond efficiency + Mechanism 4's transit_fraction. Do not emit a separate family. Fold the one genuinely-distinct sub-measure (pole_stickiness as a soft occupancy weight) into the bistability BUILD (#4).

The test it already loses. If high-dwell vs high-traversal MFE/MAE label distributions correlate ~1 with range_state_score buckets at matched location, it is redundant — and that head-to-head is already lost to an existing column.

12. Over-extension magnitude gated by antipersistence — SKIP-redundant — edge 2/5

Why it dies. Both halves already exist. Over-extension = vwap_deviation.vwap_dev_sigma/vwap_dev_atr and ema_derived deviation (already signed sigma/ATR stretch from a central attractor). Antipersistence = statistical_regime Hurst / variance_ratio / autocorr_lag1. The contribution collapses to "query existing stretch × existing Hurst jointly" — a confluence query recipe, not a new column. Emit nothing.

The recommendation that survives (as a query, not a build). Stretch in sigma AND Hurst<0.5 → fade to center; same stretch with Hurst>0.5 → skip/continuation. Worth running if the joint test separates forward reversion within the antipersistent bucket vs the persistent bucket — but it stays SKIP-redundant as a build item because it adds no measurement column.

13. Polarization / magnetization order parameter — SKIP-redundant — edge 2/5

Why it dies. M = mean of signed unit reads across TFs/indicators is a near-linear recombination of columns the Atlas already has: polarity_shift multi-TF coherence + per-TF directional signs (price-vs-EMA from ema_derived, RSI-vs-50 from rsi_derived, MACD-vs-signal from macd_derived). A cluster engine over those columns recovers |M| already. Its susceptibility (rolling variance of M near the flip) overlaps the cleaner critical-slowing formulation (#7).

What to do instead. Route the only useful pieces — polarization_slope (ordering vs demagnetizing) and the critical-point variance — into the critical-slowing-down BUILD (#7). Treat |M| itself as a query-time recombination. Build nothing new here.

The test it already loses. Correlate |M| against polarity_shift multi-TF coherence; if ~identical (and susceptibility ~ var_slope), it is a recombination → SKIP.

14. Keystone / Seam / two-pillars duality narrative — BRAND-ONLY — edge 1/5

What it is. Not a market mechanism — a naming + UX skin for the duality cluster: Jachin & Boaz = the two poles, the Keystone / Middle Pillar = the synthesizing equilibrium (the value_area middle), the Seam = the traversal of the depleted middle (the tradeable membrane). It ties value_area-as-middle + force-imbalance + bimodal-poles into one mental model, and rhymes with the existing brand soul (the zero-with-a-line-through-it membrane in 00 and 08).

Quarantine. Explicitly non-falsifiable as a market claim and must never count toward WR/R:R/frequency validation. Legitimate only as a label: "Keystone" read = price at the synthesis with imbalanced force (Mechanism 5); "Seam" setups = traversal of the depleted middle (Mechanism 4/6). Any P&L effect would be an operator-behavior/placebo effect — testable, if ever, only via a journaling A/B on adherence/override rate, kept strictly separate from the edge's own OOS.

IV. The scoreboard

The structural invariant, satisfied honestly: the BUILDs cover every limb of the skeleton — the joint (2), the nonlinearity (3), the unstable middle (6), the memory (1), the regime type (4), the time-balance (8), the frequency (9), the co-presence (10), the force-toward-synthesis (5), and the pre-tip warning (7). The skips are exactly the limbs the Atlas already measures (occupancy ≈ efficiency, extension × Hurst ≈ two existing families, alignment ≈ multi-TF coherence).

V. RECOMMENDATION

Add a polarity_structure query lens — but as a thin, derived family layered over existing Atlas columns, not a new heavy measurement build. Almost every BUILD here is computed from columns that already exist (level prices + strengths, VWAP/VPOC, orderflow pressure, returns moments, atomic indicators). That makes the whole family cheap and squarely in the operator's "measure-once-query-infinitely" mode: it is a confluence/query layer, not a re-measurement of the terrain.

The one to lead with — and the only one I'd fold into the maximal run (WF4): Mechanism 1, Hysteresis / Remanence. Reasons:

1. It is the single genuinely novel axis the entire memoryless 39-family Atlas lacks (path-dependence). Everything else is "join/transform existing columns"; this one adds a dimension.
2. Its two claims map directly to the operator's hard utility: the remanence claim raises WR (stop fading non-reversals in the committed zone); the coercive claim raises survivability (filter false flips = fewer chop losses) — exactly the levers WF4 is being graded on.
3. It has a clean kill test (matched-momentum split by inherited pole; monotone-WR-vs-penetration), so it can fail fast and visibly rather than quietly overfit.

Sequencing — be disciplined, do not over-stuff WF4.

• Fold into WF4 now: Mechanism 1 (hysteresis) as a candidate gate/feature, plus Mechanisms 2 and 5 if and only if they pass a quick pre-WF4 redundancy screen (dipole imbalance must beat single-pole distance at matched distance; force-imbalance must beat displacement-alone OOS). Those two are cheap derivations and align with the maximal-utilization mandate, but they carry real collapse risk, so they earn their seat only by passing the screen first.
• Run as a follow-on (after WF4): Mechanisms 3, 4, 6, 7, 8, 9, 10. They are legitimate BUILDs but are mostly regime conditioners (gates that improve the frequency-quality tradeoff) rather than standalone edges; adding seven conditioners into the maximal run at once invites multiple-comparisons noise and dilutes the verdict. Validate them individually on the April+May OOS as a second polarity_structure battery, promote only those that beat their named existing-column rival head-to-head (#4 vs range_state_score, #9 vs zigzag_legs_count, #6 vs band_reversion_score).
• Build nothing for #11–13 (covered) and #14 (brand). Record #12 as a query recipe (stretch × Hurst) in the confluence catalog, not a column.

Brand: adopt Keystone / Seam / Jachin–Boaz purely as the name for the polarity_structure cluster in the UI — it ties cleanly to the existing zero-line membrane brand soul — but keep it walled off from every WR/R:R/frequency number.

Bottom line. Polarity-as-structure is not empty symbolism re-skinning polarity_shift. There is one real new dimension in it — path-dependence / hysteresis — and that one is worth putting into WF4. The dipole-imbalance and force-toward-middle joins are worth a fast pre-screen for inclusion. The rest are a solid follow-on conditioner battery. Three lenses are pretty but redundant and were killed on purpose; one is brand only. Measure the magnet, not just the needle — but only the parts of the magnet the Atlas can't already see.

Companion to 12_path_of_least_resistance (the prior entry — flow down a gradient), 08_union_of_opposites (the two poles and the surviving knife-edge — this document is 08's thread dragged through the skeptic's gates), and 00_THE_ZERO_AND_THE_INFINITE (the zero-line membrane, here the reconciling middle). Where 08 sang the union, this one audits it: of all the beautiful two-poles-and-a-middle stories, exactly which ones survive a forward-return test. The bolt finds the gap (12); the magnet holds the field (13); the operator trades the seam between them.