experiments: midnight exp23 outcomes + contrast/sign-structure scripts + Milestone R (radiation as PAC ledger severance)

Adds midnight journals (exp23 outcomes, balance-coherence-QBE synthesis), exploration scripts (contrast, inconclusiveness, sign structure), and the Milestone R experiment set (radiation = PAC ledger severance).

Author: lornecodes

foundational/experiments/midnight/README.md

@@ -14,6 +14,7 @@
 - [2026-06-13 — Maximal Coherence as the Connection Primitive (theory note)](journals/2026-06-13_maximal-coherence-connection-primitive.md)
 - [2026-06-13 — exp_23 Within-Scope Local Coupling Law: pre-registration](journals/2026-06-13_exp23-preregistration.md)
 - [2026-06-13 — exp_23 outcomes (INCONCLUSIVE, suggestive; global→local correction works)](journals/2026-06-13_exp23-outcomes.md)
+- [2026-06-14 — Coherence as Balance: the QBE as the dynamics of the connection primitive (theory)](journals/2026-06-14_balance-coherence-qbe-synthesis.md)
 
 ---
 

foundational/experiments/midnight/journals/2026-06-13_exp23-outcomes.md

@@ -98,3 +98,49 @@ established.
   or rule touched).
 - Running Midnight meta-pattern intact: the **local/relational** form of a prediction carries
   signal; the **global/coordinate** form (exp_19/21) carried artifacts.
+
+---
+
+## Post-hoc diagnostic (2026-06-14): why inconclusive, and the corrected observable
+
+Exploratory follow-up (`explore_inconclusiveness.py`, `explore_contrast.py`) — **post-hoc on the
+same data, hypothesis-generating, NOT a result**; the registered verdict above stands. Captured
+because it sharply reframes the next prediction.
+
+**The inconclusiveness was structural + methodological, not low power.**
+- *Wrong test.* R1 tested a monotone IP-ordering, but leave-one-ion-out shows the "trend" is
+  carried entirely by the two IP extremes (drop AlII or CIV → rho 0.43→0.09; drop any middle ion
+  → rho rises to 0.49–0.66). The middle (SiII, CII, SiIV: 8–34 eV) is a dead-flat plateau at
+  c = −1.07 ± 0.03.
+- *Centering shadow.* That flat plateau is a mean-centering artifact: subtracting the scope mean
+  imprints a common recoil slope on passive ions. Removing the centering (direct pairwise
+  contrasts) dissolves the plateau and exposes the signal.
+
+**The corrected picture — the carriers are CIV and MgII, not an IP class.**
+Centering-free, slope of (logN_hi − logN_lo) vs local cascade phase diseq:
+
+| pair | n | slope | CI95 |
+|------|--:|------:|------|
+| CIV − AlII | 24 | +4.22 | [+2.47, +6.30] ✓ |
+| CIV − AlIII | 11 | +3.05 | [+0.55, +8.10] ✓ |
+| CIV − FeII | 28 | +2.35 | [+0.57, +4.70] ✓ |
+| **SiIV** − FeII | 19 | −0.88 | [−3.62, +1.52] |
+| CIV − **MgII** | 30 | −0.17 | flat |
+
+- **CIV** swings hard against passive ions toward cascade transitions. **SiIV** (also IP > E_H)
+  does **not** — so the E_H / IP-threshold split is the wrong axis.
+- **CIV − MgII is flat** → MgII swings *with* CIV. The active set is **CIV + MgII** — exactly the
+  two clean resonance doublets the 2026-06-08 tapestry flagged (MgII p=10⁻¹¹, CIV p≈0; FeII the
+  anomaly). Two independent analyses → the same ion set: corroboration, not coincidence.
+- AlII's strong negative coupling in the registered run was **centering recoil** (it is the most
+  passive ion, so it recoils hardest against CIV's rise) — which is why CIV−AlII is the largest
+  swing. The "sign-flip near E_H" reading is superseded by this.
+
+**Caveats (post-hoc):** single dataset, active set found by inspection; slopes are extrapolated
+(diseq never exceeds 0.44 — never a real transition); small n per pair. The bootstrap CIs
+excluding zero are real but this is the *hypothesis*, not its confirmation.
+
+**Next registered prediction (writes itself):** *the clean-doublet ions (CIV, MgII) swing in
+abundance/coherence relative to passive ions as a function of local cascade phase* — pre-register
+on **new data (DESI multi-ion absorbers), ideally reaching diseq > 0.7** (a real transition). The
+active-ion set is now independently motivated by the tapestry, so the grouping is non-circular.

foundational/experiments/midnight/journals/2026-06-14_balance-coherence-qbe-synthesis.md

@@ -0,0 +1,199 @@
+# Coherence as Balance: the QBE as the Dynamics of the Connection Primitive
+
+**Date:** 2026-06-14
+**Status:** Working draft (confidence ~0.55) — theory consolidation, no result yet
+**Consolidates:** [2026-06-13 Maximal Coherence as the Connection Primitive](2026-06-13_maximal-coherence-connection-primitive.md)
+(the C functional, null-interval C=1, revivals) + the legacy **Quantum Balance Equation**
+(`legacy-cim-qbe-origins`, `qbe-pac-unification`) + the exp_23 sign-structure finding.
+**Origin:** Peter's reading that robustness *is* symmetry/balance, and that the deep
+phenomenon is the asymmetry between balance at *admission* (emission) and balance at
+*actualization* (absorption / now).
+
+---
+
+## 0. Purpose
+
+The maximal-coherence note proposed that coherence *is* the connection primitive read as a
+magnitude `C`, with `C=1` the undeformed state, decoherence `dC/dτ < 0`, and revivals from
+redistribution. Its central open problem was: **write `C` and give it dynamics.** This note
+supplies the dynamics — they are the **Quantum Balance Equation** — and reframes the whole
+object around a single idea: **coherence is balance, and what we observe is balance frozen at
+admission and read at actualization.**
+
+---
+
+## 1. Thesis in one paragraph
+
+A connection at full strength is two ledgers in **balanced correspondence**. Coherence `C` is
+that balance read as a magnitude: `C=1` is perfect shared balance (symmetry), `C<1` is
+imbalance accrued between the two loci. The Quantum Balance Equation governs how each locus
+relaxes toward its **own local** balance; that local balancing is precisely what erodes the
+**shared** balance that linked the two loci — so **decoherence is the triumph of local balance
+over shared balance**. Things are robust because they are momentarily in balance (symmetric),
+not by nature — and a configuration balanced at the moment of admission (emission) need not be
+balanced at the moment of actualization (absorption). The delta between admission-balance and
+actualization-balance is the deformation that proper time deposits — the **memory** — and it is
+what light carries to us.
+
+---
+
+## 2. The Quantum Balance Equation (restated)
+
+```
+dI/dt + dE/dt = λ · QPL(t),      QPL(t) = Q₀ · e^(−t/τ)
+```
+
+The rate of change of information plus energy equals a coupling term that **decays
+exponentially**. Strict conservation is `dI/dt + dE/dt = 0`; the RHS is the *deviation from
+balance*, relaxing to zero over timescale `τ`. So a system left alone **relaxes toward its own
+balance**. Empirically QPL_damping = 0.02; PAC later derives the characteristic
+**0.020 Hz** through Ξ = γ + ln(φ) (`qbe-pac-unification`). The QBE is the original infodynamic
+balance law; PAC's `P + A + Δ = C` is its first-principles, per-ledger form (conservation =
+balance).
+
+---
+
+## 3. The synthesis — coherence is balance, the QBE is its dynamics
+
+Three identifications turn the QBE into the dynamics the coherence note lacked.
+
+### 3.1 `C` = shared balance
+For two loci `i, j`, let `C(i,j) ∈ [0,1]` measure how much their ledgers are in **balanced
+correspondence** — `C=1` when their potential/actual books mirror each other with no net delta.
+This is "coherence is the connection magnitude" (the prior note) restated as balance: a
+connection at full strength *is* two ledgers in symmetric balance.
+
+### 3.2 The QBE relaxes each locus to its **own** balance
+Each locus independently obeys the QBE: its imbalance source `QPL` decays, so it equilibrates
+toward its *local* balance over `τ`. This is the same move SEC makes — collapse relative to the
+**local** gradient (M13), the locality principle that rescued exp_23.
+
+### 3.3 Decoherence = local balance defeating shared balance (the key step)
+Here is the reconciliation of an apparent paradox: the QBE says systems *relax toward* balance,
+yet coherence *decays*. Both are true because they are about **different balances**. As `i` and
+`j` each pursue their **own local** balance, they drift out of correspondence with **each
+other** — the shared/global balance that linked them erodes. So:
+
+> **Decoherence is each ledger equilibrating to its local balance, at the cost of the shared
+> balance.** `C` falls not because balance is lost but because balance becomes *local* instead
+> of *shared*.
+
+The decay rate is the QBE's `1/τ`; the characteristic coherence frequency is the Ξ-derived
+**0.020 Hz**. This gives `C(τ)` a concrete form (exponential envelope at the balance frequency)
+— the prior note's missing dynamics.
+
+This is also the deepest statement of the **globalism→locality** lesson that has run through all
+of Midnight: the universe's default motion is *toward local balance*, which is why global/
+coordinate quantities decohere (die) and local/relational ones persist (survive). Decoherence
+*is* localization.
+
+---
+
+## 4. The three regimes, from balance
+
+The prior note's discriminator (one `C` must yield null-link, decoherence, revival). Each now
+follows from balance:
+
+- **Null interval → C = 1.** Zero proper time ⇒ no local relaxation has accrued between
+  emission and absorption ⇒ the two endpoints never pursued separate local balances ⇒ shared
+  balance intact ⇒ `C=1`. The photon's endpoints are one balanced connection. (Same conclusion
+  as the prior note, now *because* no local balancing intervened.)
+- **Decoherence → dC/dτ < 0.** Local relaxation (QBE, rate 1/τ) erodes shared balance. The
+  exponential `QPL` decay is the functional form.
+- **Revival → dC/dτ > 0.** The QBE is a **balance** (a signed sum), not a monotone; when the
+  geometry brings two loci's local balances back into correspondence, shared balance *re-forms*.
+  Revivals require no extra postulate — re-balancing is allowed by the same law. (The prior
+  note's "redistribution permits revivals" is just balance restoring.)
+
+---
+
+## 5. The admission ↔ actualization asymmetry (the core phenomenon)
+
+PAC's two moments: **admission** (potential declared / emission) and **actualization**
+(collapse / absorption / measurement / now). The QBE runs between them. A configuration may be:
+
+- **balanced at admission, imbalanced at actualization** — it has relaxed locally in transit;
+  we read a decohered version of a once-coherent emission;
+- **imbalanced at admission, balanced at actualization** — it has *found* balance since; we read
+  coherence that wasn't there at emission.
+
+So **robustness is not a property of a thing — it is a property of a moment.** A robust
+(symmetric) configuration now may have been unstable at admission; an unstable-looking ancient
+signal may be balanced now. **The asymmetry `Δ_balance = balance(actualization) −
+balance(admission)` is exactly the deformation proper time deposits — the memory** — and it is
+`1 − C` read from the receiver's frame (M13 parallax; M15: the observer supplies the frame /
+metric representative). Light is archaeology because it freezes the *admission* balance and we
+necessarily read it at *actualization*.
+
+---
+
+## 6. exp_23 through balance
+
+The within-scope coupling result reads cleanly as balance:
+
+- **Robust contrasts (CIV/AlII, CI excludes 0, 92% of scopes)** = configurations in shared
+  balance *at the epoch observed* — robust because symmetric *now*, which (per §5) does not mean
+  they always were.
+- **The coupling sign flips across IP near E_H ≈ 27 eV** = the **balance inversion point**.
+  Below it the within-scope ledger tilts one way, above it the other. That the inversion sits at
+  the **reservoir quantum E_H** — the same Ξ-scale where the QBE↔PAC bridge lives — suggests
+  **E_H is the balance energy**: the scale at which admission and actualization balance meet.
+- **The MgII anomaly (positive coupling at low IP, jackknife-stable)** = an ion whose
+  admission/actualization balance is offset by its multiplet structure — it reads "wrong-signed"
+  because its local relaxation differs.
+
+The cascade-phase dependence (p=0.049) is then the *shared* balance shifting as the cascade
+clock advances — the QBE running, observed.
+
+---
+
+## 7. The testable edge
+
+If `1 − C` (the admission/actualization delta, the memory) accrues via the QBE, it must
+**scale with proper time** — i.e. with **lookback**. Concretely:
+
+> **Older light carries systematically more admission/actualization offset.** The deformation
+> `Δ_balance(z)` should grow with redshift, at a rate set by the Ξ-derived balance frequency
+> (~0.020 Hz / the cascade clock).
+
+exp_23 was a single-epoch snapshot, so it sees the balance *state* (the E_H inversion now) but
+not its *accrual*. The accrual is the deep, falsifiable form of photon archaeology and is what a
+**multi-epoch baseline (DESI DR1)** can test: does the coherence deficit — and the sign-flip
+energy — drift with lookback as the φ-cascade balance predicts? This is the natural next
+*registered* experiment, distinct from this consolidation.
+
+---
+
+## 8. Open problems (honest)
+
+- **`C` still not written in closed form.** Two ingredients are now in hand — the M14
+  orbit-Hilbert-space overlap (prior note's candidate for the *static* functional) and the QBE
+  (this note's *dynamics*). Uniting them — showing the orbit-overlap evolves under the QBE as a
+  balance — is the central piece of work.
+- **Phase.** `C` must be complex (decoherence kills the off-diagonal *magnitude*, interference
+  lives in the *phase*); the QBE as written is real (I, E rates). The phase must come from SEC
+  complexification (M14, A₁→SL(2,ℂ)). Showing QBE + complexification ⇒ complex `C` is open.
+- **The 0.020 Hz attractor is not fully derived** (naive Ξ gives 0.037 Hz; 0.020 is an
+  attractor whose mechanism is open — `qbe-pac-unification`). If 0.020 Hz is the coherence
+  frequency, this gap propagates here.
+- **"Shared balance" needs an operational measure** that reduces to the orbit-overlap and obeys
+  the QBE — currently a concept, not yet a metric.
+
+---
+
+## 9. Placement
+
+This sits at the M12 (connection primitive) ↔ M13 (proper time as deformation rate) ↔ M14
+(orbit Hilbert space) seam, with the **QBE** (legacy infodynamics) supplying the dynamics and
+**PAC** supplying the conservation. It does not introduce a new primitive; it identifies
+coherence with balance and hands the coherence functional its equation of motion. The immediate
+forward actions: (1) write `C` as orbit-overlap evolving under the QBE; (2) register the
+lookback-scaling test (§7) for DESI.
+
+```
+QBE (balance law)  →  PAC (per-ledger conservation)
+      └→ C = shared balance (this note + the maximal-coherence note)
+            └→ decoherence = local balance defeating shared balance (QBE relaxation)
+                  └→ admission↔actualization asymmetry = memory = 1−C (M13 parallax, M15 frame)
+                        └→ exp_23: E_H = balance energy; next: lookback-scaling test (DESI)
+```

foundational/experiments/midnight/scripts/explore_contrast.py

@@ -0,0 +1,93 @@
+"""
+EXPLORATORY / post-hoc (context for the next registered prediction) — the FIXED exp_23 observable.
+
+exp_23 (registered f1639e08, INCONCLUSIVE) tested the MEAN-CENTERED per-ion coupling for a
+MONOTONE IP-ordering. The diagnostic (explore_inconclusiveness.py) showed: the structure is an
+ENDPOINT CONTRAST (flat plateau + opposite-signed IP extremes), the flat middle is a
+mean-centering shadow, and the real signal is the HIGH-IP vs LOW-IP coupling contrast vs local
+cascade phase, measured CENTERING-FREE.
+
+The fixed observable (principled, non-circular pivot = the Hartree reservoir quantum E_H):
+  HIGH = ions with IP > E_H (27.2 eV)     -- above the reservoir quantum (active)
+  LOW  = ions with IP < E_H               -- below (settled)
+  per scope:  contrast = mean(logN_HIGH) - mean(logN_LOW)   [within-frame difference: centering-free]
+  test: does the contrast grow with local cascade phase diseq?
+
+Balance reading: the contrast is the high/low IP imbalance; the prediction is that cascade
+transitions DRIVE that imbalance (the middle is the balance pivot, the extremes swing apart).
+
+NOT registered. Post-hoc context to design the next pre-registered test (DESI / new data).
+"""
+
+import numpy as np
+from scipy.stats import spearmanr, mannwhitneyu
+from exp_23_joint_coupling import load_xqr30_scopes
+from exp_19_ionization_coupling import ION_IP, n_at_z
+
+E_H = 27.2
+def diseq(N): return max(0.0, 1.0 - 2.0 * abs(N - round(N)))
+rng = np.random.RandomState(20260614)
+scopes = load_xqr30_scopes()
+
+
+def boot_slope(D, Y, nb=5000):
+    s = float(np.polyfit(D, Y, 1)[0])
+    bs = []
+    for _ in range(nb):
+        idx = rng.randint(0, len(D), len(D))
+        if len(np.unique(np.round(D[idx], 3))) >= 2:
+            bs.append(np.polyfit(D[idx], Y[idx], 1)[0])
+    return s, float(np.percentile(bs, 2.5)), float(np.percentile(bs, 97.5))
+
+
+def contrast(exclude=()):
+    D, Y = [], []
+    for d in scopes.values():
+        hi = [v for i, v in d['ions'].items() if ION_IP[i] > E_H and i not in exclude]
+        lo = [v for i, v in d['ions'].items() if ION_IP[i] < E_H and i not in exclude]
+        if hi and lo:
+            D.append(diseq(n_at_z(d['z']))); Y.append(np.mean(hi) - np.mean(lo))
+    return np.array(D), np.array(Y)
+
+
+print("=" * 68)
+print("FIXED observable: HIGH(IP>E_H) - LOW(IP<E_H) logN contrast vs cascade phase")
+print("=" * 68)
+
+for label, excl in [("all ions", ()), ("excl. MgII (confirmed anomaly)", ('MgII',))]:
+    D, Y = contrast(exclude=excl)
+    s, a, b = boot_slope(D, Y)
+    rho, p = spearmanr(D, Y)
+    # transition-vs-trough (avoids slope extrapolation; tapestry-style 2-sample)
+    hiD, loD = Y[D > 0.3], Y[D <= 0.3]
+    U, pu = mannwhitneyu(hiD, loD, alternative='greater')
+    print(f"\n[{label}]  n={len(D)} scopes")
+    print(f"  slope vs diseq = {s:+.3f}  CI95=[{a:+.3f},{b:+.3f}]"
+          f"{'  ** excludes 0' if a > 0 or b < 0 else ''}")
+    print(f"  Spearman(diseq, contrast) rho={rho:+.3f}  p={p:.4f}"
+          f"{'  ** p<0.05' if p < 0.05 else ''}")
+    print(f"  toward-transition (diseq>0.3, n={len(hiD)}) median={np.median(hiD):+.2f}  "
+          f"vs toward-trough (n={len(loD)}) median={np.median(loD):+.2f}  "
+          f"MannWhitney(greater) p={pu:.4f}{'  ** p<0.05' if pu < 0.05 else ''}")
+
+# pairwise: which HIGH-LOW pairs carry it (centering-free)
+print("\nPairwise high-low contrasts vs diseq (slope, CI95):")
+HIGH = [i for i in ION_IP if ION_IP[i] > E_H]
+LOW = [i for i in ION_IP if ION_IP[i] < E_H]
+rows = []
+for h in HIGH:
+    for l in LOW:
+        D, Y = [], []
+        for d in scopes.values():
+            if h in d['ions'] and l in d['ions']:
+                D.append(diseq(n_at_z(d['z']))); Y.append(d['ions'][h] - d['ions'][l])
+        D, Y = np.array(D), np.array(Y)
+        if len(D) >= 8 and len(np.unique(np.round(D, 3))) >= 4:
+            s, a, b = boot_slope(D, Y)
+            rows.append((h, l, len(D), s, a, b, a > 0 or b < 0))
+for h, l, n, s, a, b, sig in sorted(rows, key=lambda r: -r[3]):
+    print(f"  {h:>4}-{l:<4} n={n:>2}  slope={s:+.3f}  CI95=[{a:+.3f},{b:+.3f}]"
+          f"{'  **' if sig else ''}")
+
+print("\n(Exploratory/post-hoc. The significant centering-free contrast is the observable to")
+print(" PRE-REGISTER on new/held-out data — DESI multi-ion absorbers, ideally reaching diseq>0.7.)")