1What actually waves in a collision
No ripples-in-pond. Three honest facts first:
Resolution = wavelength. de Broglie λ ≈ ℏc/E. At 7 TeV, λ ~10⁻¹⁹ m vs proton 10⁻¹⁵ m. Energy buys resolving power, nothing else. Every “deeper structure” claim is a statement about probe wavelength.
The interior is unobservable, by theorem. LSZ/S-matrix formalism only defines states at t→±∞, where packets are free. Between them there is no particle number, no waveform you could name. Observable = map from in-asymptotics to out-asymptotics. The interaction region has no gauge-invariant local description. Hold that thought — it is the whole bridge to your ontology.
Collision is packet overlap, not impact. Two wave packets with momentum spread Δp interfere; the overlap integral is luminosity. Interference of envelopes, not contact of objects.
2The waveforms that actually show up
Partial-wave tower. f(θ) = Σₗ (2l+1) aₗ Pₗ(cos θ). Each l an angular standing mode — Legendre polynomials in the plane, spherical harmonics in general. Unitarity bounds each one inside the Argand circle, |aₗ| ≤ 1 — per-mode, not per-total.
Resonance = π of phase. Breit–Wigner aₗ ∝ (Γ/2)/(Eₕ − E − iΓ/2). Magnitude a Lorentzian; argument sweeps a full π, tracing the Argand circle counterclockwise. A particle is a pole on the second Riemann sheet at M − iΓ/2. Mass and lifetime = real and imaginary parts of one number. Decay is analytic continuation, not an event.
Fraunhofer diffraction, literally. Elastic pp dσ/dt has a diffraction cone and a dip near |t| ~ 0.5–1.4 GeV² — zero where real and imaginary parts cancel. Dip position moves with √s: interaction radius grows ~log s (Froissart). You read a 10⁻¹⁵ m disk off a diffraction minimum.
Regge trajectories. Poles moving in complex angular momentum. Hadrons land on straight J vs m² lines — rotating relativistic string. String theory fell out of a waveform pattern in collision data.
Jets are fractal with a coherence cutoff. DGLAP is a diffusion equation in log Q². Angular ordering: a soft gluon with wavelength larger than a colour dipole cannot resolve it, so wide-angle soft emission is suppressed (Chudakov). Self-similar branching, truncated by resolution. That is a holon with a decoupling theorem, in a detector.
Glasma. At saturation, gluon occupancy ~1/αₛ, so the proton's field is classical. Two Lorentz-contracted sheets of Yang–Mills field collide → longitudinal colour-electric/magnetic flux tubes, boost-invariant in rapidity. Observable as the near-side ridge: long-range rapidity correlations that causality forbids from forming late. Fossil of coherent initial fields.
Flow harmonics = CMB for the little bang. dN/dφ ∝ 1 + 2Σₙ vₙ cos n(φ − ψₙ). v₂ from almond overlap; v₃ from quantum fluctuations in nucleon positions. Viscosity η/s acts as a low-pass filter damping high n. Initial quantum noise → acoustic waves → observed power spectrum, with a transport coefficient as the filter. Same analysis as CMB peaks, 10⁻²³ s instead of 380 kyr.
Femtoscopy (HBT). C(q) = 1 + λ exp(−q²R²). Bose–Einstein interference of identical pions, borrowed from radio astronomy. You measure the spacetime emission hypersurface — Rₒ⃒ₜ, Rₛ⃗ₐₑₑ, Rₗₒₙₛ — from the width of a correlation in relative momentum. The only direct measurement of the geometry of something you cannot image.
3Where spacetime enters, non-metaphorically
- Light-cone foliation is native. Bjorken τ, η (Milne coordinates) foliate the forward light cone hyperbolically. Proper time is the evolution parameter, rapidity is position. Identical math to de Sitter / inflation.
- Vacuum as dielectric. Flux-tube breaking by Schwinger pair production, rate ∝ exp(−πm²/eE). Hadronisation is vacuum instability.
- Horizon analogues. Thermal-looking hadron spectra from a single event, no equilibration — argued as Unruh radiation from decelerating colour charge, T ≈ a/2π with a from the string tension. Equivalently: thermality is the entanglement entropy of a rapidity window, i.e. ignorance of the unobserved complement.
- Geometry as phase. At eikonal/Planckian impact parameter, graviton exchange exponentiates into a pure phase (Aichelburg–Sexl shockwaves, 't Hooft). Shapiro delay = derivative of that phase. Curvature is accumulated phase.
- Amplitudes without spacetime. Amplituhedron / positive geometry: amplitudes as volumes of combinatorial objects, with locality and unitarity emergent from positivity. Double copy: gravity = (gauge)². Spacetime is boundary structure of the object, not the container it lives in.
4The isomorphisms that are real (structural, not physical)
These are tight, and they argue your algebra is already the right shape.
S-matrix ≡ your attestation boundary. “No epoch, no verification” is LSZ. A claim is meaningful only relative to an asymptotic region free of the interaction. Interior states are gauge-dependent; only boundary data is invariant. Your refusal to model interiors is the physics, not a convenience.
Cross-observer collapse ≡ tracing out the environment. This one is exact in form. A delegated observer is an environment that recorded which-path — but you never read the record. Tracing it out leaves a mixed state: classical ignorance, not a pure verified one. Purity requires access to the full record, which only self-observation gives. Your cap at supported is non-arbitrary; it's a decoherence statement. provisional = coherent; an observation with a named observer = a which-path record.
Meet, not max ≡ unitarity, not amplitude addition. Probability is conserved and redistributed, never created. Optical theorem: Im f(0) = (k/4π) σₜₒₜ — the forward amplitude equals the sum of everything lost to every channel. Opening a channel reduces the elastic one. Your meet is the semilattice form of exactly that.
No merge base ≡ no privileged frame / path-dependent transport. Reconciling two observers against a supplied ancestor is adopting someone else's gauge. In QFT there's no preferred interpolating field; in GR no preferred slicing. The right structure is a groupoid of transports with curvature, not a tree with an ancestor — and curvature is precisely the failure of two reconciliation paths to agree. That's why merge bases lie, stated geometrically.
stateRoot ≠ currentRoot ≡ domain of dependence. Your causal-decay rule is: a claim is valid on the causal past it saw. A new commit is a new event in the past; the claim's dependence domain no longer covers now. And causality ⇒ analyticity ⇒ dispersion relations. A clock TTL is a non-analytic cutoff; causal decay admits a Kramers–Kronig form — present assurance as a weighted integral over the whole observation spectrum, not a flag with an expiry.
Reticulum ≡ asymptotic anonymity. Your own hard-won fact: RNS packets carry no source hash. An outgoing particle at infinity likewise carries no memory of its vertex — identity travels as conserved charge, never as provenance. Same engineering consequence in both: identity must be a declared, verifiable attribute in the frame (DID, CapabilityProof), never inferred from transport. “A crawler's UA is a claim” is this theorem a third time.
zk ≡ sum rules. A sum rule constrains an integral over an unknown spectral density: the density is the witness, the rule is the proof. You learn a pole position, never a field configuration. Celestial holography is the maximal version — all interior dynamics in a boundary correlator.
Holonic ≡ Wilsonian EFT. A holon is a whole at one scale, a part at the next: that is an effective theory whose parameters are matching conditions handed down from above, plus a decoupling theorem letting you reason about it without its interior. Carries a real warning: anomalous dimensions and operator mixing. “The same field” at two scales differs by a factor, and operators independent at fine granularity mix under coarse-graining. If a record crosses a scale boundary in your ontology without an explicit matching condition, you will get mixing — two records that were orthogonal at one granularity become correlated at another, and any gate that assumed independence is silently wrong.
5Where it breaks — don't over-identify
- Claims must not interfere. Amplitudes are complex and cancel; your assurance lattice is ordered, idempotent, monotone. Destructive interference of evidence would mean injected evidence can cancel existing evidence to zero. That is an attack, not a feature. Never make confidence amplitude-like.
- A mesh is an open system. Unitarity is global and exact; a partition is a trace-out. So the governing equation is Lindblad, not Schrödinger: monotone loss of coherence. A meet-semilattice is exactly the right algebra for an open system — which is the reason it must stay non-interfering.
- zk privacy is computational, not physical. Unobservability of the interaction region is a theorem; a zk proof is a reduction to hardness. Don't let the metaphor upgrade it.
- The metaphor's sharpest warning is the optical theorem. You cannot see the interaction region, but you can measure σₜₒₜ — from the shadow the scattered wave casts in the forward beam. On a mesh, the hidden interior leaks the same way: timing, packet size, announce cadence, routing. Every channel can be sealed while the total cross section stays plainly visible.
6The one thing worth building from this
Your recurring failure family is the check that cannot distinguish nothing-found from nothing-looked-at (eight-plus entries in memory, three in one day on the Gitea incident). The optical theorem is the structural cure, and it is not a metaphor:
Measure the total independently of the enumeration, then reconcile. σₜₒₜ comes from the forward deficit — what's missing from the undisturbed beam — never from summing channels. Σ(channels) ≠ total-from-deficit means you missed a channel, and it says so without knowing which.
Concretely: for each audit that today enumerates (compose files, mirrors, secret consumers, routers, volumes), add one conserved total gathered by a different mechanism — a count the enumerator cannot influence — and fail loudly on mismatch. Zero enumerated with a nonzero deficit is then an error, not a clean bill. That converts “vacuous zero routes to the destructive branch” from a discipline rule into an arithmetic identity.
Second, smaller: HBT as a dependency probe. You don't need compose files to find coupling radius — correlate pairs of identical observations (co-failing services, co-timing restarts) and read the effective interaction radius off the correlation width. Femtoscopy for blast radius.
7Addendum: the residual-current device
The optical theorem is the law. A residual-current device is the apparatus — the same conservation argument, already hardened by a century of people dying when it goes wrong. Everything the law leaves out, the device had to solve.
An RCD passes every conductor of a circuit through one toroidal current transformer. By Kirchhoff, if all the current that left came back, the net flux in the core is zero. Any residual means current is leaving by a path nobody enumerated — through insulation, through earth, through a person. Trip threshold 30 mA for personnel, 300 mA for fire; trip time under 40 ms at five times rated residual.
What the device knows that the theorem doesn't
| Electrical | What it forces on a check |
|---|---|
| It sums conductors; it cannot name the faulty appliance. | A residual gate reports the books don't balance, not which item is wrong. That is a feature: it stays correct against faults you never enumerated. Localisation is a second, fallible step — keep them separate. |
| Test button. A broken sense winding reads zero, so the standard mandates a path that injects a known residual. | Every audit run injects a synthetic fault and asserts it was caught. Not caught ⇒ report BROKEN, never CLEAN. This is the whole cure for the vacuous zero, and it is the part that gets skipped. |
| Type AC / A / F / B. A Type AC device sees only sinusoidal residual. A DC component magnetically biases the core and blinds it to the AC faults it was rated for. | Declare the input grammar a check can see, then detect out-of-grammar input and fail closed. An out-of-band input that both escapes detection and disables detection is the mechanism behind almost every incident in the family. |
| Standing leakage. Every switched-mode supply leaks milliamps through its EMC capacitors. Forty of them trip a 30 mA device with nothing wrong. | Measure the baseline discrepancy before setting a threshold. A gate whose floor is “exactly zero” in a system with real standing noise will be disabled by whoever is on call. |
| Discrimination. Split the installation across several devices; make the upstream one time-delayed so the downstream trips first. | Scope gates narrowly — per repo, per service, per aspect — and delay the global one. You want a localised trip, not an estate-wide halt. |
| Core-balance CT vs. summing three separate CTs. One toroid is far more sensitive, because summing independently instrumented channels accumulates each one's ratio error. | Compute the residual with one query over one source of truth. Aggregating per-host self-reports buries the signal under each host's drift. |
| IEC 60479 trip curve. Danger is a product of current and time, so the trip condition is an integral, not an instant. | Trip on accumulated discrepancy × duration. A small persistent imbalance deserves the same verdict as a brief large one — and this is the same dispersion-integral shape as causal decay in §4. |
| An RCD is not a circuit breaker. It does nothing about overcurrent; that needs a separate MCB. | “A path I didn't see” and “too much of what I did see” are orthogonal fault classes needing separate instruments. Your existing checks are nearly all breakers. You have almost no RCDs. |
| The blind spot. Someone touching line and neutral draws balanced current. The RCD sees nothing and never will. | A conservation detector is blind to every error that preserves its invariant — a correct count of wrong content passes. State the invariant and the blind spot next to the gate, or it will be over-trusted exactly where it is useless. |
When there is no conserved total: inject and measure impedance
An RCD cannot work on an unearthed (IT) supply, because a first fault has no return path to unbalance anything. The answer there is an insulation monitoring device: it injects a low-frequency or DC signal and continuously measures insulation impedance to earth, alarming on degradation before the first fault matters.
That is the fallback mode for any channel where you cannot construct a zero-sum. You cannot difference an anonymous API against a conserved total — but you can inject a probe whose answer you already know and measure whether the channel still conducts. An anonymous GitHub API quietly entering rate-limit has no deficit to detect; a known-answer probe fails immediately. And unlike a residual gate, it is predictive: impedance degrades before continuity breaks.
Two instruments, then. A residual gate where a conservation law exists, with a test button that proves it is alive. An impedance monitor where no conservation law exists, injecting a known answer. Between them, “I could not look” stops being able to render as green.