Direct vs. Inferred Observables

Almost nothing in published collider results is directly observed. The detector records a small universal set of primitives — track hits, calorimeter energy deposits, magnetic-field curvature, timing, polarization — and everything else (cross sections, masses, branching ratios, couplings, $sin^{2} θ_{W}$ , CKM elements, decay rates) is a fit-extracted parameter inferred from those primitives via a theoretical model.

This page collects the taxonomy: what is directly measured vs. what is inferred, ordered by the inference distance between raw data and quoted result. It is the conceptual companion to:

cross-sections.md — the master $d σ$ formula whose parameters are extracted by the fits described here.
decay-rates.md — the master $d Γ$ formula, same structure.
collider-measurements.md — the full theory → variable → reconstruction → result pipeline for eight concrete worked examples.

1. The Direct Observables

These are what a particle-physics detector actually records. Every other number in this folder is derived from these:

Primitive	Physical signal	Detector subsystem
Track hits	Position + time of charged-particle ionization	Silicon pixel + strip tracker
Calorimeter energy deposits	Total ionization in absorber, summed across a particle shower	ECAL ( $e^{\pm}, γ$ ); HCAL (hadrons)
Magnetic-field curvature	Bending radius of charged-particle track in known $B$ -field → particle momentum	Tracker (in solenoidal $B$ -field, $\sim 2 - 4 T$ )
Timing	Bunch-crossing timestamp; time-of-flight; decay-vertex displacement	Timing layers, trigger
Polarization (selected experiments)	Spin-precession frequency in storage rings, or decay-product angular distribution	E.g. resonant depolarization (LEP), muon $g - 2$ ring (Fermilab)

That's it. All five together fit in a single short table. Everything else in particle physics is inference from these.

Two ancillary direct measurements complete the picture:

Primitive	Physical signal
Luminosity $L$	Beam-overlap measurement via van der Meer scan (vertical/horizontal beam separation vs. rate); cross-checked against forward elastic scattering (TOTEM, LHCf). Per-experiment uncertainty $\sim 2%$ .
Beam energy $E_{beam}$	At $e^{+} e^{-}$ : resonant depolarization (to $1 0^{- 5}$ at LEP). At hadron colliders: from accelerator RF + lattice optics, with auxiliary calibration.

These two — luminosity and beam energy — are directly measured external quantities. Every other "measurement" is internal to the theory model.

2. The Inferred Observables

Every quantity below is a parameter whose value is obtained by:

Building a theoretical model that depends on the parameter + a long list of nuisance parameters.
Predicting a histogram of some kinematic variable from the model.
Maximizing the likelihood that the observed histogram of detector primitives matches the model prediction, while profiling out the nuisances.

The result is the best-fit value of the parameter, with (stat) ⊕ (calibration / detector) ⊕ (theory / PDF / scheme) uncertainties.

2.1 Cross sections, decay rates, and their derived ratios (Family A)

Observable	Inferred from	Master formula
Total cross section $σ$	$N_{events} / (L \cdot A \cdot ϵ)$ — count events, divide by luminosity, acceptance, efficiency	$d σ = (1/ F) \overline{∥ M ∥^{2}} d Π_{n}$ (cross-sections.md)
Differential cross section $d σ / d X$	Same as above, binned in $X$	same
Decay rate / width $Γ$	Exponential fit of decay-time distribution $N (t) = N_{0} e^{- Γ t /ℏ}$ , or Breit–Wigner peak width	$d Γ = (1/2 M) \overline{∥ M ∥^{2}} d Π_{n}$ (decay-rates.md)
Lifetime $τ = ℏ/Γ$	Same; sometimes measured directly from displaced-vertex statistics	same
Branching ratio $BR (X \to f)$	$Γ (X \to f) / Γ_{total}$ — ratio of fits	(ratio)
Asymmetries $A_{FB}, A_{L R}, A_{CP}$	$(σ_{F} - σ_{B}) / (σ_{F} + σ_{B})$ — counts in two regions	(ratio of cross sections)

2.2 Particle masses (Family B — propagator poles)

Observable	Inferred from	Inference distance
$m_{Z}$	Breit–Wigner peak position in $σ (s)$ scan at LEP1	Short — pole of $σ (s)$
$m_{h}$	Gaussian-broadened peak position in $m_{γγ}, m_{4 ℓ}$ invariant-mass histograms	Short — invariant-mass peak
$m_{W}$	Jacobian peak in transverse-mass $d σ / d m_{T}$ template fit	Long — needs PDFs, $W$ recoil model, FSR, $Z$ calibration
$m_{t}$	Endpoint / peak of $m_{b ℓ}, m_{bjj}$ in $t \overset{ˉ}{t}$ events; or template fit	Long — needs jet-energy scale, $b$ -tag calibration, color reconnection, scheme translation
Lattice hadron masses ( $m_{p}, m_{n}, m_{π}, \dots$ )	Exponential falloff of Euclidean two-point correlator $C (t) \sim e^{- m t}$	Short (within lattice setup)
Quark masses ( $m_{u}, m_{d}, m_{c}, m_{b}, m_{t}$ )	Combined inputs: spectroscopy + lattice + perturbative matching	Long + scheme-dependent

2.3 Coupling constants and mixing angles (Family B + global fits)

Observable	Inferred from
$α_{em} (M_{Z})$	Combined fit to QED observables; runs from low-energy $α = 1/137$ via hadronic vacuum polarization
$α_{s} (μ)$	Multiple QCD observables (jet rates, $R$ -ratio, lattice, $τ$ decay, DIS) fit together
$sin^{2} θ_{W}$	$Z$ -pole asymmetries (LEP/SLC) + parity-violating $e^{-} e^{-}$ scattering + atomic PV
$G_{F}$	Muon lifetime $τ_{μ}$ via leptonic three-body decay formula
CKM elements $∥ V_{ij} ∥$	Semileptonic decay rates × lattice form factors (overdetermined, fit together)
$δ_{CKM}$ , $sin 2 β$	Time-dependent CP asymmetries in $B$ -meson decays
PMNS angles + $δ_{CP}$	Neutrino-oscillation rates $P (ν_{α} \to ν_{β}; L / E)$

2.4 Static and vertex observables (Family C)

Observable	Inferred from
Anomalous moments $a_{e}, a_{μ} = (g - 2) /2$	Spin-precession frequency in Penning trap (electron) or storage ring (muon)
Charge radii $⟨ r^{2} ⟩$	Low- $Q^{2}$ extrapolation of elastic-scattering form factors; or atomic spectroscopy
EDMs $d_{e}, d_{n}$	Atomic-physics interferometry (ACME, JILA); upper limits in SM, BSM if positive
Form factors $F_{1} (Q^{2}), F_{2} (Q^{2})$	Multiple-energy scattering measurements stitched into a function
Axial couplings $g_{A}$	$β$ -decay angular correlations

2.5 Oscillation parameters (mixing)

Observable	Inferred from
$Δ m_{d}, Δ m_{s}$ ( $B$ -meson mass differences)	Oscillation frequency in flavor-tagged time-dependent $B$ decays
$Δ m_{12}^{2}, Δ m_{23}^{2}$ (neutrino mass splittings)	Oscillation pattern $P (ν_{α} \to ν_{β}) \propto sin^{2} (Δ m^{2} L /4 E)$ vs. $L / E$
$∣ ϵ_{K} ∣, ∣ ϵ^{'} / ϵ ∣$	Time-dependent and direct CP-violating decay-rate ratios in $K$ -meson decays

2.6 Composite / global-fit parameters

Observable	Inferred from
PDFs $f_{i} (x, μ_{F})$	Global fit to $\sim 4, 000$ data points across DIS + DY + jets + top + $W / Z$
CKM unitarity-triangle apex $(\overset{ρ}{ˉ}, \overset{η}{ˉ})$	Combined fit to all CKM-related measurements (standard-model.md §1)
Global EW fit predictions ( $m_{W}, sin^{2} θ_{W}$ from inputs)	Profile-likelihood over all $Z$ -pole + low-energy EW measurements
Higgs coupling modifiers $κ_{i}$	Profile fit over all measured Higgs production × decay channels

3. Inference Distance: Short → Long

The chain from primitives to parameter has very different lengths depending on the observable. This is the most operationally important distinction across collider measurements — it determines whether the published uncertainty is dominated by data statistics or by theory modeling.

Distance	Example	Why
Zero	Event count $N$ in a histogram bin	The universal base level — every collider measurement starts here. Counts in bins of some reconstructed variable, summed over a run. Everything below is one or more steps removed from this.
Trivial	Charge of a track ( $\pm 1$ from curvature direction)	Sign-only inference
Trivial	Asymmetry $A_{FB} = (N_{F} - N_{B}) / (N_{F} + N_{B})$	Pure ratio of two histogram-bin counts; luminosity + most efficiencies cancel
Trivial	Branching ratio $BR (X \to f) = N_{X \to f} / N_{X \to anything}$	Pure ratio; luminosity and total cross section cancel
Short	$m_{Z}$ from $σ (s)$ line-shape scan	Single dominant feature (pole position); only beam-energy calibration enters
Short	$m_{h}$ from $m_{γγ}, m_{4 ℓ}$ peak	Narrow Gaussian peak; only lepton/photon energy scale
Short	Hadron masses from lattice-QCD Euclidean correlators	Exponential falloff; only statistical noise + lattice-spacing extrapolation
Short	Lifetime $τ$ from exponential decay-time fit	Slope of $lo g N (t)$ ; needs no cross-section quantity
Medium	Neutrino $Δ m^{2}$ from oscillation pattern	Frequency in $L / E$ ; need calibrated $L, E$ + flux normalization
Medium	Total cross section $σ_{tot} = N / (L A ϵ)$	Counting + luminosity + acceptance — but the shape of the histogram is irrelevant
Long	$m_{W}$ from transverse-mass templates	Needs PDFs, $W$ recoil model, FSR, $Z$ calibration
Long	$m_{t}$ from $t \overset{ˉ}{t}$ kinematic fits	Needs jet-energy scale, $b$ -tag calibration, MC color reconnection, scheme-translation theory
Long	$α_{s} (M_{Z})$ from jet observables	Many NLO/NNLO QCD corrections, PDFs, non-perturbative effects
Very long	Higgs self-coupling $λ$	Requires di-Higgs cross section + production modeling; uncertainty $≳ 50%$ even at HL-LHC

The zero-row is worth pausing on: all collider measurements share the same base-level quantity — counts in histogram bins. Cross sections, branching ratios, asymmetries, masses, lifetimes, couplings, and discovery significances are different functionals of the same primitive. Cross sections are one common such functional — but branching ratios, asymmetries, and lifetimes are extracted without ever forming a cross-section quantity. So "everything at colliders relies on cross sections" is too strong; the correct statement is "everything at colliders relies on histograms of event counts, of which the cross section is the most familiar consumer".

In the long-inference cases (notably $m_{W}$ and $m_{t}$ ) the theoretical-modeling uncertainty competes with or exceeds the statistical uncertainty in the final published number. This is why theory progress (better PDFs, higher-order calculations) reduces published uncertainties even without new data.

4. The Renormalization-Scheme Wrinkle

For all mass and coupling observables, the inference does not stop at "best-fit parameter value" — even the meaning of the parameter depends on a renormalization-scheme choice. This is a separate inference step on top of the experimental fit.

4.1 The "which mass?" question

Quantity	Scheme	Definition
$m_{W}^{OS}$	On-shell	Real part of $W$ propagator pole
$m_{W}^{\overline{MS}} (μ)$	$\overline{MS}$	Mass parameter in $\overline{MS}$ -renormalized Lagrangian, depends on scale $μ$
$m_{W}^{pole}$	Complex-pole	$Re s_{0}$ where $s_{0} = m_{W}^{2} - i m_{W} Γ_{W}$
$m_{t}^{pole}$	On-shell	Real part of top propagator pole
$m_{t}^{\overline{MS}} (μ)$	$\overline{MS}$	Running mass
$m_{t}^{MC}$	Monte Carlo	Value of $m_{t}$ parameter inside the event generator (Pythia, Herwig, ...)

Numerical differences:

Comparison	Difference
$m_{W}^{OS}$ vs. $m_{W}^{pole}$	$\sim 26 MeV$
$m_{t}^{pole}$ vs. $m_{t}^{MC}$	$\sim 500 MeV$ (translation theory uncertainty)
$m_{b}^{pole}$ vs. $m_{b}^{\overline{MS}} (m_{b})$	$\sim 500 MeV$
$m_{c}^{pole}$ vs. $m_{c}^{\overline{MS}} (m_{c})$	$\sim 300 MeV$

All of these are comparable to or larger than current experimental precision. The published value carries an implicit scheme tag, and translating between schemes is a theoretical operation that adds its own uncertainty to the experimental fit.

4.2 The "which coupling?" question

$α_{s}$ runs by orders of magnitude across observable scales. Saying " $α_{s} = 0.118$ " is meaningless without specifying $μ = M_{Z}$ and the scheme ( $\overline{MS}$ for most modern conventions). Similarly:

$sin^{2} θ_{W}^{eff,lep}$ at the $Z$ -pole vs. $sin^{2} θ_{W}^{\overline{MS}} (M_{Z})$ vs. on-shell $sin^{2} θ_{W} = 1 - m_{W}^{2} / m_{Z}^{2}$ — these differ at the $\sim 1 0^{- 3}$ level.
$α (0) = 1/137.036$ vs. $α (M_{Z}) = 1/128.946$ — same quantity, two scales.

5. The Full Chain

Combining the §1 pipeline (collider-measurements.md) with scheme translation gives:

$actually recorded detector primitives \to reconstruction physics objects (tracks, jets, leptons, \neq E_{T}) \to measured distribution X_{reco} histogram \to inference template fit \to parameter in a scheme \hat{θ} (scheme) \to after scheme translation θ (Lagrangian) .$

The only left-end node is data. Every subsequent step is model + inference.

6. Why This Matters

Comparing measurements across experiments requires comparing apples to apples: same scheme, same inputs to nuisance parameters. A $5 σ$ tension may evaporate if one experiment used a different PDF set or a different MC tune.
Quoted uncertainty bands can mislead. The full chain has stat ⊕ syst ⊕ theory + (scheme-translation) — the latter is often not in the published uncertainty.
Theory progress directly reduces experimental uncertainties for long-inference observables. NNLO/NNNLO QCD calculations, better PDF fits, and improved MC generators have shrunk $m_{W}$ and $m_{t}$ error bars without any new data.
What counts as "data-driven" is itself ambiguous. Even the "data-driven" hadronic vacuum polarization in muon $g - 2$ uses lattice QCD as a cross-check; pure first-principles measurement is rare.

7. See Also

Cross Sections — the master $d σ$ formula whose parameters are inferred via the fits described here. See §1.1 for the five equivalent definitions of $σ$ (geometric, single-target, operational, quantum amplitude, S-matrix) and how the empirical chain (counting events in a detector → $σ$ ) ties to the theoretical chain ( $∣ M ∣^{2} \to σ$ ).
Decay Rates — the master $d Γ$ formula, with the muon-lifetime worked example showing how a direct-time-domain measurement of $Γ$ is also an inferred quantity (lifetime fit + theory model).
Collider Measurements — eight worked case studies of the theory → variable → reconstruction → result pipeline; this page is the conceptual filter on top.
Electroweak Observables — the EW-sector inventory, each entry tagged with its inference structure.
Standard Model Observables — cross-sector observables (CKM unitarity triangle, global EW fit) — all long-inference by construction.
Observables — General Map — the structural Class A/B/C classification that all of the above instantiate.

Keyboard shortcuts

youyuanwu