{"template":"article","title":"Boundary Factorization with a Partially Recorded Center","authors":[{"name":"Leona V. Haldersen","affiliations":[1],"email":"l.haldersen@ravelin.example","corresponding":true},{"name":"Masaki Tsubehara","affiliations":[2]},{"name":"Nayana Selvaratne","affiliations":[2,3]}],"affiliations":["Department of Mathematical Physics, Ravelin Institute, Bergen, Norway","Centre for Quantum Geometry, Kosenbridge Institute of Science, Sendai, Japan","Department of Theoretical Physics, Haldenmere College, Wellington, New Zealand"],"abstract":"A finite semisimple boundary algebra gives an exact model of the distinction between sewing at fixed central character and averaging over central characters before testing boundary factorization. We derive the joint amplitudes of classical boundary readouts coupled through this center and give a rank bound on the number of sectors compatible with the readouts. In an eight-sector realization the bound is saturated even though each single-boundary marginal agrees with that of the independently prepared system. A charge selection rule controls higher moments and separates the shared preparation from a mixture that agrees on all pair observables. Conditioning on a partial record of the sector lowers the rank of the defect, with full factorization recovered only when the record distinguishes all occupied sectors. We instantiate the readout using a certification procedure for an absent original and distinguish removal of the original from removal of the shared preparation. All results follow from finite sums, including the comparison between empty and partially specified boundary records.","keywords":["topological field theory","factorization","superselection sectors","boundary algebras"],"body":"\\newcommand{\\EE}{\\mathbb{E}}\n\\newcommand{\\rank}{\\operatorname{rank}}\n\\newcommand{\\Cov}{\\operatorname{Cov}}\n\n# Introduction {#sec:intro}\n\nThe product of partition functions on two boundaries in a specified theory and the mean of that product over theories are different objects [@giddings1988; @coleman1988; @marolf2020]. In JT gravity the difference has a geometrical interpretation through the double trumpet in the matrix integral [@saad2019]. The Lorentzian obstruction involving the bulk Hilbert space [@harlow2020] raises a related question with different observables. Euclidean examples with dynamical metrics depend on which data are integrated over in common [@santos2021; @blommaert2022].\n\nGiven a joint readout at two boundaries, how many sectors must a conditionally independent description have? We give an exact bound and calculate what remains of the factorization defect when a partial record of the sector is available. A two-dimensional topological field theory identifies the sewn amplitudes that reproduce these moments. The normalization is specified in @sec:algebra.\n\nThe example came from a determination that two intake points could independently certify the nonreceipt of a requisition [@allende2024]. Each used a complete local record. Instructions prohibited either point from seeing the other's copy. The absence of a submission was taken to exclude any common input. The registration mark on the certificate stock had, however, been impressed before either point opened. The requisition was never made.\n\n# The circle algebra and the boundary state {#sec:algebra}\n\nLet $Q=\\mathbb{Z}/q\\mathbb{Z}$, with $q\\geq 2$. Take the oriented circle algebra to be $\\mathcal A=\\mathbb C^Q$ with pointwise multiplication and identity $1$. Write $e_a$ for the idempotent supported on $a\\in Q$. Our Frobenius trace and the positive form it induces are specified in @eq:trace. This is the semisimple case of the oriented bordism construction [@atiyah1988; @abrams1996; @kock2004].\n\n$$\n\\tau(e_a)=q^{-1},\\qquad\n(f,g)_\\tau=\\tau(\\overline f g).\n\\label{eq:trace}\n$$\n\nThe orthonormal charge basis $\\chi_m(a)=\\exp(2\\pi i m a/q)$ identifies $\\mathcal A$ with the complex group algebra of $Q$ in its normalized coefficient trace. Related character sums occur in finite gauge fields [@dijkgraaf1990] and two-dimensional Yang–Mills at zero area [@witten1991]. We choose @eq:trace directly, since a different gauge group volume would change the bare handle weights.\n\nAt a pair of pants, sector labels on the circles must coincide. The inverse metric on a sewn circle is $q\\delta_{ab}$, so a connected genus-zero bordism with $n$ circle states gives $\\tau(f_1\\cdots f_n)$. We pair all outgoing circles with the displayed states, inserting the complex conjugate of any state regarded as a bra.\n\nLet $\\Phi_{\\rm sh}$ describe $n$ boundary units prepared in one shared unknown sector. In the split state $\\Phi_{\\rm sp}$ each unit uses its own draw. Both functionals in @eq:state are positive and normalized on $\\mathcal A^{\\otimes n}$. Setting all but one insertion to the identity gives equal marginals.\n\n$$\n\\begin{aligned}\n\\phi_a(f)&=f(a),\\\\\n\\Phi_{\\rm sh}(f_1\\otimes\\cdots\\otimes f_n)\n   &=\\frac{1}{q}\\sum_{a\\in Q}\\prod_{i=1}^n\\phi_a(f_i),\\\\\n\\Phi_{\\rm sp}(f_1\\otimes\\cdots\\otimes f_n)\n   &=\\prod_{i=1}^n\\tau(f_i).\n\\end{aligned}\n\\label{eq:state}\n$$\n\nThe connected genus-zero amplitude gives $\\Phi_{\\rm sh}$. For $n$ disjoint disks the functor gives $\\Phi_{\\rm sp}$. A connected expectation in the statistical sense subtracts products of lower moments. In particular its two-boundary value is the difference of the two functionals. The connected geometry itself also includes the constant mode. @sec:sewing gives the handle normalization.\n\nEquivalently one can take sector states in a diagonal Hilbert space, with the boundary operators acting by multiplication. A uniform incoherent mixture and a vector with component norm squared $q^{-1}$ in each sector give the same expectations of these operators. Gauge invariant algebras can acquire a common center by imposing a constraint on the physical Hilbert space [@donnelly2012; @casini2014]. Here the center comes from record preparation.\n\n# Realization by independent certification {#sec:procedure}\n\n## Local records and the common preparation\n\nA determination of nonreceipt states that no matching requisition was entered and identifies the register row in which the search ended. Both fields are completed even if the register is empty. The missing document is a request for a replacement blank. Its instance number would have been assigned at intake, so the local search uses the form type and the intake window. We use the definitions of same and separate determinations in [@allende2024].\n\nThe service copy stock has eight alternatives. Alternative $a$ offsets the register alphabet by $a$ relative to the reference column on the backing sheet. A common batch gives each point a fresh sheet prepared with the same setting. In the split case each sheet has its own setting. The visible row $b$ gives no indication of whether the shift was used. The backing sheet records $a$ and is filed under stock inspection, outside the disposition register [@kerstan2022]. After dispatch the readers work independently.\n\nA local key directs the reader at point $i$ to the main setting with fraction $\\lambda_i$. The remaining keys are divided equally among the $q$ reference rows. A reference row can coincide with the main setting, which is why @eq:channel includes its uniform term on the diagonal as well. The fractions are design parameters. A routine log in which readers choose keys after inspecting the backing sheet would have a different channel.\n\n$$\nR_i(b\\mid a)=\\lambda_i\\delta_{ba}+\\frac{1-\\lambda_i}{q},\n\\qquad 0\\leq\\lambda_i\\leq 1 .\n\\label{eq:channel}\n$$\n\nIn the nominal enumeration each point has 24 main and eight distinct reference keys. We count each combination of keys once for every setting of the backing sheet. For the split case a separate setting is counted for each point. These are exact dictionaries. A row of the dictionary is a permitted service record and carries no claim of a separate delivery.\n\nThe third and fourth intake points are appeal registers. An appeal asks whether the original determination should stand and is eligible even if the requisition was never made. We hold the appeal disposition fixed and record only its row. The appeal copies use the same batch and different local keys. Attaching a scan of an earlier determination would change the joint law [@tilvis2025].\n\n## Insertions induced by the reader\n\nFor a function $u$ of the observed row, the inserted state is its conditional mean. @eq:pullback preserves the constant mode and attenuates every nonzero mode by the same fraction.\n\n$$\n(R_i^*u)(a)=\\sum_{b\\in Q} R_i(b\\mid a)u(b),\\qquad\nR_i^*\\chi_m=\n\\begin{cases}\n1 & m=0,\\\\\n\\lambda_i\\chi_m & m\\ne 0.\n\\end{cases}\n\\label{eq:pullback}\n$$\n\nWe grant access to the backing sheet only in @sec:conditioning, after all dispositions are fixed. Keeping it out of the local observable algebra makes the number of sectors an inference from the service copies.\n\n# Observable signatures of a common sector {#sec:observables}\n\n## Pair amplitudes and a lower bound on the center\n\nLet $J_{xy}$ be the joint probability of rows $x,y$ at the two points. Evaluate @eq:state on the corresponding indicators and write $P=q^{-1}\\mathbf 1\\mathbf 1^\\mathsf T$, $\\Lambda=\\lambda_L\\lambda_R$. Each marginal is uniform. The probability of equal rows is $q^{-1}[1+(q-1)\\Lambda]$ in the shared state.\n\n$$\n\\begin{aligned}\nJ_{\\rm sh} &= q^{-1}\\bigl[P+\\Lambda(I-P)\\bigr],\\\\\nJ_{\\rm sp} &= q^{-1}P .\n\\end{aligned}\n\\label{eq:pair}\n$$\n\nIn @fig:joint the diagonal probability is $79/1024$ per cell and the off-diagonal is $7/1024$. @tbl:pair compares the split preparation to a right reader switched to reference keys. Both controls give the same probability matrix. The reference control keeps the common setting, which would become observable again if the reader resumed its main-key schedule.\n\n```chart {#fig:joint width=0.85 caption=\"Joint probability by registration row, in percent, for the shared preparation with q = 8 and main-key fraction 3/4 at both points. Each point separately has uniform row probability.\"}\n{\n  \"type\": \"heatmap\",\n  \"rows\": [\"0\", \"1\", \"2\", \"3\", \"4\", \"5\", \"6\", \"7\"],\n  \"cols\": [\"0\", \"1\", \"2\", \"3\", \"4\", \"5\", \"6\", \"7\"],\n  \"xlabel\": \"Right registration row\",\n  \"ylabel\": \"Left registration row\",\n  \"values\": [\n    [7.71484375, 0.68359375, 0.68359375, 0.68359375, 0.68359375, 0.68359375, 0.68359375, 0.68359375],\n    [0.68359375, 7.71484375, 0.68359375, 0.68359375, 0.68359375, 0.68359375, 0.68359375, 0.68359375],\n    [0.68359375, 0.68359375, 7.71484375, 0.68359375, 0.68359375, 0.68359375, 0.68359375, 0.68359375],\n    [0.68359375, 0.68359375, 0.68359375, 7.71484375, 0.68359375, 0.68359375, 0.68359375, 0.68359375],\n    [0.68359375, 0.68359375, 0.68359375, 0.68359375, 7.71484375, 0.68359375, 0.68359375, 0.68359375],\n    [0.68359375, 0.68359375, 0.68359375, 0.68359375, 0.68359375, 7.71484375, 0.68359375, 0.68359375],\n    [0.68359375, 0.68359375, 0.68359375, 0.68359375, 0.68359375, 0.68359375, 7.71484375, 0.68359375],\n    [0.68359375, 0.68359375, 0.68359375, 0.68359375, 0.68359375, 0.68359375, 0.68359375, 7.71484375]\n  ],\n  \"colormap\": \"Blues\", \"annotate\": true, \"fmt\": \".2f\", \"vmin\": 0, \"vmax\": 8,\n  \"colorbar\": \"Probability (%)\", \"size\": [5, 3.6]\n}\n```\n\nTable: Pair laws for nominal and control preparations. The final column gives the singular value of the centered probability matrix on its nonconstant subspace. A reference-only reader uses eight keys, one per row. {#tbl:pair}\n\n| Preparation | Dictionary entries | Equal rows | Probability of equal rows | Singular value |\n|:------------|-------------------:|-----------:|--------------------------:|---------------:|\n| Shared, two nominal readers | 8,192 | 5,056 | $79/128$ | $9/128$ |\n| Split, two nominal readers | 65,536 | 8,192 | $1/8$ | 0 |\n| Shared, right reader reference only | 2,048 | 256 | $1/8$ | 0 |\n\nIn @eq:pair the centered matrix has rank $q-1$ when both readers sometimes use the main key. Before centering it has rank $q$. A mixture with $s$ sectors has $J=\\sum_{a=1}^s p_a u_a v_a^{\\mathsf T}$ for stochastic columns $u_a,v_a$. Thus $s\\geq\\rank(J)$ [@iles2023]. The present construction attains the bound. A binary distinction such as whether a blank was requested cannot account for @fig:joint if it exhausts what the readers share.\n\nFor the uncentered shared matrix, a perturbation of operator norm less than $\\Lambda/q$ cannot reduce rank. The exact split result has no such stability. Even a finite count table from independently sampled sheets will usually have rank greater than one. An archival test would require a sampling model and a tolerance for weak singular values.\n\nThe mutual information is $0.9653$ bits. Weight $\\log_2(qp)$ by $p$ and $\\log_2[q(1-p)/(q-1)]$ by $1-p$, with $p=79/128$ the equality probability [@shannon1948]. The mutual information tends to zero continuously as $\\Lambda$ tends to zero, while the eight-sector rank bound holds for every $\\Lambda>0$.\n\n## Higher boundary observables {#sec:higher}\n\nA simultaneous translation of all visible rows multiplies a character product by the phase of its total charge. Averaging over the common setting projects onto neutral products, as in @eq:charge. Zero-charge insertions are identities. In the split state any nonzero charge makes the moment vanish.\n\n$$\n\\EE_{\\rm sh}\\!\\left[\\prod_{i=1}^n\\chi_{m_i}(B_i)\\right]\n=\\mathbf 1_{\\sum_i m_i=0\\ \\mathrm{mod}\\ q}\n \\prod_{i:m_i\\ne 0}\\lambda_i.\n\\label{eq:charge}\n$$\n\nFor a neutral tuple with no proper neutral subtuple, each nontrivial partition contains a block of zero moment. Its full moment is its cumulant. Examples at $q=8$ are $(1,7)$, $(1,2,5)$ and $(1,1,1,5)$. At $\\lambda=3/4$ the third- and fourth-order predictions in @fig:higher are $27/64$ and $81/256$. The main-key fraction is unchanged at appeal.\n\n![Connected moments for three neutral charge tuples with no proper neutral subtuple. Each marker is an exact enumeration on the 64-key schedule. Lines connect the enumerated points.](boundary-orders){#fig:higher width=0.84}\n\nAt $\\lambda=1/2$, select one of the three pairs with probability $1/4$ each. Assign that pair a common uniform row and the third reader an independent one. With the remaining probability $1/4$ use three independent rows. Each pair has the mixture of a uniform equal-row law of weight $1/4$ and an independent law of weight $3/4$, in agreement with @eq:pair. The $(1,2,5)$ moment vanishes because there is always an independent nonzero character. For a shared sector it is $1/8$.\n\nAt $3/4$ those pair weights would sum to more than one. The ambiguity persists. Add the real part of the $(1,2,5)$ character product times $\\delta/q^3$ to the shared three-row law. Summing over any deleted row kills the added term. The all-reference contribution keeps the law positive for $|\\delta|<(1-\\lambda)^3$. The triple moment shifts by $\\delta/2$ since the double of the tuple is nonzero in $Q^3$. Even full pair data leave the appeal amplitude undetermined absent a preparation rule.\n\n# Conditioning on the supporting copy {#sec:conditioning}\n\n## Partial records\n\nSuppose the abstract from the backing sheet is $T_d=a\\bmod d$, with $d$ dividing $q$. The available records are zero, one, two or three low bits. Within stratum $t$, the sector is uniform on $A_t=\\{t+jd:0\\leq j<q/d\\}$. We test independence within each stratum. Pooling over the strata recovers @eq:state.\n\nWrite $p_t$ for the probability vector of the uniform law on $A_t$. The reference terms cancel in the centered cross-covariance $K_t$ of row indicators. A row outside $A_t$ can still appear through a local reference key, but its conditional probability is independent of the sector. @eq:conditional follows by taking the covariance of the conditional mean vectors.\n\n$$\nK_t=\\Lambda\\bigl[\\operatorname{diag}(p_t)-p_t p_t^{\\mathsf T}\\bigr],\n\\qquad\n\\rank(K_t)=q/d-1\\quad (\\Lambda>0).\n\\label{eq:conditional}\n$$\n\nThe nonzero singular values at fixed $d$ are $\\Lambda d/q$. Each increases as the stratum narrows, even though fewer directions remain correlated. The trace decreases as $\\Lambda(1-d/q)$. To obtain the diagonal character covariance in @eq:mode, use that the sector mean of $\\chi_m$ has modulus one for multiples of $q/d$ and zero for other charges.\n\n$$\n\\Cov\\!\\left(\\chi_m(B_L),\\overline{\\chi_m(B_R)}\\mid T_d=t\\right)\n=\\Lambda\\left(1-\\mathbf 1_{m=0\\ \\mathrm{mod}\\ (q/d)}\\right),\\qquad m\\ne 0 .\n\\label{eq:mode}\n$$\n\n@fig:records and @tbl:shared-center give the conditional results. At $d=4$ four nonzero charge probes still have nonzero diagonal entries. Restricted to the two sectors in a stratum, their centered values span just one direction. If instead one averages the centered matrices across strata, their supports are orthogonal, so the rank becomes $q-d$ at $\\Lambda>0$ and the nonzero eigenvalues are $\\Lambda/q$.\n\n```chart {#fig:records width=0.91 caption=\"Conditional factorization defect with nominal readout. Left, rank within each stratum and rank after averaging the centered matrices. Right, the diagonal covariance in four charge modes, normalized by the unconditioned nonzero-mode covariance.\"}\n{\n  \"panels\": [\n    {\n      \"type\": \"line\", \"xlabel\": \"Recorded bits\", \"ylabel\": \"Covariance rank\",\n      \"x\": [\"0\", \"1\", \"2\", \"3\"],\n      \"series\": [\n        { \"label\": \"Fixed stratum\", \"y\": [7, 3, 1, 0], \"marker\": \"o\" },\n        { \"label\": \"Stratum average\", \"y\": [7, 6, 4, 0], \"marker\": \"s\", \"style\": \"dashed\" }\n      ],\n      \"legend\": \"upper right\", \"ylim\": [0, 8]\n    },\n    {\n      \"type\": \"heatmap\",\n      \"rows\": [\"1\", \"2\", \"3\", \"4\"], \"cols\": [\"0\", \"1\", \"2\", \"3\"],\n      \"xlabel\": \"Recorded bits\", \"ylabel\": \"Charge\",\n      \"values\": [[1, 1, 1, 0], [1, 1, 0, 0], [1, 1, 1, 0], [1, 0, 0, 0]],\n      \"vmin\": 0, \"vmax\": 1, \"colormap\": \"Blues\", \"annotate\": true, \"fmt\": \".0f\"\n    }\n  ],\n  \"size\": [6.6, 2.7]\n}\n```\n\nTable: Exact conditional results for the common preparation with two nominal readers. The number of possible sectors and the entropy refer to a single observed value of the supporting record. The trace is given in row-indicator coordinates. {#tbl:shared-center}\n\n| Recorded bits | Sectors compatible | Entropy (bits) | Rank within stratum | Conditional trace |\n|--------------:|-------------------:|---------------:|--------------------:|------------------:|\n| 0 | 8 | 3 | 7 | $63/128$ |\n| 1 | 4 | 2 | 3 | $27/64$ |\n| 2 | 2 | 1 | 1 | $9/32$ |\n| 3 | 1 | 0 | 0 | 0 |\n\n## Removal of the originating field\n\nThe nonreceipt procedure records a withdrawal on both supporting copies [@chakor2021]. In our model its designation is fixed before choosing the backing sheet. It changes a deterministic disposition and leaves @eq:pair and @eq:charge unchanged. If the originating number depended on the setting, removing it would also remove potential sector information.\n\nFiling the disposition alone replaces the insertion by the identity, killing every registration cumulant of order at least two involving that leg. Issuing fresh stock independently at each point gives the split state. Filing the full alignment mark gives a fixed-sector test. The fixed-sector marginals put extra mass on the observed setting, while the unstratified split marginals are uniform.\n\nThe abstractor's manual calls for an empty registration column if the backing sheet was returned unexamined [@kerstan2022]. This has $d=1$. A zero written in a completed three-bit field has $d=8$ and selects $a=0$. If the database imports both as numerical zero, it can mistake a rank-seven mixture for a factorizing stratum. Every setting generates the same number of certificates, so the count of issued certificates cannot expose this error.\n\n# Scope of the correspondence {#sec:discussion}\n\nThe rank bound assumes a mixture of local products. Direct interaction could fit the same probability matrix in a one-sector model if conditional independence failed. We rule out that interaction after dispatch by construction. Observational records from independent mailings would leave the cause of covariance unidentified. The comparison with separately prepared stock deliberately changes the outer draw while holding the two channels fixed.\n\nFor general channel matrices $U,V$ and weights $p$, $J=U\\operatorname{diag}(p)V^{\\mathsf T}$. Full column rank on both sides and positive weights give $\\rank(J)=q$, even if the charge rule fails. A convention folding two registration rows together can remove a direction from one channel. One can then bound the number of sectors from a rectangular matrix without being able to enumerate all settings available to the stock department.\n\nIn topological gravity the weights of the center can depend on a sum over topologies [@gardiner2021]. Here the distribution is supplied with the stock. Under our trace choice, handles multiply numerator and denominator of every normalized moment by the same number. The sewn amplitude evaluates the joint preparation, while the symmetric monoidal functor continues to evaluate separate disks multiplicatively.\n\n# Conclusion {#sec:conclusion}\n\nThe declaration of independence applies to the search of the local disposition registers. Extending it to the full certificates requires specifying how the supporting stock was prepared or conditioning on its setting. For an appeal filed solely to establish that the first certificate was unnecessary, the registration of that appeal is still a boundary observable of the batch.\n\n\\appendix\n\n# Handle factors and the vacuum convention {#sec:sewing}\n\nThe comultiplication is $\\Delta(e_a)=q e_a\\otimes e_a$. The handle operator $m\\circ\\Delta$ has eigenvalue $q$ in each sector. In @eq:handle the boundary insertions may be evaluated before or after the handles. The empty sphere amplitude is one and the empty torus amplitude is $q$.\n\n$$\nZ_{g,n}(f_1,\\ldots,f_n)=q^g\\tau(f_1\\cdots f_n),\\qquad\n\\frac{Z_{g,n}(f_1,\\ldots,f_n)}{Z_{g,n}(1,\\ldots,1)}\n=\\tau(f_1\\cdots f_n).\n\\label{eq:handle}\n$$\n\nFor $\\tau(e_a)=w_a>0$, the handle eigenvalue is $1/w_a$ and the unnormalized genus-$g$ weights are $w_a^{1-g}$. For $n\\geq 1$ an alternative is to leave the Frobenius weights uniform and insert the density $a\\mapsto qp_a$ on one boundary. Both constructions give the same normalized genus-zero weights for $w_a=p_a$, but only the density insertion keeps the prior $p_a$ at all genera.\n\nGluing two caps pairs their factors $\\tau(e_a)$ with the inverse metric, producing $\\sum_a (1/q)^2 q=1$. Discarding the inverse metric gives $1/q$, which is the agreement probability of independent sector draws. The counting discrepancy would persist even with no visible registrations.\n\n# Integer enumeration {#sec:integers}\n\nAt fixed $a$, 25 keys output $a$ and one outputs each other row. A cell $(x,x)$ receives $25^2+7=632$ pairs of keys after summing over sectors. The count in $(x,y)$ for $x\\ne y$ is $2(25)+6=56$. In the split case a fixed output has 32 setting-key combinations on either side, which gives $1024$ per cell. In the reference-only control any right output has one key, yielding $32$ per cell after summing over the shared sector.\n\nWith $M$ keys at each point and $k$ keys per reference row, a nonzero character has unnormalized local sum $(M-8k)\\chi_m(a)$. For a neutral $n$-tuple of nonzero charges the summed moment is $8(M-8k)^n$ before dividing by the dictionary size $8M^n$. @fig:higher uses $M=64$ and $k=8(1-\\lambda)$. The $M=32$ and $M=64$ schedules give the same values where they coincide.\n\nTo reproduce @fig:joint take the outer product of local count vectors for each $a$ and sum over the sectors. For the conditional calculation, restrict the sum to $A_t$ and divide by $(q/d)32^2$ for each $t$. Subtract the product of the two normalized conditional marginals to recover @eq:conditional.\n","bibliography":"@article{giddings1988,\n  author = {Giddings, Steven B. and Strominger, Andrew},\n  title = {Loss of incoherence and determination of coupling constants in quantum gravity},\n  journal = {Nuclear Physics B},\n  volume = {307},\n  pages = {854--866},\n  year = {1988},\n  doi = {10.1016/0550-3213(88)90109-5}\n}\n@article{coleman1988,\n  author = {Coleman, Sidney},\n  title = {Black holes as red herrings: Topological fluctuations and the loss of quantum coherence},\n  journal = {Nuclear Physics B},\n  volume = {307},\n  pages = {867--882},\n  year = {1988},\n  doi = {10.1016/0550-3213(88)90110-1}\n}\n@article{marolf2020,\n  author = {Marolf, Donald and Maxfield, Henry},\n  title = {Transcending the ensemble: baby universes, spacetime wormholes, and the order and disorder of black hole information},\n  journal = {Journal of High Energy Physics},\n  volume = {08},\n  pages = {044},\n  year = {2020},\n  doi = {10.1007/JHEP08(2020)044}\n}\n@misc{saad2019,\n  author = {Saad, Phil and Shenker, Stephen H. and Stanford, Douglas},\n  title = {{JT} gravity as a matrix integral},\n  eprint = {1903.11115},\n  archivePrefix = {arXiv},\n  year = {2019}\n}\n@article{harlow2020,\n  author = {Harlow, Daniel and Jafferis, Daniel},\n  title = {The factorization problem in {Jackiw--Teitelboim} gravity},\n  journal = {Journal of High Energy Physics},\n  volume = {02},\n  pages = {177},\n  year = {2020},\n  doi = {10.1007/JHEP02(2020)177}\n}\n@article{santos2021,\n  author = {Marolf, Donald and Santos, Jorge E.},\n  title = {{AdS} Euclidean wormholes},\n  journal = {Classical and Quantum Gravity},\n  volume = {38},\n  pages = {224002},\n  year = {2021}\n}\n@article{blommaert2022,\n  author = {Blommaert, Andreas and Iliesiu, Luca V. and Kruthoff, Jorrit},\n  title = {Gravity factorized},\n  journal = {Journal of High Energy Physics},\n  volume = {09},\n  pages = {080},\n  year = {2022},\n  doi = {10.1007/JHEP09(2022)080}\n}\n@techreport{allende2024,\n  author = {Allende Mireis, Iria},\n  title = {Independence of concurrent determinations of nonreceipt},\n  institution = {Registry Methods Unit, Brackenholt School of Administration},\n  number = {RMU-24-09},\n  year = {2024}\n}\n@article{atiyah1988,\n  author = {Atiyah, Michael},\n  title = {Topological quantum field theory},\n  journal = {Publications Math{\\'e}matiques de l'IH{\\'E}S},\n  volume = {68},\n  pages = {175--186},\n  year = {1988},\n  doi = {10.1007/BF02698547}\n}\n@article{abrams1996,\n  author = {Abrams, Lowell},\n  title = {Two-dimensional topological quantum field theories and {Frobenius} algebras},\n  journal = {Journal of Knot Theory and Its Ramifications},\n  volume = {5},\n  pages = {569--587},\n  year = {1996}\n}\n@book{kock2004,\n  author = {Kock, Joachim},\n  title = {Frobenius Algebras and 2-D Topological Quantum Field Theories},\n  publisher = {Cambridge University Press},\n  series = {London Mathematical Society Student Texts},\n  volume = {59},\n  year = {2004},\n  doi = {10.1017/CBO9780511615443}\n}\n@article{dijkgraaf1990,\n  author = {Dijkgraaf, Robbert and Witten, Edward},\n  title = {Topological gauge theories and group cohomology},\n  journal = {Communications in Mathematical Physics},\n  volume = {129},\n  pages = {393--429},\n  year = {1990},\n  doi = {10.1007/BF02096988}\n}\n@article{witten1991,\n  author = {Witten, Edward},\n  title = {On quantum gauge theories in two dimensions},\n  journal = {Communications in Mathematical Physics},\n  volume = {141},\n  pages = {153--209},\n  year = {1991}\n}\n@article{donnelly2012,\n  author = {Donnelly, William},\n  title = {Decomposition of entanglement entropy in lattice gauge theory},\n  journal = {Physical Review D},\n  volume = {85},\n  pages = {085004},\n  year = {2012},\n  doi = {10.1103/PhysRevD.85.085004}\n}\n@article{casini2014,\n  author = {Casini, Horacio and Huerta, Marina and Rosabal, Jos{\\'e} Alejandro},\n  title = {Remarks on entanglement entropy for gauge fields},\n  journal = {Physical Review D},\n  volume = {89},\n  pages = {085012},\n  year = {2014},\n  doi = {10.1103/PhysRevD.89.085012}\n}\n@techreport{kerstan2022,\n  author = {Kerstan, Ilse M. and Deyrieux, Alain},\n  title = {Registration and abstracting conventions for availability determinations},\n  institution = {Alderbasin Centre for Records Engineering},\n  number = {ACRE-22-5},\n  year = {2022}\n}\n@article{tilvis2025,\n  author = {Tilvis-Kerman, Anja and Caderel, Brice L.},\n  title = {Record identity in reviews of negative administrative determinations},\n  journal = {Annals of Registry Administration},\n  volume = {14},\n  pages = {91--111},\n  year = {2025}\n}\n@article{iles2023,\n  author = {Ilescu, Marin and Navari, Simin E.},\n  title = {Finite support bounds for common preparation across noisy boundaries},\n  journal = {Studies in Discrete Field Theory},\n  volume = {8},\n  pages = {233--250},\n  year = {2023}\n}\n@article{shannon1948,\n  author = {Shannon, Claude E.},\n  title = {A mathematical theory of communication},\n  journal = {Bell System Technical Journal},\n  volume = {27},\n  pages = {379--423 and 623--656},\n  year = {1948}\n}\n@article{chakor2021,\n  author = {Chakor, Vedika N.},\n  title = {Withdrawal dispositions for absent originating instruments},\n  journal = {Review of Documentary Procedure},\n  volume = {23},\n  pages = {141--153},\n  year = {2021}\n}\n@article{gardiner2021,\n  author = {Gardiner, John and Megas, Stathis},\n  title = {2d {TQFTs} and baby universes},\n  journal = {Journal of High Energy Physics},\n  volume = {10},\n  pages = {052},\n  year = {2021},\n  doi = {10.1007/JHEP10(2021)052}\n}\n","figures":[{"id":"boundary-orders","kind":"chart","data":"{\"type\":\"line\",\"style\":\"pgfplots\",\"xlabel\":\"Main-key fraction\",\"ylabel\":\"Connected moment\",\"x\":[0,0.125,0.25,0.375,0.5,0.625,0.75,0.875,1],\"xlim\":[0,1],\"ylim\":[0,1.05],\"series\":[{\"label\":\"Two boundaries\",\"y\":[0,0.015625,0.0625,0.140625,0.25,0.390625,0.5625,0.765625,1],\"marker\":\"o\"},{\"label\":\"Three boundaries\",\"y\":[0,0.001953125,0.015625,0.052734375,0.125,0.244140625,0.421875,0.669921875,1],\"style\":\"dashed\",\"marker\":\"s\"},{\"label\":\"Four boundaries\",\"y\":[0,0.000244140625,0.00390625,0.019775390625,0.0625,0.152587890625,0.31640625,0.586181640625,1],\"style\":\"dotted\",\"marker\":\"^\"}],\"legend\":\"upper left\",\"size\":[5.4,3]}"}],"acknowledgments":"We thank Lisette Vangean for checking the distinction between a returned abstract and an unread backing sheet.","options":{"citation_style":"numeric"}}