OctoInformBeta

OpenAI preprint, September 2026 · §8. Relational coding of the diagonal argument · §8

Admissibility, comparison, probes, and the chosen query parameter: A finite marked disjunction fixes the parameter for relational probes and three copies of their observations

Relational codes and a chosen query parameter.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

The relational diagonal construction tests data by formulas obtained from retained paths. A common probe takes a type parameter and observes a bundle containing a query parameter and a data point. Three tagged copies of probe elements supply the observations used to choose that parameter.

Fix a sort set SS. Terms have variables and sorts, applications, annotated abstractions λx:A.M\lambda x:A.M, and dependent products Πx:A.B\Pi x:A.B.✦ Each application and binder has one of the mode labels dddd, dpdp, and pppp, denoting data to data, data to proof, and proof to proof.✦ Named variables below are subject to capture-avoiding substitution. An expression Q(−)Q(-) is an open term with one distinguished input; composition Q∘fQ\circ f substitutes f(−)f(-) for that input. A nondependent arrow A→lBA\mathbin{\to_l}B is a dependent product in mode ll whose codomain BB does not depend on its bound variable.

Retained Paths and Formulas

A positive data path UU is a finite sequence of data arrows with retained domain annotations and dummy arguments.✦ Its wrapped type U(T)U(T), injection ιU\iota_U, and projection πU\pi_U are given by (1); the empty path is the identity. Here (U,D,e)(U,D,e) denotes extension by annotation DD and dummy argument ee. (U,D,e)(T)=D→ddU(T),ι(U,D,e)(t)=λddx:D.ιU(t),π(U,D,e)(w)=πU(wdde).\begin{aligned} (U,D,e)(T)&=D\mathbin{\to_{dd}}U(T),\\ \iota_{(U,D,e)}(t)&=\lambda_{dd}x:D.\iota_U(t),& \pi_{(U,D,e)}(w)&=\pi_U(w\,_{dd}e). \end{aligned}(1)

A proof path HH is a finite sequence of positive and negative extensions, acting on formulas as in (2).✦ A positive extension retains a data domain DD; a negative extension retains a formula FF. H+D(T)=D→dpH(T),H−F(T)=H(T)→ppF,1(T)=T.H_{+D}(T)=D\mathbin{\to_{dp}}H(T),\qquad H_{-F}(T)=H(T)\mathbin{\to_{pp}}F,\qquad 1(T)=T.(2)✦

A data tail VV is a tuple consisting of a sort sVs_V, a positive path UVU_V, and encoding and decoding proof paths HV+H_V^+ and HV−H_V^-.✦ Its carrier and its sending and reading operations are defined in (3). |V|=UV(sV),send⁡V(Q)=ιUV(HV+(Q)),read⁡V(w)=HV−(πUV(w)).|V|=U_V(s_V),\qquad \operatorname{send}_V(Q)=\iota_{U_V}(H_V^+(Q)),\qquad \operatorname{read}_V(w)=H_V^-(\pi_{U_V}(w)).(3)

A retained data path WW is generated by the identity, positive extensions (W,D,e)(W,D,e), and negative extensions (W,V)(W,V).✦ The parity |W||W| of a retained data path WW is the number of its negative extensions modulo two. The wrapped type W(T)W(T), builder bW(T,a)b_W(T,a), and observation oW(T,w,a)o_W(T,w,a) are the recursive operations in (4). A builder takes an open term at odd parity and a term at even parity; an observation takes a term at odd parity and an open term at even parity. Fresh bound variables do not capture the distinguished input of an open term. 1(T)=T,b1(T,a)=a,o1(T,w,Q)=Q(w),(W,D,e)(T)=D→ddW(T),b(W,D,e)(T,a)=λddx:D.bW(T,a),o(W,D,e)(T,w,a)=oW(T,wdde,a),(W,V)(T)=W(T)→dd|V|,b(W,V)(T,a)=λddx:W(T).send⁡V(oW(T,x,a)),o(W,V)(T,w,a)=read⁡V(wddbW(T,a)).\begin{aligned} 1(T)&=T,& b_1(T,a)&=a,&o_1(T,w,Q)&=Q(w),\\ (W,D,e)(T)&=D\mathbin{\to_{dd}}W(T),& b_{(W,D,e)}(T,a)&=\lambda_{dd}x:D.b_W(T,a),\\ o_{(W,D,e)}(T,w,a)&=o_W(T,w\,_{dd}e,a),\\ (W,V)(T)&=W(T)\mathbin{\to_{dd}}|V|,& b_{(W,V)}(T,a)&=\lambda_{dd}x:W(T). \operatorname{send}_V(o_W(T,x,a)),\\ o_{(W,V)}(T,w,a)&=\operatorname{read}_V(w\,_{dd}b_W(T,a)). \end{aligned}(4)

For odd paths W,W′W,W', reencoding rW,W′Tr_{W,W'}^T is the operation in (5). The open terms a↦oW(T,w,a)a\mapsto o_W(T,w,a) and w↦oW(T,w,a)w\mapsto o_W(T,w,a) are respectively evaluation and contravariant observation. rW,W′T(w)=bW′(T,a↦oW(T,w,a)).r_{W,W'}^T(w)=b_{W'}(T,a\mapsto o_W(T,w,a)).(5)✦

A logical code LL is a choice of falsity ⊥\bot and incoming and outgoing proof paths Hin,HoutH_{\mathrm{in}},H_{\mathrm{out}}.✦ A quantifier code KK is a pair of proof paths Kin,KoutK_{\mathrm{in}},K_{\mathrm{out}}.✦ Their implication and universal quantifier are (6); KYK_Y and KPK_P will be the quantifier codes for elements and predicates, respectively. A⇒B=Hout(Hin(A)→ppB),∀Kx:T.Q(x)=Kout(Πdpx:T.Kin(Q(x))).A\Rightarrow B =H_{\mathrm{out}}(H_{\mathrm{in}}(A)\mathbin{\to_{pp}}B),\qquad \forall_K x:T.Q(x) =K_{\mathrm{out}}\bigl(\Pi_{dp}x:T.K_{\mathrm{in}}(Q(x))\bigr).(6)

All logical notation below denotes the formula operations of LL in (7). A guarded universal quantifier with guard GG and body QQ is a universal quantifier whose body is G(x)⇒Q(x)G(x)\Rightarrow Q(x)✦; a finite disjunction is a disjunction associated to the right with final term ⊥\bot. The negation ¬A\neg A of a formula AA is its implication to falsity.✦ Implication from falsity to itself gives the truth formula ⊤\top.✦ The conjunction A∧BA\wedge B of formulas AA and BB is the negation of the implication from AA to the negation of BB.✦ For formulas AA and BB, negating the conjunction of their negations gives their disjunction A∨BA\vee B.✦ The logical equivalence A↔BA\leftrightarrow B is the conjunction of the implications in both directions; it expresses typed proof-term transformers in both directions.✦ Negating universal quantification over TT of the negation of Q(x)Q(x) gives the existential quantifier ∃Kx:T.Q(x)\exists_K x:T.Q(x).✦ The logical conditional cond⁡(G,A,B)\operatorname{cond}(G,A,B) is the disjunction of G∧AG\wedge A and ¬G∧B\neg G\wedge B.✦ Polarity selection sign⁡ϵ(A)\operatorname{sign}_{\epsilon}(A) is the formula AA for ϵ=1\epsilon=1 and its negation for ϵ=0\epsilon=0.✦ ¬A=A⇒⊥,⊤=⊥⇒⊥,A∧B=¬(A⇒¬B),A∨B=¬(¬A∧¬B),A↔B=(A⇒B)∧(B⇒A),∃Kx:T.Q(x)=¬∀Kx:T.¬Q(x),cond⁡(G,A,B)=(G∧A)∨(¬G∧B),sign⁡ϵ(A)={Aϵ=1,¬Aϵ=0.\begin{aligned} \neg A&=A\Rightarrow\bot,&\top&=\bot\Rightarrow\bot,\\ A\wedge B&=\neg(A\Rightarrow\neg B),& A\vee B&=\neg(\neg A\wedge\neg B),\\ A\leftrightarrow B&=(A\Rightarrow B)\wedge(B\Rightarrow A),& \exists_K x:T.Q(x)&=\neg\forall_K x:T.\neg Q(x),\\ \operatorname{cond}(G,A,B)&=(G\wedge A)\vee(\neg G\wedge B),& \operatorname{sign}_{\epsilon}(A)&= \begin{cases}A&\epsilon=1,\\ \neg A&\epsilon=0.\end{cases} \end{aligned}(7)

Double Channels and Finite Queries

A double channel D=(W1,W2)D=(W_1,W_2) is a pair of odd retained data paths.✦ Its carrier D(T)D(T) and iterates Dn(T)D^n(T) give the fourfold carrier FD(T)F_D(T), predicate code type BD(T)B_D(T), and parameter type MD(T)M_D(T) of (8). D(T)=W2(W1(T)),D0(T)=T,Dn+1(T)=D(Dn(T)),FD(T)=D4(T),BD(T)=W1(FD(T)),MD(T)=W2(BD(T)).\begin{gathered} D(T)=W_2(W_1(T)),\qquad D^0(T)=T,\qquad D^{n+1}(T)=D(D^n(T)),\\ F_D(T)=D^4(T),\qquad B_D(T)=W_1(F_D(T)),\qquad M_D(T)=W_2(B_D(T)). \end{gathered}(8)

Double channel evaluation EDE_D, its bit test βD\beta_D, the signed embedding eDϵe_D^\epsilon, and signed recovery hDh_D are defined by (9). ED(T,v,Q)=oW2(W1(T),v,bW1(T,Q)),βD(T,v)=ED(T,v,⊤),eDϵ(T,t)=bW2(W1(T),a↦sign⁡ϵ(oW1(T,a,t))),hD(T,v,Q)=cond⁡(βD(T,v),ED(T,v,Q),¬ED(T,v,Q)).\begin{aligned} E_D(T,v,Q)&=o_{W_2}(W_1(T),v,b_{W_1}(T,Q)),\\ \beta_D(T,v)&=E_D(T,v,\top),\\ e_D^\epsilon(T,t)&=b_{W_2}\bigl(W_1(T), a\mapsto\operatorname{sign}_\epsilon(o_{W_1}(T,a,t))\bigr),\\ h_D(T,v,Q)&=\operatorname{cond} (\beta_D(T,v),E_D(T,v,Q),\neg E_D(T,v,Q)). \end{aligned}(9)

A query index q∈{0,…,8}q\in\{0,\ldots,8\} has the four-bit string 𝒃(q)=(b0,b1,b2,b3)\mathbf b(q)=(b_0,b_1,b_2,b_3) of its binary digits, in increasing order of significance.✦ A query tag t∈{0,…,7}t\in\{0,\ldots,7\} has nonzero index t+1t+1.✦ Iterated signed embedding eD𝒃e_D^{\mathbf b} and the tag formula τD𝒃\tau_D^{\mathbf b} are defined in (10), where kk is the length of the remaining string 𝒄\mathbf c. For an input term denoted here by tt, write jDq(T,t)=eD𝒃(q)(T,t)j_D^q(T,t)=e_D^{\mathbf b(q)}(T,t)✦ and τDq(T,−)=τD𝒃(q)(T,−)\tau_D^q(T,-)=\tau_D^{\mathbf b(q)}(T,-). eD()(T,t)=t,τD()(T,v)=⊤,eD(ϵ,𝒄)(T,t)=eDϵ(Dk(T),eD𝒄(T,t)),τD(ϵ,𝒄)(T,v)=sign⁡ϵ(βD(Dk(T),v))∧hD(Dk(T),v,τD𝒄(T,−)).\begin{aligned} e_D^{()}(T,t)&=t,&\tau_D^{()}(T,v)&=\top,\\ e_D^{(\epsilon,\mathbf c)}(T,t) &=e_D^\epsilon(D^k(T),e_D^{\mathbf c}(T,t)),\\ \tau_D^{(\epsilon,\mathbf c)}(T,v) &=\operatorname{sign}_\epsilon(\beta_D(D^k(T),v)) \wedge h_D(D^k(T),v,\tau_D^{\mathbf c}(T,-)). \end{aligned}(10)

Iterated recovery HDn(T,Q)H_D^n(T,Q) is the open formula that equals QQ at depth zero and has successor value v↦hD(Dn(T),v,HDn(T,Q))v\mapsto h_D(D^n(T),v,H_D^n(T,Q)).✦ For an odd payload path UU, the decoder dec⁡D,U\operatorname{dec}_{D,U}, encoder enc⁡D,Ut\operatorname{enc}_{D,U}^t, and tagged query qD,Utq_{D,U}^t are the operations in (11). dec⁡D,U(T,p)=bU(T,c↦oW1(FD(T),p,jD0(T,c))),enc⁡D,Ut(T,p)=bW1(FD(T),z↦(τD0(T,z)∧HD4(T,oU(T,p,−))(z))∨τDt+1(T,z)),qD,Ut(T,m,p)=oW2(BD(T),m,enc⁡D,Ut(T,p)).\begin{aligned} \operatorname{dec}_{D,U}(T,p) &=b_U(T,c\mapsto o_{W_1}(F_D(T),p,j_D^0(T,c))),\\ \operatorname{enc}_{D,U}^t(T,p) &=b_{W_1}\bigl(F_D(T),z\mapsto (\tau_D^0(T,z)\wedge H_D^4(T,o_U(T,p,-))(z))\vee\tau_D^{t+1}(T,z)\bigr),\\ q_{D,U}^t(T,m,p) &=o_{W_2}(B_D(T),m,\operatorname{enc}_{D,U}^t(T,p)). \end{aligned}(11)

The marking formula μDt\mu_D^t, encoded point pt⁡D,U\operatorname{pt}_{D,U}, dummy parameter dDd_D, and dummy test δD\delta_D are defined by (12). μDt(T,c)(p)=oW1(FD(T),p,jDt+1(T,c)),pt⁡D,U(T,c0,c)=bW2(BD(T),p↦μD7(T,c0)(p)∧oU(T,dec⁡D,U(T,p),c)),dD(T)=bW2(BD(T),⊤),δD(T,m)=oW2(BD(T),m,bW1(FD(T),⊥)).\begin{aligned} \mu_D^t(T,c)(p)&=o_{W_1}(F_D(T),p,j_D^{t+1}(T,c)),\\ \operatorname{pt}_{D,U}(T,c_0,c) &=b_{W_2}\bigl(B_D(T),p\mapsto \mu_D^7(T,c_0)(p)\wedge o_U(T,\operatorname{dec}_{D,U}(T,p),c)\bigr),\\ d_D(T)&=b_{W_2}(B_D(T),\top),\\ \delta_D(T,m)&=o_{W_2}(B_D(T),m,b_{W_1}(F_D(T),\bot)). \end{aligned}(12)

A bundle bun⁡D(A,d,m,y)\operatorname{bun}_D(A,d,m,y) is the term in (13). bun⁡D(A,d,m,y)=bW2(W1(A),a↦cond⁡(oW1(A,a,d),oW1(A,a,m),oW1(A,a,y))).\operatorname{bun}_D(A,d,m,y)=b_{W_2}\bigl(W_1(A),a\mapsto \operatorname{cond}(o_{W_1}(A,a,d),o_{W_1}(A,a,m),o_{W_1}(A,a,y))\bigr).(13)✦

Access acc⁡D\operatorname{acc}_D, the inner probe in⁡D,U\operatorname{in}_{D,U}, and recovery access ra⁡D,U\operatorname{ra}_{D,U} retain the dummy test in (14). The terms GG and ff are open terms. acc⁡D(A,G,w,f)=oW2(W1(A),w,bW1(A,a↦cond⁡(G(a),⊤,oW2(W1(A),w,f(a))))),in⁡D,U(T,p)=bW1(MD(T),m↦cond⁡(δD(T,m),⊥,qD,U7(T,m,p))),ra⁡D,U(T,w,f)=acc⁡D(MD(T),δD(T,−),w,m↦in⁡D,U(T,f(m))).\begin{aligned} \operatorname{acc}_D(A,G,w,f) &=o_{W_2}\bigl(W_1(A),w,b_{W_1}(A,a\mapsto \operatorname{cond}(G(a),\top,o_{W_2}(W_1(A),w,f(a))))\bigr),\\ \operatorname{in}_{D,U}(T,p) &=b_{W_1}\bigl(M_D(T),m\mapsto \operatorname{cond}(\delta_D(T,m),\bot,q_{D,U}^7(T,m,p))\bigr),\\ \operatorname{ra}_{D,U}(T,w,f) &=\operatorname{acc}_D\bigl(M_D(T),\delta_D(T,-),w, m\mapsto\operatorname{in}_{D,U}(T,f(m))\bigr). \end{aligned}(14)

The structural realization Real⁡D,U(T,c,Φ)\operatorname{Real}_{D,U}(T,c,\Phi) of seven open formulas Φ0,…,Φ6\Phi_0,\ldots,\Phi_6 is the finite marked disjunction in (15). Real⁡D,U(T,c,Φ)=bW2(BD(T),p↦⋁t=06(μDt(T,c)(p)∧Φt(dec⁡D,U(T,p)))).\operatorname{Real}_{D,U}(T,c,\Phi)= b_{W_2}\left(B_D(T),p\mapsto \bigvee_{t=0}^{6} \bigl(\mu_D^t(T,c)(p)\wedge \Phi_t(\operatorname{dec}_{D,U}(T,p))\bigr)\right).(15)✦

Uniform Probes

A common probe PP is a choice of a seed sort ss, an initial retained path II of parity bb, odd raw and back paths r,ar,a, an even function path ff, a positive lifting path UU, and a data tail VV.✦ Put D=(a,r)D=(a,r).✦ For a type parameter AA, the probe’s base type A0A_0, raw type TAT_A, parameter type MAM_A, and single bundle type NAN_A are given in (16). The common probe type is ZZ. Type parameters range over the positive wrapped sort U(s)U(s).✦ The probe function type FAF_A is the data arrow from the single bundle type NAN_A to the tail carrier |V||V|.✦ A0=I(A),TA=r(A0),MA=MD(TA),NA=D(MA),FA=NA→dd|V|,Z=Πddz:U(s).f(FπU(z)).\begin{gathered} A_0=I(A),\qquad T_A=r(A_0),\qquad M_A=M_D(T_A),\qquad N_A=D(M_A),\\ F_A=N_A\mathbin{\to_{dd}}|V|,\qquad Z=\Pi_{dd}z:U(s).f(F_{\pi_U(z)}). \end{gathered}(16)

Probe specialization v[A]v[A], observation Obs⁡P\operatorname{Obs}_P, and uniform abstraction Unif⁡P\operatorname{Unif}_P are defined by (17). The open term Q(A,w)Q(A,w) has the type parameter and bundle as distinguished inputs. v[A]=vddιU(A),Obs⁡P(A,v,w)=of(FA,v[A],g↦read⁡V(gddw)),Unif⁡P(Q)=λddz:U(s).bf(FπU(z),λddw:NπU(z).send⁡V(Q(πU(z),w))).\begin{aligned} v[A]&=v\,_{dd}\iota_U(A),\\ \operatorname{Obs}_P(A,v,w) &=o_f(F_A,v[A],g\mapsto\operatorname{read}_V(g\,_{dd}w)),\\ \operatorname{Unif}_P(Q) &=\lambda_{dd}z:U(s). b_f\bigl(F_{\pi_U(z)}, \lambda_{dd}w:N_{\pi_U(z)}. \operatorname{send}_V(Q(\pi_U(z),w))\bigr). \end{aligned}(17)

Admissibility and Comparison

Fix odd logical and predicate paths ℓ,ρ\ell,\rho. The shared observation channel is C=(ℓ,r,ρ)C=(\ell,r,\rho)✦; more generally a channel is any triple of odd paths.✦ Its logical, raw, and predicate types are ℓ(T)\ell(T), r(T)r(T), and ρ(r(T))\rho(r(T)). The conversions c↓,c↑c\downarrow,c\uparrow, predicate builder [Q]T[Q]_T, evaluation ev⁡T\operatorname{ev}_T, and roundtrip round⁡T\operatorname{round}_T are defined by (18). y↓=rℓ,rT(y),c↑=rr,ℓT(c),[Q]T=bρ(r(T),c↦Q(c↑)),ev⁡T(p,y)=oρ(r(T),p,y↓),round⁡T(y)=(y↓)↑.\begin{aligned} y\downarrow&=r_{\ell,r}^{T}(y),&c\uparrow&=r_{r,\ell}^{T}(c),\\ [Q]_T&=b_\rho(r(T),c\mapsto Q(c\uparrow)),& \operatorname{ev}_T(p,y)&=o_\rho(r(T),p,y\downarrow),\\ \operatorname{round}_T(y)&=(y\downarrow)\uparrow. \end{aligned}(18)

Roundtrip respect Resp⁡T(p)\operatorname{Resp}_T(p) and predicate comparison p≃Tqp\simeq_T q are formulas given in (19). Resp⁡T(p)=∀KYy:ℓ(T).(ev⁡T(p,y)↔ev⁡T(p,round⁡T(y))),p≃Tq=∀KYy:ℓ(T).(ev⁡T(p,y)↔ev⁡T(q,y)).\begin{aligned} \operatorname{Resp}_T(p)&= \forall_{K_Y}y:\ell(T). \bigl(\operatorname{ev}_T(p,y)\leftrightarrow \operatorname{ev}_T(p,\operatorname{round}_T(y))\bigr),\\ p\simeq_T q&=\forall_{K_Y}y:\ell(T). \bigl(\operatorname{ev}_T(p,y)\leftrightarrow\operatorname{ev}_T(q,y)\bigr). \end{aligned}(19)

For a type parameter AA and query parameter mm, write Qt(p)=qD,ρt(TA,m,p)Q_t(p)=q_{D,\rho}^t(T_A,m,p) for 0≤t≤60\leq t\leq6.✦ For a predicate pp of type ρ(r(A0))\rho(r(A_0)), admissibility at query parameter mm is the formula Good⁡m(p)\operatorname{Good}_m(p) in (20). Observational equivalence y≈mzy\approx_m z quantifies only over these predicates. Good⁡m(p)=Resp⁡A0(p)∧Q0(p),y≈mz=∀KPp:ρ(r(A0)).(Good⁡m(p)⇒(ev⁡A0(p,y)↔ev⁡A0(p,z))).\begin{aligned} \operatorname{Good}_m(p)&=\operatorname{Resp}_{A_0}(p)\wedge Q_0(p),\\ y\approx_m z&=\forall_{K_P}p:\rho(r(A_0)). \bigl(\operatorname{Good}_m(p)\Rightarrow (\operatorname{ev}_{A_0}(p,y)\leftrightarrow \operatorname{ev}_{A_0}(p,z))\bigr). \end{aligned}(20)

The class predicate [y]m[y]_m and class union [y,z]m[y,z]_m use ≈m\approx_m. Class queries CtmC_t^m and edge formulas EϵmE_\epsilon^m are given by (21). The Boolean ϵ\epsilon is identified with 00 or 11. [y]m=[u↦u≈my]A0,[y,z]m=[u↦(u≈my)∨(u≈mz)]A0,Ctm(y)=Qt+1([y]m)(0≤t≤2),Eϵm(y,z)=C0m(y)∧(C1+ϵm(z)∧Q4+ϵ([y,z]m)).\begin{aligned} [y]_m&=[u\mapsto u\approx_m y]_{A_0},& [y,z]_m&=[u\mapsto(u\approx_m y)\vee(u\approx_m z)]_{A_0},\\ C_t^m(y)&=Q_{t+1}([y]_m)\quad(0\leq t\leq2),\\ E_\epsilon^m(y,z)&=C_0^m(y)\wedge \bigl(C_{1+\epsilon}^m(z)\wedge Q_{4+\epsilon}([y,z]_m)\bigr). \end{aligned}(21)

The pair predicate Pair⁡m(p,a)\operatorname{Pair}_m(p,a) is the encoded formula in (22). Its existential quantifier uses the logical channel type. Pair⁡m(p,a)=[z↦(∃KYy:ℓ(A0).(E0m(y,z)∧ev⁡A0(p,y)))∨E1m(a,z)]A0.\operatorname{Pair}_m(p,a)= \left[z\mapsto \left(\exists_{K_Y}y:\ell(A_0). (E_0^m(y,z)\wedge\operatorname{ev}_{A_0}(p,y))\right) \vee E_1^m(a,z)\right]_{A_0}.(22)✦

The truth marker c⊤(A)c_\top(A), point parameter Point⁡A(y)\operatorname{Point}_A(y), bundle Bundle⁡A,m(y)\operatorname{Bundle}_{A,m}(y), and diamond observation ◇A,m(v,y)\Diamond_{A,m}(v,y) are defined by (23). c⊤(A)=br(A0,⊤),Point⁡A(y)=pt⁡D,ρ(TA,c⊤(A),y↓),Bundle⁡A,m(y)=bun⁡D(MA,dD(TA),m,Point⁡A(y)),◇A,m(v,y)=Obs⁡P(A,v,Bundle⁡A,m(y)).\begin{aligned} c_\top(A)&=b_r(A_0,\top),\\ \operatorname{Point}_A(y)&= \operatorname{pt}_{D,\rho}(T_A,c_\top(A),y\downarrow),\\ \operatorname{Bundle}_{A,m}(y)&= \operatorname{bun}_D(M_A,d_D(T_A),m,\operatorname{Point}_A(y)),\\ \Diamond_{A,m}(v,y)&= \operatorname{Obs}_P(A,v,\operatorname{Bundle}_{A,m}(y)). \end{aligned}(23)

For a type parameter AA, query parameter mm, and probe vv of type ZZ, the probe section Part⁡A,m(v)\operatorname{Part}_{A,m}(v) is the predicate on ℓ(A0)\ell(A_0) with body ◇A,m(v,y)\Diamond_{A,m}(v,y). The section pair query Sb⁡A,m\operatorname{Sb}_{A,m} and element lift Lift⁡A,m\operatorname{Lift}_{A,m} are defined in (24). Part⁡A,m(v)=[y↦◇A,m(v,y)]A0,Sb⁡A,m(v,a)=Q6(Pair⁡m(Part⁡A,m(v),a)),Lift⁡A,m(a)=bℓ(Z,v↦Sb⁡A,m(v,a)).\begin{aligned} \operatorname{Part}_{A,m}(v)&=[y\mapsto\Diamond_{A,m}(v,y)]_{A_0},\\ \operatorname{Sb}_{A,m}(v,a)&= Q_6(\operatorname{Pair}_m(\operatorname{Part}_{A,m}(v),a)),\\ \operatorname{Lift}_{A,m}(a)&= b_\ell(Z,v\mapsto\operatorname{Sb}_{A,m}(v,a)). \end{aligned}(24)

For a predicate ii on ℓ(Z)\ell(Z), the predicate pullback Pull⁡A,m(i)\operatorname{Pull}_{A,m}(i) is the predicate on ℓ(A0)\ell(A_0) obtained by evaluating ii after element lifting.✦ Its recovery Rec⁡A,i\operatorname{Rec}_{A,i} and the uniform predicate probe ziz_i are defined by (25). Pull⁡A,m(i)=[a↦ev⁡Z(i,Lift⁡A,m(a))]A0,Rec⁡A,i(w)=ra⁡D,ρ(TA,w,m↦Pull⁡A,m(i)),zi=Unif⁡P((A,w)↦Rec⁡A,i(w)).\begin{aligned} \operatorname{Pull}_{A,m}(i)&= [a\mapsto\operatorname{ev}_Z(i,\operatorname{Lift}_{A,m}(a))]_{A_0},\\ \operatorname{Rec}_{A,i}(w)&= \operatorname{ra}_{D,\rho}(T_A,w,m\mapsto\operatorname{Pull}_{A,m}(i)),\\ z_i&=\operatorname{Unif}_P((A,w)\mapsto\operatorname{Rec}_{A,i}(w)). \end{aligned}(25)

Put X=ℓ(Z)X=\ell(Z)✦ and 𝒫=ρ(r(Z))\mathcal P=\rho(r(Z)). Probe order i≼xi\preceq x, element comparison x∼Xyx\sim_X y, extensionality Ext⁡(i)\operatorname{Ext}(i), and validity Valid⁡(x)\operatorname{Valid}(x) are the formulas in (26). i≼x=oℓ(Z,x,zi),x∼Xy=∀KPi:𝒫.((i≼x)↔(i≼y)),Ext⁡(i)=∀KYx:X.∀KYy:X.((x∼Xy)⇒(ev⁡Z(i,x)↔ev⁡Z(i,y))),Valid⁡(x)=∀KPi:𝒫.∀KPi′:𝒫.((i≃Zi′)⇒((i≼x)↔(i′≼x))).\begin{aligned} i\preceq x&=o_\ell(Z,x,z_i),\\ x\sim_X y&=\forall_{K_P}i:\mathcal P. ((i\preceq x)\leftrightarrow(i\preceq y)),\\ \operatorname{Ext}(i)&= \forall_{K_Y}x:X.\forall_{K_Y}y:X. \bigl((x\sim_X y)\Rightarrow (\operatorname{ev}_Z(i,x)\leftrightarrow\operatorname{ev}_Z(i,y))\bigr),\\ \operatorname{Valid}(x)&= \forall_{K_P}i:\mathcal P.\forall_{K_P}i':\mathcal P. \bigl((i\simeq_Z i')\Rightarrow ((i\preceq x)\leftrightarrow(i'\preceq x))\bigr). \end{aligned}(26)

Tagged Observations and Structural Formulas

A three-copy construction is a choice of odd source and first paths Ws,WcW_s,W_c and inner and outer double channels Di,DoD_i,D_o.✦ Its first, middle, and final carriers at TT are Wc(T)W_c(T), Di(Wc(T))D_i(W_c(T)), and Do(Di(Wc(T)))D_o(D_i(W_c(T))). For the shared probe take Ws=ℓW_s=\ell, Wc=rW_c=r, T=ZT=Z, and put J=r(Z)J=r(Z), N=Di(J)N=D_i(J), Y=Do(N)Y=D_o(N). The tagged embedding jtj_t is the expression in (27), for t∈{0,1,2}t\in\{0,1,2\}; [t=k][t=k] denotes its Boolean test. jt(x)=eDo[t=2](N,eDi[t=1](J,rℓ,rZ(x))).j_t(x)=e_{D_o}^{[t=2]}\left(N, e_{D_i}^{[t=1]}(J,r_{\ell,r}^{Z}(x))\right).(27)✦

The inner and outer markers βi,βo\beta_i,\beta_o, observation OiO_i, comparison Cmp\operatorname{Cmp}, and tagged observational equivalence ≃Y\simeq_Y are defined by (28). The observation uses the uniform probe ziz_i. βo(y)=βDo(N,y),βi(y)=hDo(N,y,u↦βDi(J,u)),Oi(y)=hDo(N,y,u↦hDi(J,u,w↦or(Z,w,zi))),Cmp⁡(y,z)=∀KPi:𝒫.(Oi(y)↔Oi(z)),y≃Yz=(βi(y)↔βi(z))∧((βo(y)↔βo(z))∧Cmp⁡(y,z)).\begin{aligned} \beta_o(y)&=\beta_{D_o}(N,y),\\ \beta_i(y)&=h_{D_o}(N,y,u\mapsto\beta_{D_i}(J,u)),\\ O_i(y)&=h_{D_o}\bigl(N,y,u\mapsto h_{D_i}(J,u,w\mapsto o_r(Z,w,z_i))\bigr),\\ \operatorname{Cmp}(y,z)&=\forall_{K_P}i:\mathcal P. (O_i(y)\leftrightarrow O_i(z)),\\ y\simeq_Y z&=(\beta_i(y)\leftrightarrow\beta_i(z))\wedge \bigl((\beta_o(y)\leftrightarrow\beta_o(z))\wedge \operatorname{Cmp}(y,z)\bigr). \end{aligned}(28)

Fix a labelled type term A*A_*. The specialized base is I(A*)I(A_*)✦; write E(v,y)=ev⁡I(A*)(v,y)E(v,y)=\operatorname{ev}_{I(A_*)}(v,y).✦ Observation invariance G(v)G(v), class representation Ct(v)C_t(v), and edge representation Et(v)E_t(v) are the formulas in (29). Class indices range from 00 to 22; edge indices used below are 11 and 22. G(v)=∀KYy:Y.∀KYz:Y.((y≃Yz)⇒(E(v,y)↔E(v,z))),Ct(v)=∃KYx:X.∀KYw:Y.(E(v,w)↔(w≃Yjt(x))),Et(v)=∃KYx:X.∀KYw:Y.(E(v,w)↔((w≃Yj0(x))∨(w≃Yjt(x)))).\begin{aligned} G(v)&=\forall_{K_Y}y:Y.\forall_{K_Y}z:Y. \bigl((y\simeq_Y z)\Rightarrow(E(v,y)\leftrightarrow E(v,z))\bigr),\\ C_t(v)&=\exists_{K_Y}x:X.\forall_{K_Y}w:Y. (E(v,w)\leftrightarrow(w\simeq_Y j_t(x))),\\ E_t(v)&=\exists_{K_Y}x:X.\forall_{K_Y}w:Y. \bigl(E(v,w)\leftrightarrow ((w\simeq_Y j_0(x))\vee(w\simeq_Y j_t(x)))\bigr). \end{aligned}(29)

The left-copy predicate ivi_v, relational closure R(v)R(v), and ordered family Φ\Phi of seven structural formulas are defined by (30). iv=[x↦E(v,j1(x))]Z,R(v)=Ext⁡(iv)⇒∀KYx:X.(Valid⁡(x)⇒(E(v,j2(x))⇒(iv≼x))),(Φ0,…,Φ6)=(G,C0,C1,C2,E1,E2,R).\begin{aligned} i_v&=[x\mapsto E(v,j_1(x))]_Z,\\ R(v)&=\operatorname{Ext}(i_v)\Rightarrow \forall_{K_Y}x:X. \bigl(\operatorname{Valid}(x)\Rightarrow (E(v,j_2(x))\Rightarrow(i_v\preceq x))\bigr),\\ (\Phi_0,\ldots,\Phi_6)&=(G,C_0,C_1,C_2,E_1,E_2,R). \end{aligned}(30)

The Definition

Definition 1. A relational code ℛ\mathcal R over SS is a common probe PP with odd logical and predicate paths ℓ,ρ\ell,\rho, as in (31). ℛ=(P,ℓ,ρ),|ℓ|=|ρ|=1.\mathcal R=(P,\ell,\rho),\qquad |\ell|=|\rho|=1.(31)✦ An observation code 𝒪\mathcal O over SS is a relational code ℛ\mathcal R with inner and outer double channels Di,DoD_i,D_o, as in (32). 𝒪=(ℛ,Di,Do).\mathcal O=(\mathcal R,D_i,D_o).(32)✦ A specialized relational code ℬ\mathcal B over SS is an observation code 𝒪\mathcal O with a labelled type term A*A_*, as in (33). ℬ=(𝒪,A*).\mathcal B=(\mathcal O,A_*).(33)✦ For logical code LL and quantifier codes KY,KPK_Y,K_P, the chosen structural query parameter m0m_0 of ℬ\mathcal B is the term in (34). m0=Real⁡D,ρ(TA*,c⊤(A*),(Φ0,…,Φ6)).m_0=\operatorname{Real}_{D,\rho} \bigl(T_{A_*},c_\top(A_*),(\Phi_0,\ldots,\Phi_6)\bigr).(34)✦

What Is Checked

These definitions are stated over labelled terms, retained data and proof paths, and the logical and quantifier codes, with the initial, raw, back, and function path parities specified by the common probe.