# PFABC v0.5.3 Formal Math **Presence-For / Attribution Boundary Calculus** **Contested-Bridge Patch: disagreement is not pooled, and it is not absence** Consolidated specification. This document is standalone: it carries the v0.5.2 bridge-state semantics forward and folds in the v0.5.3 contested-state delta, so nothing else is required to read it. --- ## 0. Executive statement PFABC does **not** calculate consciousness. PFABC calculates the permitted movement of a claim across evidence layers under explicit operating modes, theory families, and warrant purposes. The v0.5.1 correction made PFABC vector-valued and graph-structured rather than scalar-valued: $$ \boxed{PFABC \text{ is a multi-lane warrant object, not a verdict.}} $$ The v0.5.2 correction added bridge-state semantics: $$ \boxed{\text{unmeasured} \ne \text{failed} \ne \text{blocked}.} $$ The v0.5.3 correction adds contested-bridge semantics: $$ \boxed{\text{contested} \ne \text{unmeasured} \ne \text{failed} \ne \text{passed}.} $$ Plainly: > PFABC must not confuse "not yet measured," "measured and failed," "measured repeatedly with results that disagree," "invalidly promoted," and "not supported." These are different epistemic states. A bridge that has been measured more than once, with results that disagree about the threshold, is in a state of its own. It has not failed — some measurement passed. It has not passed — some measurement failed. It is not unmeasured — it has been measured repeatedly. And it must not be pooled into a single number, because pooling is the scalarizing move PFABC already refuses at the verdict level, applied one level down. Refusing to collapse disagreement between measurements is the same operation as `single_status_refused`, applied to the measurement layer. --- ## 1. Core objects Let: $$ \begin{aligned} C &:= \text{set of claims},\\ E &:= \text{space of evidence bundles},\\ \mathcal{T} &:= \text{set of theory families},\\ \tau &\in \mathcal{T} \quad \text{a theory family},\\ \mu &\in \mathcal{M} \quad \text{an operating mode},\\ P_f &:= \text{protected target called presence-for}. \end{aligned} $$ Modes: $$ \mathcal{M}=\{\text{target-realist},\text{ illusionist},\text{ theory-indexed}\}. $$ PFABC keeps the target assumption explicit: $$ T0:\quad \exists P_f \text{ such that } P_f \not\equiv \text{report, memory, access, processing, or behavior by default}. $$ This is not proof of $P_f$. It is the disclosed non-eliminativist operating prior. --- ## 2. Claim-promotion graph PFABC evaluates movement through a claim-promotion graph: $$ G=(V,A) $$ where: - $V$ is the set of claim nodes; - $A$ is the set of directed promotion edges; - each edge is theory-indexed and lane-typed. An edge has the form: $$ a=(p,t,\tau,\lambda,\theta,\delta) $$ where: - $p$ is the proxy/source node; - $t$ is the target claim node; - $\tau$ is the theory family; - $\lambda$ is the lane; - $\theta$ is the bridge threshold; - $\delta\in\{0,1\}$ marks whether the edge is deflationary. A promotion is not allowed merely because the source node is true. $$ Proxy(p,t)\wedge \neg Bridge_\tau(p\to t,E)\Rightarrow BlockTruthPromotion_\tau(p\to t) $$ --- ## 3. Bridge states Evidence no longer supplies at most one strength per edge. It supplies a finite multiset of measurements: $$ M_E(a)\subset\mathbb{R},\qquad a\in Measured(E)\iff M_E(a)\ \text{is declared.} $$ **Non-emptiness side condition.** A declared measurement set must be non-empty: $$ a\in Measured(E)\Rightarrow |M_E(a)|\ge 1. $$ An edge for which nothing was measured is expressed by *omission* ($a\notin Measured(E)$), never by declaring $M_E(a)=\varnothing$. A declared empty set is ill-formed input and must be rejected, not evaluated — note that $\forall m\in\varnothing:m\ge\theta_a$ holds vacuously, so an unguarded reading of the **passed** case turns no measurement into warrant. *Conformance:* enforced by the reference engine as of **v0.5.3.1**. The engine as originally authored evaluated $\varnothing$ as **passed** — finding **F1** in `HOSTILE_REVIEW_v0_5_3.md`. Measurements are real: $M_E(a)\subset\mathbb{R}$. A non-real value has no position relative to $\theta_a$ and is ill-formed on the same grounds. `NaN` is the sharp case — it compares false against $\theta_a$ in *both* directions, so it satisfies neither the **passed**, **failed**, nor **contested** condition, and an unguarded state function drops it into **failed** by elimination. *Conformance:* `NaN` rejected as of **v0.5.3.2** — finding **F6**. $\pm\infty$ is **not** rejected: strictly outside $\mathbb{R}$, but it holds a definite position relative to any finite $\theta_a$, so it does not misfire. A conformance nit, recorded rather than closed. For a bridge edge $a$, evidence may now put it in one of **five** states: $$ state_E(a)\in\{\textsf{passed},\textsf{unmeasured},\textsf{contested},\textsf{failed},\textsf{invalid}\}. $$ The state function is: $$ state_E(a)= \begin{cases} \textsf{invalid}, & a\in Invalid(E),\\[2pt] \textsf{unmeasured}, & a\notin Measured(E),\\[2pt] \textsf{contested}, & \exists m,m'\in M_E(a):\ m\ge\theta_a\ \wedge\ m'<\theta_a,\\[2pt] \textsf{passed}, & \forall m\in M_E(a):\ m\ge\theta_a,\\[2pt] \textsf{failed}, & \forall m\in M_E(a):\ m<\theta_a. \end{cases} $$ The three measurement-bearing cases are exhaustive and mutually exclusive given $|M_E(a)|\ge1$. A missing measurement is still not a failed bridge. It is an unresolved bridge. A disagreeing set of measurements is neither. **No pooling operator is defined.** There is deliberately no $$ pool:\ M_E(a)\to\mathbb{R} $$ in the calculus. Mean, median, weighted average, inverse-variance meta-analysis, and "most recent measurement wins" are all excluded. A disagreement is a fact about the evidence base, and PFABC reports it rather than resolving it. **Contest is threshold-relative, not dispersion-relative.** $M=\{0.801,0.799\}$ at $\theta=0.8$ is **contested**; $M=\{1.0,0.81\}$ at the same threshold is **passed**, despite far wider spread. This is intended — PFABC asks whether the measurements disagree *about the promotion*, not whether they disagree about the world — but it means threshold placement, not measurement quality, determines contest. A threshold moved by $\varepsilon$ can change the state. Thresholds must therefore be fixed before measurements are declared, and this requirement is now load-bearing in a way it was not in v0.5.2. A threshold changed after the fact is a discourse-audit event in its own right. --- ## 4. Path states Define the within-path severity order: $$ \textsf{SATISFIED}\prec\textsf{UNMEASURED}\prec\textsf{CONTESTED}\prec\textsf{FAILED}\prec\textsf{INVALID}. $$ For a path $\pi=(a_1,\dots,a_n)$ from an evidence-supported source to a target, the path state is the maximum edge severity: $$ PathState_E(\pi)=\max_{\prec}\ \{\,\sigma(state_E(a_i))\ :\ a_i\in\pi\,\} $$ where $\sigma$ lifts an edge state to its path-level name. Equivalently, the worst edge governs the path: one contested edge makes the whole path contested, and $$ \exists a_i:state_E(a_i)=\textsf{contested}\ \Rightarrow\ PathState_E(\pi)\ne\textsf{SATISFIED}. $$ A contested bridge can never satisfy a path. One passing measurement does not carry a claim across a bridge that another measurement says it cannot cross. Special case: direct evidence for a non-protected target may produce a zero-edge satisfied path. Direct evidence for the protected target $P_f$ is not allowed to seed itself. $$ P_f\in Seeds(E)\Rightarrow RemoveSeed(P_f)\wedge Audit(\text{protected-target-as-evidence}) $$ --- ## 5. Practical truth status For a claim $c$ under theory $\tau$, let $\Pi_{\tau}(c,E)$ be all paths from evidence-supported seeds to $c$ in the truth lane. Define the **claim-level precedence** order: $$ \textsf{WARRANTED}\succ\textsf{CONTESTED}\succ\textsf{UNRESOLVED}\succ\textsf{UNWARRANTED}\succ\textsf{BLOCKED} $$ and take the first status whose condition is met: $$ T^{prac}_{\tau}(c,E)= \begin{cases} \textsf{WARRANTED}, & \exists\pi\in\Pi_{\tau}: PathState_E(\pi)=\textsf{SATISFIED},\\[2pt] \textsf{CONTESTED}, & \exists\pi: PathState_E(\pi)=\textsf{CONTESTED}\ \wedge\ \nexists\pi': PathState_E(\pi')=\textsf{SATISFIED},\\[2pt] \textsf{UNRESOLVED}, & \exists\pi: PathState_E(\pi)=\textsf{UNMEASURED}\ \wedge\ \nexists\pi': PathState_E(\pi')\in\{\textsf{SATISFIED},\textsf{CONTESTED}\},\\[2pt] \textsf{UNWARRANTED}, & \exists\pi: PathState_E(\pi)=\textsf{FAILED}\ \wedge\ \nexists\pi': PathState_E(\pi')\in\{\textsf{SATISFIED},\textsf{CONTESTED},\textsf{UNMEASURED}\},\\[2pt] \textsf{BLOCKED}, & \Pi_{\tau}\ne\varnothing\ \wedge\ \forall\pi\in\Pi_{\tau}: PathState_E(\pi)=\textsf{INVALID},\\[2pt] \textsf{UNWARRANTED}, & \Pi_{\tau}=\varnothing. \end{cases} $$ **The two orders are not duals.** The within-path order of §4 and the claim-level order above must not be read as one ordering seen twice. Within a path the *worst* edge governs, because a path is a conjunction: every bridge on it has to hold. Across paths the ordering is by **salience**, not by optimism, because the set of paths is a disjunction and the output has to say which fact about the evidence base most needs reporting. **Why `CONTESTED` outranks `UNRESOLVED`.** A live disagreement is more informative than an unmeasured gap. Reporting a contested claim as "nobody has measured this yet" would understate the evidential situation and misdirect the next test. A field that has measured a bridge four times with conflicting results is not in the position of a field that has never measured it, and the output must say so. Mixed-invalid rules, carried from v0.5.2 and extended to the new state: $$ \exists\pi_i:\textsf{INVALID}\ \wedge\ \exists\pi_c:\textsf{FAILED}\Rightarrow T^{prac}=\textsf{UNWARRANTED}+Audit $$ $$ \exists\pi_i:\textsf{INVALID}\ \wedge\ \exists\pi_c:\textsf{UNMEASURED}\Rightarrow T^{prac}=\textsf{UNRESOLVED}+Audit $$ $$ \exists\pi_i:\textsf{INVALID}\ \wedge\ \exists\pi_c:\textsf{CONTESTED}\Rightarrow T^{prac}=\textsf{CONTESTED}+Audit $$ not `BLOCKED` in any of the three. `BLOCKED` is reserved for invalid-only truth support. **Settled (finding F2).** The rule above lets a single $\textsf{SATISFIED}$ path return $\textsf{WARRANTED}$ while a sibling path to the same target is $\textsf{CONTESTED}$. This is **correct and is now settled**, not a holding position. The objection was that PFABC refuses to let one passing measurement carry a bridge, so it should equally refuse to let one passing bridge carry a claim. That parallel does not hold, and seeing why fixes the shape of the calculus: - **Within an edge**, the members of $M_E(a)$ are competing estimates of *one quantity* — the strength of bridge $a$. When they disagree, that quantity is not known. Refusing to pool is refusing to fabricate a determinate value where none exists. The edge is a **conjunction** of its measurements agreeing. - **Across paths**, $\pi_1$ and $\pi_2$ are *different objects*. "$\pi_1$ is satisfied" and "$\pi_2$ is contested" are two independent facts, both fully known, and not in tension. $\Pi_{\tau}(c,E)$ is a **disjunction** of routes. There is no ambiguity to collapse, so "refusing to collapse" has nothing to act on. Treating a disjunction as though it were a conjunction would be its own category error — the mirror image of the one PFABC exists to prevent. Two further reasons the decision goes this way: 1. **The calculus already answers it for a worse case.** $\textsf{INVALID}$ is strictly more damning than $\textsf{CONTESTED}$: it is a category error in the promotion, not a disagreement about a measurement. Yet $\exists\pi:\textsf{INVALID}\ \wedge\ \exists\pi':\textsf{SATISFIED}\Rightarrow\textsf{WARRANTED}+Audit$ — the invalid sibling raises an audit and does not demote the satisfied route. Demoting for a contested sibling while not demoting for an invalid one would be incoherent. 2. **Demotion would be exploitable.** If a contested sibling could reduce a claim's status, anyone could degrade any warranted claim by publishing conflicting measurements on a *different* bridge, leaving the clean route untouched. Warrant would become a function of how much noise surrounds a claim rather than of the evidence for it. **What the decision does not license.** The contest is not thereby unimportant — it stays in $X_{\tau}$, in the theory notes, and in the next-test list, none of which key off $T^{prac}_{\tau}$. The correct reading of the output is *"this claim has an established clean route, and a separate route to it is in dispute."* Both halves are reported, and neither is allowed to erase the other. That was the real defect, and it is fixed. **No mixed status is introduced,** on the same grounds: a compound value such as `WARRANTED_CONTESTED` would encode two independent facts in one enum, which is the scalarizing move the product type exists to refuse. What is **not** open, and is implemented as of engine **v0.5.3.3**, is that the contest stays visible whatever the status. Whenever any path evaluation is contested, the disagreement is reported in the theory notes, named in `next_tests`, and enumerated in the contested-bridge set $X_{\tau}$ of §9 — keyed off the contested set, never off the claim-level status. A satisfied sibling path does not erase a live disagreement. --- ## 6. BLOCK is lane-local A blocked truth promotion does not erase research or precaution warrant. $$ Invalid^{truth}(c,E)\Rightarrow Audit(c,E) $$ but: $$ Research(c,E) \text{ and } Precaution(c,E) \text{ still evaluate independently.} $$ Example: $$ \text{biohybrid tissue}\to\text{consciousness} $$ may be blocked in the truth lane while still being strong in research and caution in precaution. --- ## 7. Truth lane split Truth is not one thing. $$ Truth_{\mu,\tau}(c,E)=\langle T^{prac}_{\tau}(c,E),T^{meta}_{\mu,\tau}(c,E)\rangle. $$ For the protected target $P_f$: $$ T^{meta}_{\mu,\tau}(P_f,E)= \begin{cases} \textsf{DEFLATED}, & \mu=\text{illusionist},\\ \textsf{UNBRIDGED}, & \mu\in\{\text{target-realist},\text{theory-indexed}\}. \end{cases} $$ For non-protected targets: $$ T^{meta}_{\mu,\tau}(c,E)=\textsf{NOT\_APPLICABLE}. $$ Thus a functional theory may warrant a practical claim under its own standard while metaphysical presence remains unbridged. --- ## 8. Deflation flag A deflation flag fires when a path toward the protected target relies on a deflationary bridge. $$ F_{\mu,\tau}(c,E)=1 $$ iff: $$ c=P_f \wedge \mu=\text{target-realist}\wedge \exists\pi\in\Pi_{\tau}(P_f,E): \Big(Deflationary(\pi)\wedge PathState_E(\pi)\in\{\textsf{SATISFIED},\textsf{UNMEASURED},\textsf{CONTESTED}\}\Big). $$ $\textsf{CONTESTED}$ is new to this set in v0.5.3 and belongs there for the same reason $\textsf{UNMEASURED}$ does: the deflationary route to the protected target is live but not settled. Omitting it produced an incoherence — the flag fired when nobody had measured the bridge, and stayed silent when some measurements said it passed, which is the stronger case. $\textsf{FAILED}$ and $\textsf{INVALID}$ stay out, unchanged: a bridge that failed did not reach the target, so nothing was deflated. A deflation flag does not mean the theory is useless. It means the theory reaches the protected target by redefining it as function, access, behavior, integration, or architecture. --- ## 9. Full warrant object The PFABC output remains a product type. v0.5.3 adds one field to the per-theory report, so that contested bridges are enumerable rather than recoverable only by walking path evaluations: $$ W_{\mu,\tau}(c,E)= \left\langle T^{prac}_{\tau}, T^{meta}_{\mu,\tau}, R_{\tau}, P^{prec}, A^{disc}, F_{\mu,\tau}, X_{\tau} \right\rangle. $$ Where: - $T^{prac}$ = theory-indexed practical truth; - $T^{meta}$ = metaphysical presence status; - $R$ = research warrant; - $P^{prec}$ = precaution warrant; - $A^{disc}$ = discourse audit; - $F$ = deflation flag; - $X$ = contested-bridge set. **Notation.** Two symbols were previously overloaded, and are disambiguated here. $\Pi$ denoted both the precaution warrant and the path set $\Pi_{\tau}(c,E)$; $A$ denoted both the edge set of the graph $G=(V,A)$ and the discourse audit. The rule now is that **a symbol is either structural or lane-valued, never both**: structural symbols keep the bare letter, and the lane-valued reading takes a lane superscript, matching the $T^{prac}/T^{meta}$ convention already in use. So $A$ is the edge set and $A^{disc}$ the audit; $\Pi_{\tau}$ is the path set and $P^{prec}$ the precaution warrant. $R$ and $F$ never collided and are unchanged. The contested-bridge set is: $$ X_{\tau}=\{a\ :\ a\in\pi\ \text{for some}\ \pi\in\Pi_{\tau}(c,E),\ state_E(a)=\textsf{contested}\}. $$ $X_{\tau}$ collects the edges whose **own** state is contested, not every edge of a contested path. An edge that is both contested and invalid is invalid — the state function of §3 tests $Invalid(E)$ first — and does not appear in $X_{\tau}$. No field is allowed to erase another. $X_{\tau}\ne\varnothing$ in particular must not be erased by $T^{prac}_{\tau}=\textsf{WARRANTED}$. *Conformance:* implemented as of **v0.5.3.3** — finding **F3**. `TheoryReport.contested_bridges` carries $X_{\tau}$; `PathEvaluation.contested_edges` records which edges of a given path are in disagreement. Both serialize. --- ## 10. Top-level PFABC function $$ PFABC_{\mu}(c,E,\mathcal{T})= \left\{ \begin{array}{ll} mode &: \mu,\\ claim &: c,\\ discourse\_audit &: A^{disc}(c,E),\\ per\_theory &: \{\tau\mapsto W_{\mu,\tau}(c,E)\mid \tau\in\mathcal{T}\},\\ conflict\_profile &: Conflict(\{W_{\mu,\tau}\}),\\ depth\_profile &: D(E),\\ safer\_language &: Safer(c,E),\\ next\_tests &: Tests(c,E),\\ single\_status\_refused &: \textsf{true}. \end{array} \right\} $$ The contested-bridge set is per-theory and therefore lives inside $W_{\mu,\tau}$, not at top level. The top-level shape is unchanged from v0.5.2. --- ## 11. Depth remains vector-valued Depth of consequence is a profile, not a score. Let: $$ D=\{d_1,\dots,d_n\} $$ where each dimension $d_i$ has: $$ \langle apparent_i, proxy_i, discount_i\rangle. $$ The discounted dimension is: $$ value_i = apparent_i(1-proxy_i). $$ But there is no authorized scalar: $$ \nexists S:D\to\mathbb{R} \text{ that PFABC treats as a verdict.} $$ A display may count cleared dimensions, but only as a display summary: $$ cleared(D)=\{d_i:value_i\ge\theta_i\wedge proxy_i\le \rho_i\}. $$ This must not be presented as consciousness probability. --- ## 12. Homecoming as context-altered relevance Let $O_t$ be orientation state at time $t$, $e_t$ an admitted episode, and $\pi$ a valid provenance path. A lawful deformation is: $$ O_{t+1}=Deform(O_t,e_t,\pi). $$ A future datum $x$ matters through orientation: $$ M_t(x)=Evaluate(x,O_t). $$ Homecoming-like structure requires: $$ M_{t+k}(x)=Evaluate(x,O_{t+1})\ne Evaluate(x,O_t) $$ and the difference must be counterfactually dependent on $e_t$ under controls. Thus: $$ Homecoming(e_t,x)= \begin{cases} \textsf{candidate}, & \text{altered relevance survives no-raw/no-summary controls},\\ \textsf{unwarranted}, & \text{effect vanishes under retrieval ablation},\\ \textsf{contested}, & \text{repeated runs disagree about the bridge},\\ \textsf{unresolved}, & \text{bridge is unmeasured}. \end{cases} $$ The contested case is new in v0.5.3. Repeated tests of the same homecoming bridge that disagree about its threshold are not pooled into an effect size, are not reported as an unmeasured bridge, and are not reported as a failed one. --- ## 13. Learned-carrier control as bounded engineering analogue A governed learned-carrier lane has the form: $$ \text{episode}\to\text{non-source receipt}\to\text{admission}\to\text{scoped carrier}\to\text{later behavior}. $$ The engineering claim is not consciousness. It is: $$ CarrierBehavior(y) \text{ depends on admitted receipt } r $$ while: $$ RawText, Summary, RAG, Persona, WrongCarrier, ProtectedRoute \not\Rightarrow y. $$ A positive result in this lane can support research warrant for engineered returning alteration, but it does not bridge metaphysical presence-for. --- ## 14. Summary of corrections Carried from v0.5.2 and still in force: | Problem | v0.5.2 correction | |---|---| | scalar verdicts | product/vector warrant object | | lane fusion | parallel truth/research/precaution/audit | | global block | lane-local block and audit | | untyped warrant | practical vs metaphysical truth split | | unmeasured = failed | explicit `UNRESOLVED` bridge state | | invalid path dominates clean failing path | clean failing path gives `UNWARRANTED + AUDIT` | | protected target seeds itself | protected target seed removed and audited | | deflation hidden | deflation flag on deflationary paths to protected target | | depth score | depth vector/profile only | Added in v0.5.3: | Problem | v0.5.3 correction | |---|---| | repeat measurements pooled into one strength | finite multiset of measurements per edge; no pooling operator defined | | one passing measurement carries the bridge | `contested` edge can never satisfy a path | | disagreement reported as absence | `CONTESTED` outranks `UNRESOLVED` at claim level | | disagreement reported as failure | `CONTESTED` distinct from `UNWARRANTED` | | contest erased by a satisfied sibling path | contested-bridge set reported whatever the claim-level status | | "run one more study and declare victory" | next-test text names the disagreement as the object of study | --- ## 15. Plain-English version PFABC v0.5.3 does not ask for one answer. It asks: 1. What claim is being made? 2. What evidence supports which lower nodes? 3. Which bridge would be needed for the claim to travel? 4. Is the bridge passed, unmeasured, contested, failed, or invalid? 5. If it has been measured more than once, do the measurements agree about the threshold? 6. Which lane is affected: truth, research, precaution, discourse audit? 7. Does any theory reach the protected target only by deflating it? 8. What remains unresolved, and what remains contested? The core correction: > A missing bridge measurement is not a failed bridge. A set of bridge measurements that disagree is neither a missing measurement nor a failed one, and it is never averaged away. An invalid promotion is not a global veto. And a successful functional bridge to presence-for is not metaphysical proof unless the target has been deflated. As in every version: PFABC does not calculate consciousness. It calculates what a claim is permitted to do, in which lane, under which theory, on the evidence actually declared. --- ## 16. Scope note — what v0.5.3 does not fix The v0.5.3 patch header cited three reference-engine defects by identifier. All three are now resolved and the scope note is empty: two were defined nowhere and their citation was deleted (finding F7), and the third — `PracticalTruth.PARTIAL` remaining unreachable — was resolved by **removing the member** in engine v0.5.3.7 (finding F8). It was absent from v0.5.1, added to the enum in v0.5.2, and never assigned by any version: dead from birth rather than a regression. The enum now contains exactly the five statuses the engine can emit, and a reachability test enforces that. The v0.5.3 patch header also cited two further reference-engine defects by identifier. They were defined nowhere in the lineage, their referents are lost, and the citation has been **removed** rather than carried forward — an identifier with nothing behind it can be neither closed nor shown still open (finding F7, closed). Neither was a defect in the calculus. `HOSTILE_REVIEW_v0_5_3.md` is the authoritative findings ledger and continues past this document. The findings that bear on this specification are: - **F1** — empty declared measurement set evaluated as passed. *Engine conformance gap against §3.* **Closed in v0.5.3.1.** - **F2** — cross-path contest collapse. Two separable questions. The reporting half — that a contest stays visible in notes, in `next_tests`, and in $X_{\tau}$ regardless of claim-level status — is **closed in v0.5.3.3**. The status half — whether a satisfied sibling path should change the claim-level status — is **settled in favour of $\textsf{WARRANTED}$**: paths are a disjunction, so there is no ambiguity to collapse. See §5. - **F3** — $X_{\tau}$ specified in §9 and absent from the engine. *Reporting gap.* **Closed in v0.5.3.3.** - **F4** — `bridge_strengths` type annotation still read `Dict[str, float]` after the field began accepting sequences. **Closed in v0.5.3.3.** - **F5** — threshold placement now carries more weight. Not a defect but a design consequence, written into §3. - **F6** — `NaN` measurements evaluated as failing rather than rejected. *Engine conformance gap against §3.* **Closed in v0.5.3.2.** The calculus is v0.5.3 throughout; the patch version tracks engine conformance only. No question about the calculus remains open. With F1 and F6 closed, no ill-formed measurement input can reach a state assignment. With F3 and the reporting half of F2 closed, no contest can be silently erased from the output.