Work

"If It Were There, Would We Have Seen It?": Dharmakīrti's Fail-Closed Pramāṇa Logic, Eleven-Mode Anupalabdhi, 9-Cell Hetucakra, and Anyāpoha Heyting Algebra

Year: 2026
Type: preprint
Abstract: ## Abstract When does failing to observe an object justify certifying its absence, and when must the exact same empty observation trigger epistemic abstention? In the 7th century CE, **Dharmakīrti** gave a remarkably tight, fail-closed answer across the *Nyāyabindu* (`NB`), *Pramāṇavārttika* (`PV`), *Hetubindu* (`HB`), and *Sambandhaparīkṣā* (`SP`): non-perception certifies absence *only* when the stated observation conditions would have revealed the target had it been present (*upalabdhilakṣaṇaprāpta*, `NB` II.32); otherwise (`NB` II.46–48), the inference must abstain. Likewise, observing zero counterexamples in a finite sample (*adarśanamātra*) never certifies a universal rule without a structural grounding link (*svabhāvapratibandha*). Building on Dignāga's 3×3 *Hetucakraḍamaru*—where Z3 verifies an exact **$S \leftrightarrow \neg S$ involution symmetry** swapping valid cells (`#2, #8`) identically onto *viruddha* cells (`#4, #6`) while preserving the five-cell *anaikāntika* verdict class setwise (fixing diagonal cells `#1, #5, #9` and swapping `#3 <-> #7`)—`S2-P39` turns Dharmakīrti's reasoning into a reusable executable verifier (`core/dharmakirti_pramana_engine.py`) inside Epistemic Life Engine v4.1 (`epistemic_v4.1_harness`), extending our Sautrāntika calculus from `S2-P35` and modeling Dharmakīrti's anti-realist contrast with Nyāya *abhāva* (real absence) via intuitionistic double-negation closure ($\neg\neg u \neq u$). Strictly partitioning zero-parameter symbolic verification (**Module 1, Tier M1**) from disclosed pedagogical simulations (**Module 2, Tier M3**, `is_zero_parameter_derivation = false`) and a pre-registered OOD empirical benchmark (**`AXIS_B_PENDING`**, **Tier B4**), our executable harness certifies six audited results: - **Theorem DK.I (Z3 *Hetucakra* Involution Symmetry, *Trairūpya* Side-Condition, *Adarśanamātra* Gap, & *Viruddhāvyabhicārin* Impossibility, Tier M1; Tier B1).** Z3 verifies all $3 \times 3 = 9$ cells of Dignāga's *Hetucakraḍamaru* as realizable over finite domains (`2` valid `#2, #8`; `2` *viruddha* `#4, #6`; `5` *anaikāntika* `#1, #3, #5, #7, #9`) and proves an exact **$S \leftrightarrow \neg S$ involution symmetry**: exchanging *sapakṣa* $\leftrightarrow$ *vipakṣa* maps Valid $\leftrightarrow$ *Viruddha* (`2 <-> 4`, `8 <-> 6`) while preserving the *Anaikāntika* class setwise (`1, 5, 9` fixed; `3 <-> 7` swapped; `UNSAT` counterexample check). Z3 proves universal *trairūpya* entailment with independent *sapakṣa* witness $w \neq p$ (a well-formedness side-condition blocking Cell 5 *asādhāraṇa*), proves Dharmakīrti's **inductive gap theorem (*adarśanamātra*)** via a `SAT` unobserved *vipakṣa* countermodel $u_{\ast}$, and proves **`NB` III.110–116**: two simultaneously grounded *svabhāvapratibandha* reasons proving $S(p)$ and $\neg S(p)$ are jointly `UNSAT`.- **Theorem DK.II (Two *Modus Tollens* Schemas Across 11 Canonical *Anupalabdhi* Instantiations & *Adṛśya* Abstention, Tier M1; Tier B1/B2).** Organizing Dharmakīrti's **11 canonical modes (`NB` II.31–44)** as instantiations of two *modus tollens* schemas (`11/11 UNSAT`), our Z3 encoding distinguishes **4 perceptibility-conditioned non-perception instantiations** (`#1, #2, #3, #9`, requiring `upalabdhilaksanaprapta = True`) from **7 opposite-perception (*viruddhopalabdhi*) instantiations** (`#4–8, #10–11`, deriving $\neg X$ from positively perceiving an incompatible opposite, doctrinally unified under *svabhāvānupalabdhi* at `NB` II.43–44). Violating perceptibility (`NB` II.46–48, the imperceptible *Piśāca* rule) leaves both $X(p)$ and $\neg X(p)$ `SAT`, triggering `ABSTAIN_ADRSYA_IMPERCEPTIBLE` in `DharmakirtiPramanaAuditor` and fail-closed `ABSTAIN_UNDERSPECIFIED` in `EpistemicProvenanceLedger` (contrasted against `FORMALLY_PROVEN` for perceptible *dṛśya* pot absence at `NB` II.32).- **Theorem DK.III (*Anyāpoha* Heyting Algebra, Nyāya *Abhāva* Contrast [$\neg\neg u \neq u$], Glivenko Regularized Join, & Zero-Universal *Ekārthakriyā*, Tier M1; Tier B1/B2).** Modeling Dharmakīrti's contrast with Nyāya *abhāva* (real ontological absence where *abhāvābhāva* collapses directly to positive presence) alongside *anyāpoha* double exclusion (`Tier B2` algebraic analogy), we verify all `125/125` residuation triples on the 5-element Alexandrov open-set topology $\mathcal{O}(X)$ and `27/27` triples on the 3-element epistemic chain $H_3 = \{\bot < u_{\text{adṛśya}} < \top\}$, certify all **Glivenko / Lawvere–Tierney double-negation nucleus axioms** ($a \le \neg\neg a$, $\neg\neg\neg a = \neg a$, $\neg\neg(a \wedge b) = \neg\neg a \wedge \neg\neg b$), exhibit strict failure of Excluded Middle ($u \vee \neg u \neq \top$) and strict separation of double absence from raw presence ($\neg\neg u = \top \neq u$), and prove that $\operatorname{Reg}(\mathcal{O}(X))$ exits closure under raw set union ($\{c_1\} \cup \{h_1\} \notin \operatorname{Reg}(\mathcal{O}(X))$) yet becomes a complete Boolean algebra specifically under the **Glivenko-regularized join** $a \vee_{\text{Reg}} b := \neg\neg(a \cup b)$. Over 6 pairwise distinct momentary particulars (*svalakṣaṇas*), double exclusion recovers exact *ekārthakriyā* task classes with **`0` reified universal entities**, while SymPy proves $Q_{\text{apoha}}^2 - Q_{\text{apoha}} = \mathbf{0}_{6\times 6}$ (`rank = trace = 2`, 50-step discrete residual `0.0`).- **Proposition DK.IV (Disclosed-Parameter Four-Regime *Apoha* & *Bhrānti* Simulations, Tier M3; Tier B3).** With explicitly disclosed pedagogical constants (`ksana_drift_omega = 0.45`, `apoha_learn_rate = 0.16`, `bhranti_noise_low = 0.10`, `bhranti_noise_high = 0.85`, `ood_shift_step = 25`, `exclusion_margin_gain = 0.35`, `horizon_steps = 60`; zero unused parameters, `is_zero_parameter_derivation = false`), four regimes verify: (1) strict momentary drift (`min_svalaksana_step_norm = 0.426213 > 0`) paired with exponential *apoha* convergence (`apoha_error_t60 = 2.862577e-05`, analytical diff `3.39e-21`); (2) *bhrānti* sensory distortion transition (`0.0` vs. `0.21` error rate); (3) post-OOD collapse of *adarśanamātra* correlation (`1.0` false-positive rate vs. `0.0` under *svabhāvapratibandha*); and (4) *apoha* exclusion margin rejection of novel *atadvyāvṛtti* distractors (`1.0` vs. `0.0` for an unbounded positive-prototype half-space).- **Theorem DK.V (Layer $-1$ Farkas Deadlocks, Nyāya *Svarūpa-Sambandha* Steel-Man Control, & Layer 2b Schema-Consistency Pushout, Tier M1; Tier B1/B2).** With every `>= 1` inequality normalized to `<= -1` (so inequality multipliers remain strictly positive while `-1` multiplies the equality constraint), exact integer Farkas duals certify three deadlocks matching §4.1's constraint order verbatim: *Sattvānumāna* permanent-existent impossibility `(1, 1, 1, -1, 2)`, *Sambandhaparīkṣā* cap-1 relation regress `(1, -1, 1)`, and grounded *viruddhāvyabhicārin* antinomy `(1, 1, -1, 1)`—each canceling identically to $0 \le -1$—while Filter A rejects an unanchored recall-vs.-cost preference. Steel-manning Nyāya via a Z3 control (`nyaya_svarupa_closure_sat = true`) isolates **ontological parsimony** (`0` extra *dravyas*/class in Dharmakīrti vs. `+2` in Nyāya *sāmānya* + *samavāya*) as the decisive separation. The 3-domain Layer 2b canonical schema-consistency colimit (`26` coords, `26` eqs, `ASSUMED = 0`, `DOF = 0`, `Phi_syn = 1.0`, `connected_components = 1`, pre-pinned Gate 1 SHA-256 `d905faea...`, `3/3` Farkas mutation kills) passes all five `PostPushoutVerifier` gates, while both Red-Team controls (`E8_Hetucakra_9_Root_Numerology_Slot` and `Nyaya_Samanya_Dravya_Universal_Slot`) fail Gate 5 fail-closed (`Phi_syn = 0.0`, `connected_components = 2`).- **Theorem DK.VI & Protocol DK.VII (Layer 6 SAT Manifold, Five Zero-Z3 DPLL Proofs, Minimax Axiom-Profile Discrimination Test, & Tier B4 Kill-Switch, Tier M1 / M2 / B4).** A 10-variable, 12-clause canonical *Pramāṇa*/*Sattvānumāna* CNF matches exhaustive $2^{10} = 1024$ enumeration identically (`50` models, `frozen_vars = []`, `overlap_mean = 0.1755`, `REPLICA_SYMMETRIC`), paired with **5 standalone zero-Z3 pure-Python DPLL + explicit Resolution proofs** (`VALID_RESOLUTION_REFUTATION`), including a 6-step *Sattvānumāna* refutation and a 6-step multi-hop refutation proving an eternal universal cannot be an object of direct perception (*pratyakṣa*). A Layer 7 diagnostic harness test of `ActiveRuleDiscoverer` confirms single co-occurrence leaves all 4 formal epistemological profiles indistinguishable (`AMBIGUOUS_HYPOTHESIS_CLASS`), whereas 3 minimax probes (combining one *pañcakāraṇī* OOD counterfactual probe with two explicit design-invariant checks on abstention and parsimony) isolate `H1_DHARMAKIRTI_PRAMANA_SVABHAVAPRATIBANDHA` uniquely (`METAMORPHIC_PASS`). Genuine empirical falsification sits in **Protocol DK.VII**, registering a pre-registered OOD contrastive-exclusion Pass Condition and $H_0$ Null Kill-Switch (`AXIS_B_PENDING`, Tier B4).
Author: Oo, Myo
Language: English
Open Access status: green
OpenAlex ID: W7237335291
Cites:
Cited by:
Related to: