# Publication navigation: direct standalone route and historical tracks The current researcher entry is the [Conditional Separation source package](conditional_separation/README.md) (subdirectory `conditional_separation/`), with a [mathematical introduction to R](conditional_separation/PROOF_MAP.md#research-note) and a [two-root partial-lemma exchange recipe](conditional_separation/README.md#partial-work-across-machines). The known public starting revision is [`a86a6ce9720b71a6bd5439c90d86edd6b6bc25b6`](https://github.com/khanukov/p-np2/tree/a86a6ce9720b71a6bd5439c90d86edd6b6bc25b6/conditional_separation); this documentation follow-up retains that PR1822 ancestry and does not merge it. For any distributed package, identify the exact reviewed release revision and complete package inventory; branch names and old CI do not certify new bytes. | Current standalone surface | Boundary | |---|---| | [`OpenProblem.R`](conditional_separation/ConditionalSeparation/OpenProblem.lean) | UNPROVED: one fixed threshold exponent k defeats every polynomial degree of nonuniform DAG promise-search solvers on full tables | | [`R_iff_original`](conditional_separation/ConditionalSeparation/Equivalence.lean) | Definitional equality of the exact problem, not a lower bound | | [`close (hR)`](conditional_separation/ConditionalSeparation/Conditional.lean) | Conditional SAME-k/SAME-language Ordinary NP, non-PpolyDAG and non-Ordinary-P conclusion; no legacy ContentVerifierBridge premise | | [Partial lemma workflow](conditional_separation/README.md#partial-work-across-machines) | Compile a useful closed lemma and its named check; transfer source/identity; final checker must still reject the partial-only attempt | This is **Infrastructure**, not a proof of R, unconditional P≠NP, a novel magnification theorem, an established equivalence to every external model, or Lean-team/mathlib/institutional endorsement. Source review, exact-head native Linux/macOS/install evidence and release/publication decisions remain separate. The [old Preprint](Preprint.latex) describes a historical SAL/anti-checker route; it is not the accompanying direct-route paper. Its retained claims are not current release evidence. The research note states the current quantifiers, codec, prefix language and literature limitations explicitly. ## Historical PNP3 RC notes (17 August 2026) The following material is retained for the legacy PNP3 release/audit track. Its `ResearchGapWitness` and verifier/runtime/advice boundaries are not extra hypotheses of the direct standalone endpoint above. Links and historical audit commands below preserve that scope; they are not fresh execution receipts. - Release type: **RC / framework milestone**, not a final unconditional claim. - Status date: 2026-08-17. - Source of truth for release wording: `RELEASE_RC.md`. - Source of truth for closure progress: `CHECKLIST_UNCONDITIONAL_P_NE_NP.md`. ## Claim matrix | Claim | Status | |-----------------------------------------------------------------------|--------------------------------------------------------------| | Active `pnp3/` has no project-local axioms and no `sorry`/`admit` | Verified by `scripts/check.sh` policy | | `P_subset_PpolyDAG` internalized | Proved, but coarse polynomial-size inclusion | | `ResearchGapWitness` → `P_ne_NP_final` | Proved conditional bridge | | Old support-bounds / multi-switching route | Formally refuted; retained only as audit / compatibility | | `FormulaSupportBoundsPartial_fromPipeline_fixedParams` | Candidate contract shape, not a proved source theorem | | Deprecated pnp3 AC0 endpoint | Vacuous enriched-package statement; not a standard AC0 lower bound | | Unconditional `P != NP` | **Not claimed** | ## Historical PNP3 public closure boundary The current public default endpoints (in `pnp3/Magnification/UnconditionalResearchGap.lean`) are: ```text NP_not_subset_PpolyDAG_final (gap : ResearchGapWitness) P_ne_NP_final (gap : ResearchGapWitness) ``` The only mathematical input is `ResearchGapWitness.dagSeparation : NP_not_subset_PpolyDAG`. Legacy `hMag` / `hMS` / support-bounds / multi-switching / provider entry points are retained only as explicit audit / compatibility wrappers; their source predicates are formally refuted by the falsifiability audit. ## What this release does not include 1. An unconditional in-repo theorem `P != NP`. 2. A non-vacuous proof of the formula-side support / locality source theorem. 3. A proof that `FormulaSupportBoundsPartial_fromPipeline_fixedParams` is realizable for realistic AC0 parameters. 4. A zero-argument final theorem with no external research payload. 5. A claim that deprecated `gapPartialMCSP_not_in_AC0` is a standard AC0 lower bound. It only negates an enriched package whose `params` and `easyData` fields already imply `False` without solver correctness. ## Audit-only theorem surfaces — not claims The final-result modules also contain quarantined audit / compatibility endpoints that compile but do not represent publishable claims. Per Research Constitution Rule 6, every such endpoint carries one of three explicit name prefixes: | Prefix | Meaning | Module | |-----------------|--------------------------------------------------------------|----------------------------------------------| | `RefutedRoute_` | Consumes a refuted predicate (e.g. `MagnificationAssumptions`, `FormulaSupportBoundsFromMultiSwitchingContract`, ...). Vacuous as a path to unconditional closure. | `FinalResultAuditRoutes.lean`, plus the `_supportBounds_TM` companions in `FinalResultLegacyTM.lean` | | `AuditOnly_` | Legacy `_TM` compatibility wrapper not consuming refuted predicates directly, but kept off the publishable surface. | `FinalResultLegacyTM.lean` | | `Vacuous_` | Vacuous via the typeclass-payload audit channel (`FinalPayloadProvider`, `DefaultFormulaSource`, ...). Records the channel concretely so it cannot quietly come back. | `FinalResultAuditRoutes.lean` | These prefixes are enforced visually: no `RefutedRoute_*` / `AuditOnly_*` / `Vacuous_*` endpoint should ever be presented as the public closure boundary. Treat them as the analogue of an `audit/` directory inside the type-theoretic API: their job is to keep historical shapes importable while making it impossible to mistake them for unconditional results. ## How to verify ```bash ./scripts/check.sh for f in pnp3/Tests/AxiomsAudit.lean \ pnp3/Tests/BarrierAudit.lean \ pnp3/Tests/BarrierBypassAudit.lean \ pnp3/Tests/BridgeLocalityRegression.lean \ pnp3/Tests/WeakRouteSurfaceTests.lean \ pnp3/Tests/FormulaSupportBoundsFalsifiabilityProbe.lean; do lake env lean "$f" done ``` ## Where to read next | Question | File | |-----------------------------------------|----------------------------------------------| | What is the project, in two pages? | `README.md` | | What is closed and what is open? | `STATUS.md` | | What blocks unconditional closure? | `CHECKLIST_UNCONDITIONAL_P_NE_NP.md` | | Route map for auditors | `PROOF_OVERVIEW.md` | | Common questions | `FAQ.md` | | Verified vs. not-claimed claim table | `TECHNICAL_CLAIMS.md` | | Axiom / `sorry` hygiene only | `AXIOMS_FINAL_LIST.md` | | Detailed frontier FAQ | `pnp3/Docs/Unconditionality_FAQ.md` | | Route policy lock | `pnp3/Docs/CLOSURE_ROUTE_POLICY.md` | | Release wording guardrail | `RELEASE_RC.md` |