O3 OPEN: every fixed A>4 smooth equality and negligible-budget route

Investment state: active. This describes research progress; claims have separate evidence grades.

Contribution to the goal

Register the exact O3 completion obligation from accepted source2597 as OPEN work. Full statement/domain: Original Eqs.23-24: fixed A>4, smooth equality and o(y/log y), not merely fixed A=12. Preserve the complete original statement reproduced in report_md and its locator original-manuscript-verbatim.md, lines 389–413. This is a proposed source/dependency/formalization path, not a novel theorem or a claim of completed source verification. No task for the full arbitrary-fixed-A smooth conclusion found. The parameter limits and fixed-A12 budget already proved must be reused.

Prior work and proposed difference

Updated route237rev10/7052 and route236rev12/6920, reused the source comparison, and consulted Hildebrand-Tenenbaum1993 Cor1.3 (https://tenenb.perso.math.cnrs.fr/PPP/Survey_friables.pdf, PDFp7/printedp417), pinned Mathlib SmoothNumbers.lean and official Lean/cache sources. Analytic smooth-number theory is prior art;2597 supplies the fixed-A12 final theorem and5728 the A-general budgets. This contribution is the missing exact-pinned final composition and fixed-prefactor adapter, not a new analytic estimate. Scope/queries and limitations are recorded in prior-work.json. The source2601 Eq24 equality requires a lower analytic half; neither an upper-only input nor a fixed finite budget proves it. No global absence claim.

Central uncertainty

The exact full statement is not established by the accepted three-claim package. No task for the full arbitrary-fixed-A smooth conclusion found. The parameter limits and fixed-A12 budget already proved must be reused. The first look must determine whether already available primary statements and compatible checked interfaces answer the missing step; source access or compatibility may be the blocker.

Next experiment

Can the supplied finite smooth-number budgets be upgraded to explicit little-o statements along the manuscript parameters, and can the full Eq24 two-sided asymptotic be bound to a separate, explicitly stated lower analytic input without mistaking an upper budget for an equality?

Reuse2597,5728 and this return's PrimaryMainA without rebuilding their final-completion argument. For A>240/23, use finite_Psi_mass_le_A and its free exponent margin to prove for every eta>0 an eventual bound2Psi(m+2,z1)<=etay/log(y). Under InputH9Upper(1/2), do the analogous all-A>4 argument using the third-power cost. Separately define the matching lower half of the uniform Hildebrand-Tenenbaum input and use the existing ulog(u)/L2 limit to prove an eventual sandwich X(log y)^(-A-eta)<=Psi(X,Z)<=X*(log y)^(-A+eta), with X=m+2 and Z=z1, for every eta>0. Preserve cutoff/floor conventions and source2601's exact Eq24 scope. State every lower-input assumption explicitly; do not attempt to prove or silently import O2's analytic input. Compile/audit the new declarations and targeted assumption/domain controls at the same pin.

Required evidence

Unaccepted premises remain conditional.

Evidence behind continued investment

These investigations led to the current experiment. Their claims retain their own evidence grades.

Investigation history

  1. Compiled SmoothBudgetA.lean & Kernel Replay (C05–C10): SmoothBudgetA.lean (0ad2ee9758587795dfceb9476eda18a1066b9adcc955c6f097ce2a48a361c64d, 16,650 B, zero sorry/axiom) proves rpow_pow_exp_eq, eventual_power_comparison, eventual_power_comparison' (via Mathlib's tendsto_rpow_atTop), decay_exponent_le_A, parameter_logZ_le_A, finite_Psi_mass_le_A, finite_smooth_budget_A (exhibiting 23 / 48 and repaired cost exponent 23 / 48 * A' - 5), eventual_smooth_budget_A_margin, eventual_smooth_budget_A, eventual_smooth_budget_A_unfolded, eventual_smooth_and_uncovered_budget_A, eventual_uncovered_budget_A (exhibiting 240 / 23), and the conditional printed Eq. 9 theorems eventual_smooth_budget_of_inputH and eventual_smooth_and_uncovered_budget_of_inputH for 4 &lt; A under SmoothBudgetA.InputH9Upper (1 / 2). Audit9088.lean (audit_9088.out 46e51906f0e127e864d518553e3b6078d23e7d311b8d7165bde7611eb3b6f45c, exit=0) confirms [propext, Classical.choice, Quot.sound] on all 19 declarations, and leanchecker -v SmoothBudget SmoothBudgetA (leanchecker_9088.out c9c6f25d08c4c6c57a65c4e21269a3488e5074dd99bc57607a3ce0313a7eeb81, exit=0) replays both .olean modules.
  2. Negative Controls (C11, C13): control_corrupt_9088.lean (8118673e60ef93c9b143790c75f65c323f3f21d290d749c474bb7acf8b08add2) mutates 240/23 -&gt; 10, 23/48 -&gt; 1/2, c-5 -&gt; c-4, and 4 &lt; A -&gt; 3 &lt; A; lean rejects all 4 (control_corrupt_9088.out 2afcfaa1b277b7f337170c33371efba985254eef0818173555d75e0c89043ab8, exit=1, surfacing sorryAx). check_9088.py passes 15/15 checks (check_9088.out 8167713ad072a5e87d469f433b2ac2b60b93d79aa644a154b333d32321d29730) and catches 15/15 corruptions (check_9088.corrupt.out 30eb9ddee9b79716e064e66914a990259f5efd06b273bd8565549f6a4c5093f2).
  1. NOT ANSWERED BY NEW CANDIDATE #5707 (C03-C05, C08). Return #5707 (Route 310, job #9129, recorded/recorded, created 2026-10-24T16:35:53.308Z, after #5703 at 16:14:30.549Z) packages Rh41.TakeCount (#4434) as a lean-kernel-v1 statement proposal: byte-exact sources (Rh41/GreedyResidual.lean 4d5dc128..., Rh41/TakeCount.lean 52fde740...), a fresh pinned-executor rebuild (leanprover/lean4:v4.35.0-rc3 + Mathlib 331d5244...), statement bundle, drafted verification plan, kernel object/custody records, axiom audit and controls; it claims no new mathematics. All 23 served attachments re-hash to their return digests (C04). Across the return JSON (its job_brief field excluded) and all 23 attachments, with the served step text stripped, there are 0 occurrences of decay_exponent_le_A, finite_Psi_mass_le_A, finite_smooth_budget_A, eventual_smooth_budget_A, eventual_smooth_budget_of_inputH, proposal_smooth_budget_A_ec, 23/48, 240/23, 503/46, 529/96, and also 0 of SmoothBudget, ShiftedRankin, SmoothRankin, SmoothLogLimit, SmoothParameters, eventual_smooth_budget (C05); #5707 never imports or compiles SmoothBudget.lean. Positive control (C07): the identical scanner detects all 10 target tokens in the reused #5703 corpus (counts 20-49 per token).

  2. REUSED INVARIANTS AND CURRENT STATUS (C01, C02, C06). Route custody at revision 7: state active, origin #2601, last_return_id: 5703, updated_at: 2026-10-24T16:14:30.549Z (exactly #5703 creation), 7 events; no newer return exists on Route #237 itself. #5703's audit is reused unchanged (0 O3 token occurrences outside job_brief across #5691, #5683, #5699 and the 8 comparison returns; exact rationals 1/2 - 1/48 = 23/48, crossover 4 - (23/48)A &lt; -1 &lt;=&gt; A &gt; 240/23, A=12 fixed margin 529/96 = 11/2 + 1/96 (503/46 if eps=1/2 kept), free-eps recovery for all A &gt; 240/23, fixed-(1-delta) Input H limit 5/(1-delta) &gt; 5 vs printed Eq. 9 closing all A &gt; 4; open findings #53828, #120070, #143613, #146586 do not touch SmoothBudget.lean). Fresh fetch of #5581 (scope: r237-a-general): still status: "pending", final_rung: null, research_route_id: null, no review ids, 16 attachment names unchanged from #5703's verified audit (C06); pending unreviewed work is not settled evidence.

WHY NOT known: the step demands a compilation experiment (restore the A=12 package of #2597 from its 63 served files, confirm the pre-state build and clean #print axioms, compile the four free-eps A-general declarations plus the printed Eq. 9 conditional). No return on record runs it; the only on-record artifact of that shape (#5581) is pending and unreviewed. The step stays OPEN and is copied byte-for-byte as next_step; held pursuit job #9088 is released with this note and #5707 never holds it again.

CHECKS: check_step11526.py 8/8 PASS (exit 0); --corrupt 6/6 single-field controls decisive (exit 1); 23/23 of #5707 and 5/5 of #5703 attachment digests re-verified; #5703's 138-attachment digest audit reused, not repeated. Read-only: cpu_hours 0, no Lean build, no experiment.

  1. NOT ANSWERED BY NEW CANDIDATE #5691 OR CONTEXT RETURNS #5683, #5699 (C08-C11).
  1. REUSED INVARIANTS AND SCOPING NOTES FOR PURSUIT JOB #9088 (C04-C07, C14).

CHECKS: check_step11509.py 15/15 PASS (exit 0); --corrupt 15/15 single-field controls decisive (exit 1); 138/138 attachment SHA-256 digests (+ 16/16 on pending #5581) verified.

  1. NOT ANSWERED BY THE 8 COMPARISON RETURNS (C08-C10). Across #5554 (route 239, O5 rescue, recorded), #5528 (route 239, O5 annotation audit, accepted/verified), #5295 (route 236, O2 Rankin U0 table, recorded), #4926 (route 235, O1 5-lemma tail pin, recorded), #4492 (route 240, O6 statement check, needs_capability), #4472 (route 241, O7 V(z)=2M(z)^2 P(z) identity, recorded), #4461 (route 239, O5 SieveUpperBound.lean compile, accepted/measured), and #4427 (route 238, O4a O4aDyadicReduction.lean compile, accepted/proven), plus all 52 attachments, there are 0 occurrences (outside quoted job_brief prompts) of decay_exponent_le_A, finite_Psi_mass_le_A, finite_smooth_budget_A, eventual_smooth_budget_A, eventual_smooth_budget_of_inputH, proposal_smooth_budget_A_ec, 23/48, 240/23, 503/46, or 529/96. Six returns (#5554, #5528, #5295, #4926, #4492, #4472) ran no Lean build. The two that ran Lean (#4427, #4461) compiled O4a (O4aDyadicReduction.lean 5ec2c501...) and O5 (SieveUpperBound.lean 0da7043b...); #4427 built #2597's unmodified A=12 SmoothBudget.lean (4427__build.out line 30) and audited only the four O4aDyadic.* declarations (4427__o4a_iso4.out). Positive control (C11): the identical scan on #4411 detects all 10 target tokens (eventual_smooth_budget_A: 27, 23/48: 68, 240/23: 91).

  2. PRE-STATE AND RATIONAL INVARIANTS RE-VERIFIED (C04-C06). Served SmoothBudget.lean (d69ec8f9..., 10,322 B) hard-wires chooseB 12 (10 occurrences), declares 9 fixed-A=12 theorems, and has 0 occurrences of eventual_smooth_budget_A. Exact rationals confirmed: 1/2 - 1/48 = 23/48, threshold 4 - (23/48)A &lt; -1 &lt;=&gt; A &gt; 240/23, (23/48)(23/2) = 529/96 = 11/2 + 1/96, fixed-margin threshold 240/23 + 1/2 = 503/46, and fixed-(1-delta) Input H threshold 5/(1-delta) &gt; 5 (10 at delta=1/2, 20/3 at delta=1/4, 50/9 at delta=1/10), so #5295's fixed-delta U0 tightening cannot replace printed Eq. 9 (Psi = X u^{-(1+o(1))u}) for 4 &lt; A &lt;= 240/23.

  3. SCOPING NOTES FOR THE PURSUIT (C07, C13). (a) #4427 and #4461 resolve #4411 section 5's Mathlib provenance note: Mathlib commit 331d5244f0d3aad530d9ab00ded135b4c7691502 on leanprover/lean4:v4.35.0-rc3 (commit 470d5ce1...) has lake-manifest.json 479029d8..., lean-toolchain bc84812c..., and lakefile.lean 7a0f6151... byte-identical to #2597's descriptors, and builds the 36-module closure via LEAN_PATH="$ROOT/rootbuild:$MATHLIB_LP". (b) research_context lists unreviewed pending #5581 (status: "pending", final_rung: null, research_route_id: null, scope r237-a-general 5448efe7..., files AGeneral.lean.txt, AGeneral__SmoothBudgetA.lean.txt, Audit.lean.txt, PreState.lean.txt). Pending unreviewed work is not settled evidence (known would stop Route #237 without review), so the pursuit is released with a note to inspect and verify #5581's candidate files at the pinned toolchain. (c) Open findings #53828, #143613, #146586 do not touch SmoothBudget.lean.

CHECKS: check_step11473.py 14/14 PASS (exit 0); --corrupt 14/14 single-field controls decisive (exit 1); 110/110 attachment SHA-256 digests verified.

  1. THE A=12 FILE RE-IMPOSES A>=12 IN TWO PLACES (DERIVED, exact rationals; anchored in served SmoothBudget.lean sha d69ec8f9...). (a) Hypothesis: decay_exponent_le consumes hut : (23/2)*L2 &lt;= u log u. That 23/2 is 12 - 1/2: the ratio limit at A=12 with eps=1/2, not a general constant. A general run can only supply A'*L2 &lt;= u log u with A' &lt; A; copying the same 1/2 margin forces A - 1/2 &gt;= 23/2, i.e. A>=12, precisely the step's failure clause. (b) Target: the file closes at -(11/2)T, which is (23/48)(23/2) = 529/96 = 5.510416... rounded down to 11/2 (gap exactly 1/96). The sharp constant is 23/48 = 1/2 - 1/48 (1/2 from the leading -(u logu)/2 of ShiftedRankin.actual_finite_Psi_bound, 1/48 from SmoothBudget.correction_le). Generalising the rounded target to (23/48)(A-1/2) moves the threshold to 240/23 + 1/2 = 503/46, losing (240/23, 503/46].

  2. THE EXACT REPAIR (DERIVED). Carry the margin free: 0 &lt; eps &lt; A - 240/23, A' := A - eps, target -(23/48)A'T, hypothesis A'T &lt;= u logu. Then (23/48)A' &gt; 5 &lt;=&gt; A' &gt; 240/23 &lt;=&gt; A &gt; 240/23 + eps, satisfiable for every A > 240/23: threshold exactly 240/23, no A>=12, failure clause does not fire. Budget test 4 - (23/48)A < -1 has its exact crossover at A = 240/23 (equality there gives -1, not <-1). The only new plumbing is the closing comparison t^(4-c) &lt;= 1/24 for c > 5; A=12 avoids it because 4 - 11/2 = -3/2 collapses to 1/(t sqrt t) (exp_decay_identity).

  3. PART (4): THE CONDITIONAL MUST USE Eq. 9 AS PRINTED (READ+DERIVED). Printed Input H is Psi = X u^{-(1+o(1))u}, so the effective exponent is (1+o(1)) u logu = (A+o(1))L2, i.e. A, and the manuscript's Eq.24 reads the ratio as L^{4-A+o(1)}, closing for every A>4. The step's fixed 1-delta paraphrase is strictly weaker: 4-(1-delta)A &lt; -1 &lt;=&gt; (1-delta)A &gt; 5 &lt;=&gt; A &gt; 5/(1-delta) &gt; 5, so no fixed delta>0 closes A<=5 and the conditional cannot claim 4<A<=240/23 under that reading. Scoping correction to the step's wording, not a change to the route.

  4. ONLY ONE FILE IS FIXED-A (READ, byte-anchored). The called modules are already A-parameterised: SmoothParameters.mass_bounds (y A B), log_mass_bounds, u_range, SmoothLogLimit.eventual_u_log_u (A) (hA:4&lt;=A) (eps) (heps:0&lt;eps), SmoothRankin.actual_finite_rankin (y A), ShiftedRankin.actual_finite_Psi_bound (y A), ReserveBudget.parameter_uncovered_budget (y A). SmoothBudget.lean uses chooseB 12 10 times; no served file declares eventual_smooth_budget_A. Formalisation is confined to SmoothBudget.lean (five declarations plus the new conditional). Patch text: proposal_smooth_budget_A_ec.md (explicitly uncompiled).

  5. NOT DONE, STATED: no compiled development. This host has Lean v4.34.1 and no Mathlib; the package pins v4.35.0-rc3; and neither a project lakefile nor a Mathlib revision pin is served (the served lake-manifest is Mathlib's own, listing its deps). A build at a revision I chose would not be the checked package. No kernel evidence is claimed.

CHECKS: 12/12 PASS exit 0 (check_ec.out); 12/12 single-field corruption controls decisive (check_ec.corrupt.out); 18/18 byte anchors present and their files re-hash to the pins.

NOT ANSWERED BY THE RECORD. The three new candidates (route 236, all recorded, all after #3577): #3651 (job 7480, 0.03 cpu-h: an explicit Rankin bound gives HalfRangeUpper PNT-free, so de Bruijn (1966) II's error terms are not load-bearing for the upper side), #3627 (job 7405: the primary upper paper is hash-pinned but image-only here; the Theorem-5 error is localized as the load-bearing input) and #3594 (job 5511: primary HT Cor 1.3 extraction plus an explicitly UNCOMPILED Lean draft, UniformPsiInterface.lean). Across the comparison set's served bytes, none of the three contains eventual_smooth_budget, eventual_smooth_budget_A, finite_smooth_budget, decay_exponent_le, ShiftedRankin or the constant 23/48; SmoothParameters occurs once each in #3651/#3594 as a citation of #2651's package, SmoothLogLimit once in #3594. The only answer-shaped token outside the served step text is one 240/23 in #3576, a row of its comparison table quoting the setter ("#2651 ... a reported finite-Rankin consequence for A>240/23, and a proposed general-A formalization"); its window names 2651: a citation, not a result.

THE SCAN STRIPS THE STEP'S OWN TEXT. The success criterion names the theorem TO BE PRODUCED (eventual_smooth_budget_A), so the setter's method/success strings contain it and a naive scan finds the step itself. After stripping the served step's string values the result is empty. Positive control, same scan: #2651 and #3577 carry eventual_smooth_budget/240/23 outside the step.

PACKAGE ON RECORD HAS ONLY A=12. SmoothBudget.lean as served by #2651 (fetched by sha, re-hashed; sha in the pins file) hard-wires SmoothParameters.threshold 12 (chooseB 12), u y 12, X y 12, Z y 12; its budget theorems are finite_Psi_mass_le, finite_smooth_budget and eventual_smooth_budget (2Psi(m y (chooseB 12)+2)(z1 y 12) <= y/(24log y)). eventual_smooth_budget_A occurs 0 times there.

COVERAGE / LIMITS. All returns named by route 237 (events, dependencies, basis) and all six named by route 236's event list (#2600, #2645, #3576, #3594, #3627, #3651) were fetched and scanned, plus the assignment's 22-item relevance list, both route records and the open finding #53828 (§10 text repair on paper/kk-lower-bound.md; it does not touch the budget). There is NO return-listing endpoint (GET <project>/returns 404), so this is not a global sweep: a return producing the general-A theorem without naming it and without being cited by either route would be invisible.

SCOPING FOR THE PURSUIT (from the record, no new work). (1) #3651 (recorded, not accepted) reports a PNT-free Rankin substitute for the HALF-RANGE UPPER bound -- the shape of the hypothesis the step's part (4) names for 4<A<=240/23 -- while #3627 (its predecessor) localized the opposite reading; decide which to adopt. (2) #3594's UniformPsiInterface.lean is uncompiled: re-verify, never cite. (3) Do not re-derive the A-general parameter map or the 23/48 / 240/23 arithmetic (#2651, re-checked by #3577). (4) The failure clause is live: if the argument needs A>=12 or the transfer cannot be uniform in A, record that the finite route does not cover A>240/23 and keep the fixed-A12 budget as the only proved instance.

READ-ONLY: cpu_hours 0, no Lean build, no experiment, no published computation reproduced. Every served byte used is pinned (32 files) and machine-checked: 17/17 checks PASS exit 0, 17/17 controls decisive.

Bindings (all hash-verified on fetch against the return-2597 files listing; files_report.json).

Parameter implications are complete for all A≥4 (checked, not new). With L=log y, L₂=log L, L₃=log L₂, B=chooseB(A)=1+6·survivorConstant·A⁴, X=m+2, Z=z₁, u=logX/logZ:

The one missing input is the smooth estimate for Ψ, and it splits by A. The package proves only a finite Rankin upper bound, and only at A=12 (SmoothBudget.eventual_smooth_budget, eventual_smooth_and_uncovered_budget hard-wire 12). Deriving its A-general form (Eq.14 here): Ψ ≤ C·X·logZ·exp(-½ u logu + 6log4·√u·logu); the shift is ≤ u logu/48 when √u≥288log4, so exponent ≤ -(23/48)u logu, and with u logu=(A+o(1))L₂ and X logZ≤2yL⁴, Ψ ≤ 2C·y·L^{4-(23/48)A+o(1)}. That is o(y/L) iff 4-(23/48)A &lt; -1, i.e. A>240/23≈10.4348. Asymptotically the best this route gives is exponent A/2, so A>10 at minimum. Checked anchor at A=12: -(23/48)(23/2) = -529/96 ≤ -11/2, giving 2Ψ≤y/(24L), consistent with the package. By contrast Input H (Eq.9) Ψ=(m+2)L^{-A+o(1)} with (m+2)/y∼L³L₃²/(B L₂⁴) yields ratio to y/L ∼ L^{4-A+o(1)}L₃²/(B L₂⁴) → 0 exactly for A>4.

Conclusion. For A>240/23 the finite upper-bound route suffices (bounded formalization, no new analytic input). For 4<A≤240/23 the finite route is too weak; the earliest missing input is the printed uniform smooth equality Input H (Eq.9), which the manuscript marks OPEN (O2). Eq.24's full equality remains OPEN for all A>4.

Controls / limits. check_gr.py 31/31 exit 0; --corrupt fails (3 FAIL). No Lean compiled, no proof executed, no new theorem claimed. O2's own dependency on O1/O4 is not determined here. A genuine O2 route id was not confirmed in this session, so none is cited. One mathematical object (Eqs.23-24) and one package revision (return 2597) inspected.