erdos-problems 0.3.1 → 0.3.3

package/packs/number-theory/problems/848/WEAKEST_CASE_BUDGET.md

@@ -0,0 +1,183 @@
+# Problem 848 Weakest-Case Budget
+
+This note freezes the current best source-backed picture of the weakest branch
+in Sawhney's proof.
+
+Scope:
+- this is only the `0.0377` branch from the proof
+- this is not yet an explicit threshold proof
+- this is the budget sheet that the next explicit-threshold step must respect
+
+## Where the `0.0377` branch comes from
+
+Source:
+- Sawhney, `Problem_848.pdf`, Section 3, pages 2-3
+
+Branch description:
+- assume `A*` contains an even element
+- then bound `A*` using Lemma 2.1
+- bound `A7 ∪ A18` using Lemma 2.2
+- add the two contributions
+
+Public bound recorded in the note:
+- `|A| / N <= 0.0377`
+
+Target extremal density:
+- `1/25 = 0.04`
+
+Recorded slack to the target:
+- `0.04 - 0.0377 = 0.0023`
+
+## Main-term decomposition from the note
+
+### A. `A*` contribution
+
+Source:
+- Sawhney, `Problem_848.pdf`, pages 1-2, especially the proof lines leading to the first case split
+
+Displayed bound:
+- `|A*| / N <= (23/25) * (1 - prod_{p ≡ 1 mod 4, p >= 13} (1 - 2/p^2)) + o(1)`
+
+Rounded numerical contribution recorded in the note:
+- `0.0252`
+
+### B. `A7 ∪ A18` contribution
+
+Source:
+- Sawhney, `Problem_848.pdf`, pages 1-2, first Lemma 2.2 application in the proof
+
+Displayed bound:
+- `|A7 ∪ A18| / N <= (2/25) * (1 - prod_{p != 2,5} (1 - 1/p^2)) + o(1)`
+
+Rounded numerical contribution recorded in the note:
+- `0.0125`
+
+### C. Combined branch
+
+Displayed combination:
+- `|A| / N <= 0.0252 + 0.0125 = 0.0377`
+
+Important caution:
+- the note records four-decimal rounded numerics, not a repo-frozen exact Euler-product evaluation
+- therefore the currently usable slack is not yet a rigorously frozen `0.0023`
+- before spending the full `0.0023 N` on explicit error terms, the repo should freeze the exact main-term constants or adopt a deliberate safety reserve
+
+## Non-explicit inputs in this branch
+
+### 1. Lemma 2.1 large-prime tail
+
+Source:
+- Sawhney, `Problem_848.pdf`, page 1
+
+Public proof move:
+- choose `T = floor(sqrt(log N))`
+- control the tail by
+  `sum_{T <= p <= N^(1/2)} N / p^2 << N / T`
+
+Current status:
+- structurally explicit
+- not numerically frozen here
+
+### 2. Lemma 2.1 small-prime inclusion-exclusion remainder
+
+Source:
+- Sawhney, `Problem_848.pdf`, page 1
+
+Public proof move:
+- truncate to primes `p <= T`
+- note `prod_{p <= T} p^2 <= N^(o(1))`
+- conclude the inclusion-exclusion remainder is `N^(o(1))`
+
+Current status:
+- this is one of the main non-explicit steps in the branch
+- the proof architecture is clear, but the repo does not yet have a concrete bound replacing `N^(o(1))`
+
+### 3. Lemma 2.2 large-prime tail
+
+Source:
+- Sawhney, `Problem_848.pdf`, page 2
+
+Public proof move:
+- again choose `T = floor(sqrt(log N))`
+- bound the tail by
+  `sum_{T <= p <= N} (N / p^2 + 1) << N / T`
+
+Current status:
+- structurally explicit
+- not numerically frozen here
+
+### 4. Lemma 2.2 reduction to Lemma 2.1
+
+Source:
+- Sawhney, `Problem_848.pdf`, page 2
+
+Public proof move:
+- for each prime `p <= T` with `(p, qb) = 1`, the condition `p^2 | (ab + 1)` determines one residue class
+- the remaining count is then handled by Lemma 2.1
+
+Current status:
+- this means the `0.0377` branch inherits the non-explicit part of Lemma 2.1 twice:
+  once directly for `A*`, and once indirectly through Lemma 2.2 for `A7 ∪ A18`
+
+### 5. The `eta` absorption step
+
+Source:
+- Sawhney, `Problem_848.pdf`, pages 1-2
+
+Public proof move:
+- assume `|A| >= (1/25 - eta) N` for a small absolute `eta`
+- absorb `eta` and the `o(1)` errors into the strict inequality between the branch bound and `1/25`
+
+Current status:
+- no explicit admissible `eta` is frozen here
+- explicit threshold extraction must budget for both the analytic error terms and the `eta` slack used in the contradiction
+
+## Honest current bottleneck
+
+The next true bottleneck is not yet “make Lemma 2.2 explicit.”
+
+The first bottleneck is:
+- freeze the exact main-term constants behind `0.0252` and `0.0125`
+- decide how much of the nominal `0.0023` slack is actually safe to spend on explicit error terms
+
+Until that is done, any explicit-threshold extraction remains numerically underdetermined at the branch level.
+
+## What this changes
+
+- The route no longer needs a generic “find the weakest branch” task; that is now done.
+- The next honest move is to replace rounded branch numerics with an exact or conservatively certified usable slack budget.
+
+## Repo numerical freeze
+
+Date:
+- April 5, 2026
+
+Method:
+- multiply the Euler-product factors directly over all primes up to `5,000,000`
+- use a crude tail envelope from `sum_{n > P} 1 / n^2 < 1 / P`
+- treat this as a conservative numerical budget note, not yet as a fully formal proof artifact
+
+Numerical outputs:
+- `A*` main-term contribution: `0.0251587645...`
+- `A7 ∪ A18` main-term contribution: `0.0124525434...`
+- combined weakest-branch main term: `0.0376113079...`
+- target density: `0.04`
+- numerical slack to the target: `0.0023886920...`
+
+Crude omitted-tail envelope from the `5,000,000` cutoff:
+- total branch-level tail contribution below about `3.84e-7`
+
+Conservative numerical reading:
+- the branch-level main term appears safely below `0.037612`
+- the real branch slack appears closer to `0.002389` than to the rounded `0.0023`
+- the rounded note therefore seems to hide roughly `8.8e-5` of additional branch slack
+
+What this still does **not** do:
+- it does not make Lemma 2.1 explicit
+- it does not make Lemma 2.2 explicit
+- it does not yet absorb the `eta` term
+- it does not produce a final explicit threshold `N0`
+
+What it **does** do:
+- it removes uncertainty about whether four-decimal rounding is eating a significant fraction of the apparent branch budget
+- it gives the next explicit-extraction step a more honest target: roughly `0.002388` of branch slack before analytic error absorption

package/packs/number-theory/problems/848/context.yaml

@@ -0,0 +1,44 @@
+problem_id: "848"
+family_role: finite_check_number_theory_workspace
+harness_profile: decidable_gap_workspace
+default_active_route: finite_check_gap_closure
+bootstrap_focus: Treat Problem 848 as a finite-gap closure problem and freeze the certificate surface before claiming bounded verification progress.
+route_story: Problem 848 is already asymptotically resolved in public, but not yet closed here for all N. The honest job is to shrink the finite remainder with audited threshold tracking and reproducible bounded verification.
+frontier_label: finite_check_gap_closure
+frontier_detail: The repo has chosen bounded finite verification as the next closure lane under the best imported threshold currently tracked.
+checkpoint_focus: Preserve the distinction between existential N0, explicit N0, sample finite checks, and full closure.
+next_honest_move: Freeze the bounded finite-verification program and its certificate requirements before claiming any new covered interval.
+related_core_problems:
+  - "844"
+literature_focus:
+  - squarefree values
+  - sieve bounds
+  - explicit threshold extraction
+  - bounded verification
+artifact_focus:
+  - STATEMENT.md
+  - REFERENCES.md
+  - EVIDENCE.md
+  - FORMALIZATION.md
+  - THRESHOLD_LEDGER.md
+  - BOUNDED_VERIFICATION_PLAN.md
+  - VERIFICATION_REGIMES.md
+  - VERIFICATION_CERTIFICATE_SPEC.md
+  - EXTERNAL_VERIFICATION_LEDGER.md
+question_ledger:
+  open_questions:
+    - What exact interval should the repo try to cover first in a bounded-verification lane?
+    - Which imported threshold value is trusted enough to size the finite remainder operationally?
+    - What certificate format is strong enough to prevent repeat low-trust verification claims?
+    - Which public verification attempts are worth auditing rather than rebuilding from scratch?
+  active_route_notes:
+    - Keep the asymptotic theorem separate from the unresolved finite closure.
+    - Keep imported threshold claims separate from repo-owned audited claims.
+    - Prefer interval certificates to raw "checked up to X" claims.
+    - Use past public verification criticism as a design constraint, not just background chatter.
+  route_breakthroughs:
+    - Public asymptotic resolution by Sawhney is already in hand.
+    - The threshold ledger separating existential and explicit claims is now frozen.
+    - The repo now has an audited explicit candidate package distinct from the imported public threshold timeline.
+    - The repo has now chosen bounded finite verification as the next closure lane.
+  problem_solved: []

package/problems/848/AGENT_START.md

@@ -0,0 +1,33 @@
+# Agent Start
+
+Fast start:
+- `erdos problem show 848`
+- `erdos workspace show`
+- `erdos preflight`
+- `erdos continuation use route`
+- `erdos checkpoints sync`
+
+Working assumptions:
+- Open problem: no
+- Active route: finite_check_gap_closure
+- Repo status: active
+- Harness depth: dossier
+- Site status: decidable
+- Route breakthrough: yes
+- Problem solved: no
+
+Read in this order:
+- `STATEMENT.md`
+- `REFERENCES.md`
+- `EVIDENCE.md`
+- `FORMALIZATION.md`
+- `PUBLIC_STATUS_REVIEW.md`
+- `AGENT_WEBSEARCH_BRIEF.md`
+
+First honest move:
+- isolate the exact finite remainder left after Sawhney's sufficiently-large-`N` theorem
+- determine whether the missing closure is an explicit threshold extraction problem, a pure
+  finite computation problem, or a mixed lane
+- keep every note honest about the gap between `decidable` and fully `solved`
+- read the Sawhney note before touching the forum heuristics, then use the forum only to
+  map candidate threshold-improvement strategies and public finite-check coverage

package/problems/848/AGENT_WEBSEARCH_BRIEF.md

@@ -0,0 +1,21 @@
+# Erdős Problem #848 Agent Websearch Brief
+
+Why this exists:
+- do not rely on erdosproblems.com alone as the canonical public truth surface
+- compare the site status with current publicized discussion, literature, and formalization chatter
+
+Suggested queries:
+- "Erdos Problem #848"
+- erdos problem 848
+- "Erdos Problem #848" erdos
+- "Erdos Problem #848" explicit N0
+- "Erdos Problem #848" finite check
+- "Problem848.lean"
+- "Problem_848.pdf" Sawhney
+
+Review prompts:
+- Does the problem still appear publicly open?
+- Are there recent solution claims, partial claims, or major status updates?
+- Are there recent formalization artifacts, surveys, or project pages worth pulling into the dossier?
+- Has anyone publicly converted the existential threshold into a concrete usable `N0`?
+- Is there any public finite verification beyond the sample checks mentioned in the forum?

package/problems/848/CHECKPOINT_NOTES.md

@@ -0,0 +1,23 @@
+# Checkpoint Notes
+
+- Problem: 848
+- Repo status: active
+- Harness depth: dossier
+
+Checkpoint prompts:
+- What changed in the active route since the last honest checkpoint?
+- Which claim level is justified right now: Exact, Verified, Heuristic, or Conjecture?
+- Which upstream/public truth changed, if any?
+- What did the public-status review and agent websearch brief surface beyond erdosproblems.com?
+- Which artifact or literature bundle should the next agent read first?
+- What route, evidence, and formalization notes should be promoted out of scratch space into canonical dossier files?
+- Did we improve the explicit threshold side, the finite-computation side, or only our
+  understanding of the gap between them?
+- Are we optimizing the right thing right now: a smaller imported `N0`, or a smaller
+  remaining finite range to verify?
+- Is any new `N0` claim actually checked line-by-line against Sawhney's note, or is it still
+  forum-level heuristic discussion?
+- Have we cleanly separated the best imported threshold claim from the repo's own audited
+  explicit candidate?
+- What exactly does the current public Lean development certify, and what does it still leave
+  open?

package/problems/848/EVIDENCE.md

@@ -0,0 +1,112 @@
+# Evidence
+
+- This dossier was seeded for Erdos Problem #848 from a pull bundle.
+- Imported record included: yes
+- Site snapshot included: yes
+- Public status review included: yes
+- Repo status at seed time: active
+- Harness depth at seed time: dossier
+- Imported status: `decidable`
+- Imported formalization state: `yes`
+
+Current public evidence captured locally:
+- The live site snapshot from 2026-04-05 states that the problem is resolved up to a finite
+  check.
+- The site records van Doorn's density upper bound and Weisenberg's slight refinement as
+  context for why the extremal density is constrained near `1/25`.
+- The site reports that Sawhney solved the problem for all sufficiently large `N`, with a
+  stability statement forcing near-extremal sets into the `7 mod 25` or `18 mod 25`
+  classes.
+- The public search review surfaced a current Lean formalization thread and a suggested
+  arXiv reference for the sufficiently-large-`N` result.
+- Sawhney's public four-page note proves the existence of an `N0`, but does not give an
+  explicit threshold in the statement of Proposition 1.1.
+- The public forum discussion records an explicit-threshold timeline:
+  - `N0 = 7 x 10^17` on 2026-03-21
+  - `N0 = 3.3 x 10^17` on 2026-03-22
+  - `N0 = 2.64 x 10^17` on 2026-03-23
+- The same thread also records earlier larger threshold exploration, including
+  `exp(1958)`-scale discussion and `exp(1420)` as a plausible intermediate explicit route
+  rather than a final clean resolution.
+- The same thread suggests the real technical bottleneck is improving the error terms in
+  Lemma 2.1 and especially Lemma 2.2 enough to make the explicit threshold and the finite
+  range practically closable.
+- The public Lean thread claims a formalization of the sufficiently-large-`N` theorem and
+  mentions finite verification for `N = 50` and `N = 100`, but not a full finite closure.
+- The local number-theory pack now has a one-sided explicit Lemma 2.1 remainder ledger.
+  It shows that, at the public threshold candidates `exp(1420)` and `exp(1958)`, the
+  small-prime inclusion-exclusion remainder is negligible while the large-prime tail still
+  exceeds the frozen weakest-branch slack.
+- The local truncation scan also shows that the one-sided explicit route no longer needs
+  `T = floor(sqrt(log N))`; larger choices of `T` leave the discrete term negligible while
+  materially shrinking the live `A*` tail.
+- The local weakest-branch witness budget at `T = 250` now shows that, after the explicit
+  Lemma 2.1 and Lemma 2.2 tail terms are paid, the branch still appears to retain about
+  `0.001775` slack before the Lemma 2.2 prime-count term is inserted, and about `0.000366`
+  witness margin after that term is bounded with a Dusart-style explicit prime-count estimate.
+- The local weakest-branch assembly note now fixes a working witness
+  `T = 250`, `N >= exp(1420)`, `eta = 10^-4`, with about `0.000266` visible reserve left in
+  that branch after all currently frozen lemma-level costs.
+- The local branch-comparison ledger now indicates that the same witness also closes the
+  public branches with rounded bounds `0.0358`, `0.0336`, and `0.0294` at the repo's current
+  explicit level.
+- The local proposition-level candidate note now packages that shared witness as the current
+  repo candidate for an explicit stability statement, while still refusing to mark the problem
+  fully solved.
+- The local proof-obligation ledger now identifies the next real blocker as numerical
+  certification of the displayed decimal inputs, not another branch estimate.
+- The local certified numerical ledger now replaces the main displayed decimal inputs with
+  conservative machine-interval bounds, leaving a certified visible reserve of about
+  `0.000266` after the working `eta = 10^-4`.
+- The local theorem-style note now assembles the current explicit candidate into a single
+  proof-shaped artifact suitable for paper-writer mode and future public review.
+- The current candidate is now surfaced in both the paper bundle and a dossier-level public
+  review note, with a short share-ready summary.
+- The local paper bundle now has drafted introduction, preliminaries, and related-work
+  sections, and the bundle has been refreshed so that no placeholder paper text remains.
+- The current repo candidate is weaker than the best imported threshold value currently
+  visible on the public thread, so its main value is auditability and handoff structure,
+  not "best known `N0`" status.
+- The repo has now chosen bounded finite verification as the next closure lane and frozen
+  first-pass notes for verification regimes, certificate requirements, and imported
+  verification audit.
+
+Claim-safe local posture:
+- Exact: the public status is `decidable`, not `open` and not fully `solved`.
+- Exact: a formalized statement is publicly claimed by the imported atlas.
+- Exact: the remaining public gap is not a new asymptotic theorem, but the bridge from
+  existential `N0` to an explicit or fully checked all-`N` statement.
+- Exact: as of 2026-03-23, the best imported public-thread threshold visible in the dossier
+  is `N0 = 2.64 x 10^17`.
+- Exact: the repo's own audited candidate is currently weaker than that imported threshold,
+  so it should not be described as the best-known public bound.
+- Exact: the current repo explicitization of Lemma 2.1 isolates the large-prime tail as the
+  live analytic blocker in the weakest branch.
+- Exact: within the one-sided explicit route, enlarging `T` is now justified and appears to
+  be the right next analytic move.
+- Exact: a concrete witness value `T = 250` now survives the live tail bookkeeping in the
+  weakest branch at an explicit witness-budget level.
+- Exact: the weakest branch now has a concrete working `eta` witness, not just a floating
+  slack budget.
+- Exact: no other public branch currently appears to outrun that witness in the repo ledger.
+- Exact: the repo now has a proposition-level explicit candidate, but not yet a publication-
+  ready explicit theorem artifact.
+- Exact: the next real mathematical hardening step is numerical certification of the displayed
+  decimal inputs.
+- Exact: the current repo candidate is now numerically hardened at the ledger level.
+- Exact: the repo now also has a theorem-style proof artifact for the current candidate.
+- Exact: the current candidate package is now surfaced for both paper-writing and public
+  review workflows.
+- Exact: the paper bundle for the current candidate is now fully drafted at the section-shell
+  level and no longer depends on placeholder text in its public-facing sections.
+- Heuristic: the lowest-friction route from here is to work both sides of the bridge:
+  sharpen the explicit threshold while separately understanding how far direct finite
+  verification can reasonably go.
+
+Next maintainer step:
+- keep the dossier centered on the finite-check gap rather than re-opening the asymptotic
+  theorem lane
+- preserve the current claim-safe package as a clean review unit
+- treat imported threshold improvements as external progress markers unless they are
+  re-audited inside the repo
+- freeze the bounded finite-verification program before counting any interval as covered

package/problems/848/EXPLICIT_CANDIDATE_REVIEW.md

@@ -0,0 +1,57 @@
+# Problem 848 Explicit Candidate Review
+
+## Current repo candidate
+
+The current repo candidate is:
+
+- `N >= exp(1420)`
+- `T = 250`
+- `eta = 10^-4`
+
+Public context:
+- this is the repo's audited explicit candidate, not the best imported public threshold claim
+- the public forum thread records a stronger external threshold timeline through
+  2026-03-23, ending at `N0 = 2.64 x 10^17`
+
+and the claim-safe conclusion is:
+
+- if `A subseteq [N]` and `ab + 1` is never squarefree for all `a, b in A`
+- and `|A| >= (1/25 - 10^-4) * N`
+- then `A` is contained in either the `7 mod 25` class or the `18 mod 25` class
+
+## Why this is review-ready
+
+The repo now has explicit support notes for:
+- the weakest branch main term
+- explicit Lemma 2.1 and Lemma 2.2 bounds
+- the prime-count term
+- a weakest-branch assembly ledger
+- a branch-comparison ledger for the other public branches
+- a certified numerical ledger replacing bare display decimals
+
+Read in this order:
+- `packs/number-theory/problems/848/PROPOSITION_EXPLICIT_CANDIDATE.md`
+- `packs/number-theory/problems/848/THEOREM_STYLE_EXPLICIT_NOTE.md`
+- `packs/number-theory/problems/848/CERTIFIED_NUMERICAL_LEDGER.md`
+- `packs/number-theory/problems/848/WEAKEST_BRANCH_T250_ASSEMBLY.md`
+- `packs/number-theory/problems/848/BRANCH_COMPARISON_LEDGER.md`
+
+## What this is **not** claiming
+
+This review artifact does **not** claim:
+- that Problem 848 is fully solved in the repo
+- that the finite range below `exp(1420)` has been closed
+- that `exp(1420)` is optimal
+- that the current note set is already publication-ready
+
+The current status is:
+- explicit repo candidate: yes
+- publication-ready explicit theorem artifact: not yet
+- full all-`N` closure in this repo: not yet
+
+## Suggested reviewer questions
+
+- Is every displayed decimal used in the witness backed by the certified ledger?
+- Does the theorem-style note consume every branch cleanly without hidden extra losses?
+- Is the distinction between repo candidate and solved/public-truth claim clear enough?
+- Should the next surface be the paper bundle, a public review post, or both?

package/problems/848/FORMALIZATION.md

@@ -0,0 +1,16 @@
+# Formalization
+
+- Local status: statement-formalized
+- Imported formalized state: yes
+- Imported formalized last update: 2026-02-01
+- Public formalization chatter now includes a Lean development for the sufficiently-large-`N`
+  theorem discussed on the public problem thread.
+- Public forum claims indicate that the Lean work covers `∃ N0, ∀ N ≥ N0, Problem848Statement N`
+  and includes some sample finite checks (`N = 50`, `N = 100`), but not yet a publicly
+  stated full finite closure.
+
+Local formalization target:
+- keep the statement and the finite-check remainder cleanly separated
+- identify exactly which part of Sawhney's result is already formalized
+- decide whether this repo should mirror a public formal artifact, link it, or rebuild a
+  cleaner local version around the finite completion lane

package/problems/848/PUBLIC_STATUS_REVIEW.md

@@ -0,0 +1,16 @@
+# Erdős Problem #848 Public Status Review
+
+Fetched at: 2026-04-05T06:30:32.165Z
+Provider: duckduckgo
+
+Queries run:
+- "Erdos Problem #848"
+- erdos problem 848
+- "Erdos Problem #848" erdos
+
+Top public results:
+- [Erdős Problem #848 - erdosproblems.com](https://duckduckgo.com/l/?uddg=https%3A%2F%2Fwww.erdosproblems.com%2F848&rut=ab6658c984a06053ad3348d748646858fbf56e4f660020ca7ee7bfbfef6ff482)
+- [erdos-banger/formal/lean/Erdos/Problem848_FINAL.lean at main - GitHub](https://duckduckgo.com/l/?uddg=https%3A%2F%2Fgithub.com%2FThe%2DObstacle%2DIs%2DThe%2DWay%2Ferdos%2Dbanger%2Fblob%2Fmain%2Fformal%2Flean%2FErdos%2FProblem848_FINAL.lean&rut=6c864fa6a79183ffd94cce666451b7057530bd905d84b4213d80af6058f78dad)
+- [Problem #848: Is the maximum size of a set $A\subseteq \ {1,\ldots,N ...](https://duckduckgo.com/l/?uddg=https%3A%2F%2Fwww.dumbprime.com%2Ferdos%2Fproblem%2F848%2F&rut=cc6844492d2a1f0f0bc1cea1fa6dbd2c398f0734f7949a83f97a898ac20125fc)
+- [List of Erdős problems](https://duckduckgo.com/l/?uddg=https%3A%2F%2Fwww.math.ucsd.edu%2F~erdosproblems%2Ferdos%2F&rut=8e8ff012d3b3d9f37713038bf55c27bacfac0778281dbaa59eba31fb4ddce652)
+- [Erdos Problem Atlas — Interactive Exploration of 1,184 Classical Problems](https://duckduckgo.com/l/?uddg=https%3A%2F%2Ferdosatlas.org%2F&rut=9519b09e19680c6de399a4eb6980a1262a6b967ba8cca0f04fdab077275b7a8a)

package/problems/848/REFERENCES.md

@@ -0,0 +1,25 @@
+# References
+
+- Public problem page:
+  - <https://www.erdosproblems.com/848>
+- Original source pointer recorded on the public page:
+  - `[Er92b, p.239]`
+- Publicly linked sufficiently-large-`N` note:
+  - <https://www.math.columbia.edu/~msawhney/Problem_848.pdf>
+- Public discussion thread with explicit-threshold and finite-check remarks:
+  - <https://www.erdosproblems.com/forum/thread/848>
+  - useful recent timeline:
+    - 2026-03-21: `N0 = 7 x 10^17`
+    - 2026-03-22: `N0 = 3.3 x 10^17`
+    - 2026-03-23: `N0 = 2.64 x 10^17`
+- Publicly surfaced formalization discussion:
+  - <https://github.com/The-Obstacle-Is-The-Way/erdos-banger/blob/main/formal/lean/Erdos/Problem848.lean>
+- Public formalization milestone thread:
+  - <https://github.com/The-Obstacle-Is-The-Way/erdos-banger/blob/main/formal/lean/Erdos/Problem848_FINAL.lean>
+- Publicly suggested reference from the discussion thread:
+  - <https://arxiv.org/abs/2511.16072>
+- External imported atlas data:
+  - <https://github.com/teorth/erdosproblems>
+  - `data/problems.yaml`
+- Related problem:
+  - <https://www.erdosproblems.com/844>

package/problems/848/ROUTES.md

@@ -0,0 +1,78 @@
+# Routes
+
+## Status Ladder
+
+- Open problem: no
+- Active route: finite_check_gap_closure
+- Route breakthrough: yes
+- Problem solved: no
+
+## Active route
+
+- `finite_check_gap_closure`
+- Goal: convert the public sufficiently-large-`N` theorem into a complete all-`N`
+  resolution.
+- Current support:
+  - the public page reports an asymptotic solution by Sawhney
+  - the public page describes a stability statement near density `1/25`
+  - a public Lean formalization thread exists, but the exact finite closure surface still
+    needs review
+
+## Optimization target
+
+- Primary objective: decide the full truth value by closing the finite remainder.
+- Secondary objective: lower the explicit threshold `N0` only insofar as it shrinks the
+  finite remainder that still has to be checked.
+- Hygiene objective: keep imported public threshold improvements separate from the repo's own
+  audited candidate statements.
+
+## Subroutes
+
+- `external_threshold_tracking`
+  - record the best imported public `N0` timeline without automatically adopting it as
+    canonical repo truth
+  - current imported timeline:
+    - `7 x 10^17` on 2026-03-21
+    - `3.3 x 10^17` on 2026-03-22
+    - `2.64 x 10^17` on 2026-03-23
+- `explicit_threshold_extraction`
+  - extract or improve an explicit `N0` from Sawhney's proof
+  - likely pressure points:
+    - sharpen Lemma 2.1 with better explicit squarefree-counting bounds
+    - sharpen Lemma 2.2, which the public thread identifies as the harder obstacle
+    - test whether Montgomery-Vaughan style large-sieve inputs materially beat the current
+      coarse error budget
+- `bounded_finite_verification`
+  - once an explicit `N0` is available, close `N < N0` by direct computation
+  - public evidence suggests that naive finite verification has only reached relatively
+    small ranges so far
+- `formalization_coverage_audit`
+  - determine exactly what the public Lean files certify
+  - separate:
+    - asymptotic theorem coverage
+    - sample finite checks
+    - genuine full finite closure, if any
+
+## Route discipline
+
+- Do not widen `decidable` into `solved` without an explicit finite completion artifact.
+- Do not confuse "best imported public threshold" with "repo-owned audited threshold."
+- Count a route breakthrough only if we either:
+  - extract a fully explicit threshold that reduces the remainder to a bounded finite check, or
+  - finish the finite range directly.
+- Treat formalization progress as support evidence unless it clearly covers the exact
+  remaining finite gap.
+
+## Immediate next move
+
+- The repo now has a committed, claim-safe review package for its own audited explicit
+  candidate.
+- The imported public thread currently reports a better external threshold
+  `N0 = 2.64 x 10^17` on 2026-03-23.
+- The repo has now chosen the bounded finite-verification lane for the next cycle.
+- The next concrete task is to freeze:
+  - a regime split for the finite remainder
+  - a certificate format for bounded interval claims
+  - an audit ledger for imported verification work
+- The right optimization target remains the size of the remaining finite gap, not the
+  threshold race in isolation.

package/problems/848/SHARE_READY_SUMMARY.md

@@ -0,0 +1,36 @@
+# Problem 848 Share-Ready Summary
+
+Current repo candidate:
+- `N >= exp(1420)`
+- `T = 250`
+- `eta = 10^-4`
+
+Important context:
+- this is the repo's current audited candidate package
+- it is not the best imported public threshold currently visible on the forum thread
+- as of 2026-03-23, the public thread reports `N0 = 2.64 x 10^17` as a stronger external
+  threshold claim
+
+Claim-safe conclusion:
+- if `A subseteq [N]` and `ab + 1` is never squarefree for all `a, b in A`
+- and `|A| >= (1/25 - 10^-4) * N`
+- then the current repo candidate forces `A` into either the `7 mod 25` class or the
+  `18 mod 25` class
+
+What is ready now:
+- a theorem-style proof note in the paper bundle
+- a dossier-level explicit-candidate review note
+- supporting explicit ledgers for the branch bounds and numerical witness
+- a bounded finite-verification lane with regimes, certificate requirements, and external
+  audit notes
+
+What is not being claimed:
+- not full all-`N` closure in the repo
+- not a publication-ready proof artifact
+- not an update from `decidable` to `solved`
+- not the current best public `N0`
+
+Best pointers:
+- `problems/848/EXPLICIT_CANDIDATE_REVIEW.md`
+- `packs/number-theory/problems/848/PROPOSITION_EXPLICIT_CANDIDATE.md`
+- `packs/number-theory/problems/848/THEOREM_STYLE_EXPLICIT_NOTE.md`