# The stretch decomposition: S_k = [q_k^2, q_{k+1}^2) formalized, the Stretch Postulate placed strictly above the strong Zone Postulate, the delta split into a bounded rho(2) deficit and a diverging width margin, occupancy certified to 9.0e15, and the square anchor read exactly — real QR structure, zero advantage *(2026-08-21. Staging note; nothing here is integrated into a live document. Producer, formally embedded: `research/stretch-01.js` (0.9 s; code-sha256 446167b9..., out-sha256 fb0be535...; the tail carries a forced re-embed stamp from adding two printed figures before first publication — 0 of 152 figures in the replaced block changed, stamped in the fingerprint). The frame is Chris's, from live discussion: between consecutive prime squares the active destroyer set is frozen, the stretch is a finite window of the periodic tile, and it is a finality window, so tile survivors there are true twins. The producer verifies his hand anchors before printing anything — 11 wakes at 121 with first prime 127, 13 wakes at 169 with first prime 173, and [169, 289) feels only the moire of {7, 11, 13} — ABORTING on any mismatch, then calibrates against natal-onset-01 (999 channel pairs, 203 survivors over 1..1e4), zonegap-01 (Z2(11) = 30, Z2(13) = 30, head and tail digit-exact) and the adopted TOS/A113274 custody (sha-checked, 4051 rows, 75 starred records, record 41 = 8040 at 65095731749). Calibration marked per claim: PROVEN, VERIFIED by exact computation, MEASURED, OPEN, REFUTED. Conventions: pairs named by opener a; a pair is in S_q iff q^2 <= a and a+2 < q'^2; offsets t = a - q^2.)* ## 0. The object, and whose vocabulary it is For consecutive primes q < q', the **stretch** is S_q = [q^2, q'^2). This is the GLOSSARY's *onset shell* of q read as a first-class interval object; "stretch" is Chris's term for it and is adopted here. The stretches partition [4, infinity), no twin pair straddles a boundary (a < q'^2 <= a+2 forces a+2 = q'^2, composite, or a+2 = q'^2 + 1, even — VERIFIED at every boundary in the sieve range), and from q = 7 on the stretch is an UNWRAPPED window of the tile T_q, since q'^2 < q# (121 < 210, 169 < 2310, 289 < 30030, then ln q# ~ q against 2 ln q'). ## 1. The decomposition, formalized and verified [PROVEN + VERIFIED] - **Freeze (S1).** For every n in S_q, the destructive influencer set {r prime : r^2 <= n} is exactly {r <= q} — the wheel-frame reading is Chris's frozen {7..q}. One line from the q^2 first-fresh-kill rule (`research/natal-onset-01.js` §0 frame); verified integer by integer at all 25 stretches to q = 97. This is zonegap-02's D1 onset desert made into the grid itself: a stretch's interior contains no onset tread by construction. - **Finality (S2).** Every destroyer of every position in S_q is active, so wheel-survivors are certified: openers a in S_q with lpf(a) > q and lpf(a+2) > q are EXACTLY the twin-prime openers there, as sets — the Zone Restriction Lemma (`zonegap-02-reduction.md` §1, Lemma A) restricted to S_q, which sits inside zone q. Verified set-equal at all 25 stretches. - **Zone identity (S3).** With r0 the least prime whose square exceeds p, (p, p'^2) = (p, r0^2) ⊔ S_{r0} ⊔ ... ⊔ S_p, exactly pi(p) - pi(sqrt p) full stretches plus a head fragment (the fragment is the tail of the stretch containing p itself). The zone's LAST stretch is S_p = [p^2, p'^2). Verified digit-exact at p = 7, 11, 13, 17, 23, 53, 97. Conversely one stretch serves many zones: S_q lies wholly inside zone p for every prime p in [q, q^2) — the overlap redundancy zonegap-01 §3 measured is exact and structural on this grid. ## 2. The Stretch Postulate, and its exact logical position > **Stretch Postulate (SP).** Every stretch S_q contains a twin pair. > **Weak form.** Infinitely many stretches do. - **SP implies the strong Zone Postulate [PROVEN, containment].** S_p is a subset of zone p, and the tile's first hole above 1 is p' (zonegap-02 §1), so a pair in S_p is an in-zone pair: SP at p forces zone p occupied. - **The converse containment FAILS [PROVEN].** No zone fits inside any stretch (p >= q^2 and p' <= q' force q^2 <= p <= q' < q^2 for q >= 3), so the strong Zone Postulate does not deliver SP by any interval argument — a zone can be occupied entirely below p^2 while its last stretch starves. Whether some non-containment derivation closes the gap is OPEN; the honest grade is **SP >= strong ZP, strictly positioned above it**. - **Weak SP <=> TPC [PROVEN, both directions, elementary].** Occupied stretches are disjoint and finite, so infinitely many occupied stretches give infinitely many twins; conversely every twin pair with opener >= 5 lies in exactly one stretch, and infinitely many twins cannot fit in finitely many finite stretches. The weak forms of SP and ZP coincide at TPC, exactly as ZONE-POSTULATE §2 has it for zones. - **The upper neighbour [PROVEN implication, conjecture-conditional premise].** S_q is a union of consecutive integer-square windows [n^2, (n+1)^2), n = q..q'-1. So the A091592-completeness conjecture (no twin-free window (n^2, (n+1)^2) for n > 122; OEIS, tested to 1e7, keyword hard — ZONE-POSTULATE §5a carries the prior art) plus the finite check for q <= 109, which SEC B2's direct occupancy sweep to q = 9973 subsumes, implies SP. The chain is A091592-complete ==> SP ==> strong ZP ==> weak ZP <=> TPC <=> weak SP. - **The honest grade.** SP is TPC-strength (it implies TPC) and per-level strictly more demanding than zone occupancy: SP(p) demands a pair in the zone's DEEPEST stretch — the tread-free tail where zonegap-01 measured the binding gap loading. It is adopted as structure, not as need; the programme's minimal requirement stays the weak form, per ZONE-POSTULATE §2's own discipline. ## 3. The delta, split honestly [MEASURED, guardrail built in] **REFUTATION FIRST, the guardrail.** The density delta is a BOUNDED constant, and it is ADVERSE. Route B (density advantage at the origin/zone scale) is CLOSED (`REFUTED.md`; the rho(2) reversal, ZONE-POSTULATE §6) and nothing below reopens it: no claim toward TPC is made from any density in this note. - **The bounded constant [MEASURED, four bands to 1e8].** The stretch's survivor density is the tile density d(q) = (1/10) prod_{7<=r<=q}(1-2/r); the measured twin density in the stretches, divided by d(q), reads 0.7980, 0.7978, 0.8021, 0.7946 across q-bands [101,313] to [3163,9973], against the limit rho(2) = e^{2gamma}/4 = 0.793055 — with measured/HL at 0.9813..1.0057, i.e. Hardy-Littlewood accounts for everything. The stretch sits at u = 2, the MINIMUM of the survival curve: certification is bought at the worst point of the only curve in the problem. The tile overcounts its own certified twins by the bounded factor 1/rho(2) = 1.261 and no more. - **The diverging margin [MEASURED, decades 10^1..10^8].** M = width/env, width = q'^2 - q^2 ~ 2q ln q, env = largest published record gap wholly below the stretch top (= the zone envelope Z2(q); zonegap-03's identity). Per decade, min M runs 0.80 (at q = 29) -> 1.45 -> 2.81 -> 13.96 -> 66.44 -> 351.26 -> 2175.84, mean M reaches 67968.9 at decade 10^7..10^8, and env/ln^3 q holds 3.67..4.65 from decade 10^2 on (the zonegap band constants, echoed). The mean margin diverges like q/ln^2 q — the attack's budget — and the MIN margin like q/ln^3 q, owned by twin-q stretches of minimal width 4q + 4. Per the square-window lesson (ZONE-POSTULATE §5a), none of this divergence is evidence at the gap scale; it is the budget, not the proof. - **Occupancy, certified nearly eight decades past the 1e8 sieve [VERIFIED from adopted data].** Direct: all 1229 stretches with q <= 9973 (tops to 1e8) are occupied; minimum 1 pair, at the degenerate pre-wheel S_2 = [4, 9). The finer-than-zones pass the task priced cost one bit sieve to 1e8 and under a second. Beyond the sieve, the straddle criterion: an empty S_q forces an in-table twin gap >= width - 2 starting below q^2, so width - 2 > runmax(q^2) certifies occupancy from the adopted TOS table alone. That inequality holds at 5,484,595 of 5,484,596 stretches with q'^2 < 2^53 ~ 9.007e15; the SOLE exception is q = 29 (width - 2 = 118 against record 8's 150 at F = 659), where the direct sieve holds 2 pairs. **Every stretch with q'^2 < 2^53 is occupied**, conditional only on the adopted TOS/A113274 custody (zonegap-01 §1). Zone occupancy to 1e11 (zonegap-01 §2) does not give this — the stretch grid is finer — which is why the dedicated pass existed. ## 4. The square anchor: real structure, exactly bookkeeping - **(i) The QR kill law [PROVEN; VERIFIED r = 7..31, all prime anchors to 2000].** Inside S_q, active prime r kills opener q^2 + t iff t ≡ -alpha or -alpha - 2 (mod r), with alpha = (q mod r)^2 a NONZERO quadratic residue. So only (r-1)/2 of the r generic forbidden-offset pairs can ever occur at a square anchor, and any offset class t with -t and -t-2 both nonresidues (or 0) is IMMUNE to r at EVERY square anchor: {0, 2} mod 7, {1, 3, 9} mod 11, up to 8 of 31 classes mod 31. A generic window has no immune classes; the anchor's kill pattern is genuinely non-generic. - **The guardrail is a theorem [PROVEN].** Summing the kill incidence over all r offset classes gives 2(r-1) exactly — ensemble mean 2/r, identical to a generic window. The QR structure REDISTRIBUTES kills over offsets; it removes none. There is no density advantage at the square anchor, and Route B stays closed from this side too. - **(ii) The measured signature [MEASURED, 439,644 openers, heights 1e5..1e8].** Zero twin openers land in forbidden classes (certification restated in offset coordinates). The aggregate offset histograms are non-flat exactly as the character null predicts: against the CONDITIONAL null (realized anchors and per-stretch totals fixed, only offset placement tested) chi2/df reads 0.52..1.63 across r = 7..23, worst single class 2.40%. The marginal null's excess (chi2/df 21.85 at r = 7) is anchor-sampling variance, not offset structure. **Verdict: the square anchor's moire is real, deterministic, character-determined — and it is EXACTLY the CRT null. Structure, not advantage.** - **The deep-end reconciliation [MEASURED + cited].** zonegap-01's u -> 1 loading (binding gap just under p'^2) is NOT square-anchor attraction: Z2 = env identically (zonegap-03 §1, D = 0), the deep end is an onset desert (zonegap-02 §4), and here the record process itself is square-blind — record starts sit uniformly inside their stretches (mean placement fraction 0.479 against null 0.5, deciles consistent over 75 records), 8 record intervals contain a prime square against a null expectation of 9.92 (all at n <= 20 where gaps are widest relative to square spacing), and no record contains two consecutive prime squares (none may — that would be an occupancy failure). The frontier p'^2 IS a prime square, so "the binding gap sits near the squares" is the envelope entering at the frontier: bookkeeping, exactly as zonegap-03 resolved. - **(iii) Adjacent published machinery, named in the owning convention (`research/SEARCH-CONVENTIONS.md` discipline; no absence claim is made here).** The forbidden condition "q lands on a root of x^2 ≡ -t (mod r)" belongs to the *roots of quadratic congruences to prime moduli* literature: Duke-Friedlander-Iwaniec, Ann. of Math. 141 (1995) 423-441 (MR1324141), equidistribution of those roots, with a living successor line (e.g. arXiv:2107.13301). Per-modulus pattern discrepancy is character-sum territory: Polya-Vinogradov and Burgess bounds, and the Burgess least-nonresidue bound r^{1/(4 sqrt e) + eps} governs the first killable offset class. Named as adjacent instruments only — nothing here imports a theorem, so no page-image obligation is triggered; the honest limit is that ALL of this machinery is per-modulus, while the wall is the JOINT placement over all r <= q, where per-character bounds compose into exactly the C^{pi(z)}-class loss the corpus already owns (`history/staging/import-l1l2.md`). "The quadratic point is equidistributed at gap scale" would be the Holt-Conjecture-2.1 failure shape ZONE-POSTULATE §7 warns on, and it is not argued here. ## 5. The attack statement, with its honest hardness > **Target (SP, kill-side form).** For every prime q, the frozen moire of > {7..q} cannot place a twin-slot gap covering the square-anchored window > [q^2, q'^2) — the tile window anchored at the quadratic CRT point > (q^2 mod r)_{r<=q} always retains a slot. - **Budget.** Width ~ 2q ln q against danger env ~ 3.7..4.7 ln^3 q (measured, §3): margin ~ q/ln^2 q in the mean, q/ln^3 q at the twin-q minimum, both diverging; occupancy is already certified to 9.0e15 from adopted data, so any counterexample lives beyond the reach of every table now published. - **Provably harder than it looks [the honest sentence].** The tile HAS gaps far wider than the stretch: G2(q#) reads exponent 1.50 on the trusted terms (`exponent-control.md` §5 via ZONE-POSTULATE §5) and carries the proven lower bound G2(x#) >> x ln x lnlnln x / lnln x (ZONE-POSTULATE §3) — both >> 2q ln q. **No whole-tile gap bound can ever prove SP.** The question is pure PLACEMENT: whether the tile's rare wide gaps can sit exactly at the quadratic point. That is the anchored-versus-worst-case split again — the wall's exact shape, third frame (zones, origin, now the square anchor) — and the density delta is even ADVERSE here (rho(2), §3), so no statistical argument helps. - **What would prove it.** (1) An anchored-family cap at the quadratic point: the unified per-prime caps of `attack-anchored-01.md` transplant from the origin comb (anchor vector 0) to the stretch window (anchor vector -q^2 mod r, known and QR-structured); the caps are history-blind and per-prime, which is the only proof form that has ever certified survivors in this corpus. (2) The stretch's own destruction ledger: the Phi* transplant (zonegap-02 §3.4, W -> q'^2) bounds every prime's fresh kills in the stretch exactly — but it bounds destruction, not placement, and zonegap-02 §5 already says so. (3) QR equidistribution of the anchor against the tile's big-gap set — measure-versus-placement, the conjecture-shaped step, named above and not available. - **Cheapest falsifiable next probe.** Extend SEC C3 to records 76-82 (starts above 2^53; BigInt isqrt, minutes): seal the placement-fraction band [0, 1) uniform, 7 fresh points, before computing. A clustering of the unswept records against their stretch boundaries would be the first measured square-anchor coupling anywhere; a uniform read closes the fourth decade of the square-blindness measurement. Second probe, same cost class: replicate the SEC C2 conditional-null test on stretches q in [10^4, 3.16*10^4] (heights 1e8..1e9) with the bands sealed first. ## 6. NOT REACHED - Records 76-82 (starts beyond 2^53) are outside SEC C3; the straddle criterion stops at q'^2 < 2^53 although maxF = 9.99e15 would carry it slightly further with BigInt widths. - The offset signature is measured at heights 1e5..1e8 only; no second decade, no sealed bands (that is §5's probe, deliberately left). - The anchored-cap transplant to the quadratic point is named, not attempted; no cap was computed at any square anchor. - Whether strong ZP implies SP by any non-containment route is untouched beyond the disproof of the interval argument. - The Phi* ledger was not instantiated on the stretch grid (it is proven for zones at zonegap-02 §3.4; the W -> q'^2 substitution is verbatim but unexecuted here). - Nothing anywhere about WHERE survivors sit inside a stretch beyond the offset-class marginals; the supply question is the postulate, unchanged. --- *Producer and custody: `research/stretch-01.js`, embedded (`node research/qc/embed.js --check` passes bit-honest; the tail carries a forced-re-embed stamp, disclosed above). Calibration inside the run, all abort-on-mismatch: TOS sha256 prefix 78767cad d001b1b0 and 4051-row parse with 75 starred records equal to A113274 1..75 at the anchors checked (records 1-6, 41, 75); Chris's hand anchors 127 and 173; natal-onset window totals 999/203; zonegap Z2(11) = 30 and Z2(13) = 30 with head/tail digit-exact; rho(2) recomputed from gamma. Cited, not recomputed: the Z2 = env identity (zonegap-03), the zonegap band constants and u-deciles (zonegap-01), G2's exponent 1.50 and lower bound (ZONE-POSTULATE §3, §5), the unified floors (attack-anchored-01) — per the standing compute rule.*