docs: recover scientific frontier and score production mechanisms

[LightChainr]

科学信息恢复＋#275 原始观测量与可识别预测路由

继续保持 Draft。本 PR 不合并 main，不关闭、锁定或批量改名 Issue，不把优先级变成许可系统，也不 rebase/force-push。

已完成的机制链

固定 Kreg=K2+K0 与完整 original-U moving-root/slope functional 的 N100/N400 macro-window pilot 已按合同停止：两个 simultaneous intervals 均含零、点估计异号，N400 生产投影超过 2M 上限。D17/D21 未评估；不 top-up、不换 window、不拟合自由指数。64 个旧 counter 的 pivotal preflight 也按合同停止。

共同物理扇区与 edge/contact 链已经前移：

• c958b30b 在同一 N25 exact population 中保存 joint 12-port identity；118/1,390 sectors、106/106 pooled rectangles 与 30/30 complete two-geometry rectangles exact nonzero。same-sector determinants 仅占 coarse determinant 0.2101274%，所以 coarse amplitude 以 cross-sector mixing 为主。
• 22f01e33 的 kernel-changing graph 有 6,846 classes、740,950 fibres，signed mass -4.9488399165e-6，判决 TWO_INDEPENDENT_DEFECT_GAIN_REJECTED。在固定 NN-thermal selected set 内 no-contact mask 恰为空；automatic two-independent-defect/six-arm 路线被排除。
• df4a64f6 保留 stage×contact，exact N25 matrix 为 [-2.88380e-6,-2.93729e-6; -5.32257e-6,+6.19482e-6]，det=-3.3498535471e-11。幸存对象是 contact fusion × topological completion rank two，而不是 scalar contact loading。

Held-out N65 transmission 与 full-T share 已完成

Commit f46c38c3 使用冻结的 20M N65 block、四 shards、100 batches、独立 P45 baseline 与 no top-up，得到：

• matrix [-1.57857e-7,-9.22766e-8; -3.09130e-7,+3.69680e-7]；
• 四个 simultaneous marginal signs 保持 frozen -- -+；
• Delta_cs=-8.6882161e-14，SE 1.4139777e-14，95% CI [-1.1459612e-13,-5.9168198e-14]；
• literal decision CONTACT_FUSION_COMPLETION_TRANSMITS。

这些 cells 是 pooled-root Schur T_t numerator allocations；尚未逐格除以 M_t 或乘 A_N，不能改名为完整 cellwise J/U 或机制份额。

Branch-only 95a695c7 保存完整 P45/new-MC covariance，并做明确标注的 adaptive shape check：single-amplitude N25→N65 模型 chi2=10.3371/3、p=.01591。它与 f46c38c3 使用相同 seed 与 byte-identical shards，是一次 production 的依赖重分析，不是第二票。

同一 branch 的 bab37f21 把 determinant 写成 noncommuting completion commutator，并给出 post-reveal N^-3/N^-29/8 fingerprint；这是机制候选，不是 field identification。Current branch head f9ba1ff6 用相同 N65 sufficient statistics 重建 complete response：

• J65=-.00162250989+/-.00018553008；
• J65/J25=.29396323+/-.03361398；
• carrier share 从 full T25 的 5.892% 降到 full T65 的 2.551%；
• pooled-root 相对 prescribed reference-p 的位移只改变约 .0155% 的 |J65|。

所以 typed carrier 确实传输，但不饱和 original response；carrier closure 不再是默认解释。所有 N65 结果属于同一 dependency group。

PR #509 仍是 OPEN Draft，live head ac5f5fe7；PR #544/#545 保持仓库团队既有的 closed-unmerged 生命周期。本 PR 不复活、合并或改写这些对象。

当前 P0：#275 已完成第一轮 production forward-identifiability score

仓库团队已把 #275 调为 OPEN P0，把 #537 保留为 OPEN P1 theory。本 PR 只观察并记录该优先级变化。

本分支已经完成 raw-observable/normalizer dictionary，并把现有 norm-4
K1/K2 production 的完整选定 12 x 12 aligned covariance 用于第一次
机制判别；没有生成新 Monte Carlo：

• fixed ordinary q2 semisimple kappa=.5：chi2=11.792466/4，p=.018963，排除这个具体参数化；
• Jordan affine log：chi2=6.432527/4，p=.169092，尚未排除；
• free kappa：最优 2.114984508，即 delta=-1.080647096，是相对增长模，不支持“更 irrelevant bulk singlet”的读法；
• physical decaying semisimple 0<kappa<1 没有内部最优，infimum 落在 kappa -> 1- 的 Jordan collision boundary；
• fixed semisimple/Jordan 总秩各为 8、联合秩 12、像交维数 4；由于 unrestricted semisimple closure 在 kappa -> 1 包含完整 Jordan image，字面结论是 PARTIALLY_IDENTIFIABLE、uniform_identifiability=false。Jordan compatibility 不是 Jordan operator identification。

因此下一交付不再寻找更多描述性坐标，而只补一个
restricted_trace_modulus_or_phase_transport：同一 B source、original
q/E、pooled-root physical normalizer 下的 rank-0/rank-2 restricted-trace
传输关系。优先理论推导 independent-singlet 与 Jordan-top 的 modulus/
character vectors，再评分已有 rho-child 或 P43/P57 assets；只有理论不能
给出该关系时，才考虑同一 B column 的 phase-calibrated second rotation。

并行 C3 rank audit 也已经给出精确边界：单一物理角度、任意 complex
amplitude 时 H4/H8 的 real design images 都是 R^2，所以即使现有
Etop_r1 显著非零也严格不可识别。只有先冻结跨角 gain law，第二旋转
才有信息；signed-real contract 推荐 delta=pi/24（7.5 度），而
pi/12（15 度）exact alias。任意独立 complex gain 下，任何第二角度仍
不可识别。

一个 unnormalized numerator character zero 不能自动穿过 twist-dependent normalization。声称 common vacuum/Ward differential operator 的候选，必须证明它在 rank0/rank2 restricted traces 上按同一规则作用；否则必须给出明确 defect/contact/normalization 项。Q4/Jordan 候选也必须给 restricted-sector prediction，而不是借用同-spin field label。

并行 P1 / P2

• [P1 theory] Full original-U thermal/contact asymptotics after the completed one-defect gate #537 保持 P1，目标是 nonlocal remainder 与 uniform near-critical/pooled-root transport theorem。
• exact N25 all-z ledger、N85 normalized-tensor comparison 与 fresh N130 F=C+R partition 都是 conditional support；只在 [P0 theory] Original Matching-One observable: homology-marked trace, normalization, and identifiable predictions #275 冻结原始 observable/normalizer 与相反预测后再启动，不进入当前 production queue。
• joint-map witness 与 annulus theorem/counterexample 保留为 P2 theory/support；只在 physical joint-incidence-free complement 被证明后，six-arm tail 才获得适用对象。

仓库与执行边界

README、STATUS、RESEARCH-MAP、ROADMAP、NEXT-TARGETS、DECISION-EXPERIMENTS、TEAM-COORDINATION 及 machine-readable ledger/registry 统一记录 lifecycle、observer/sector、dependency、full-T share 与新的 #275 prediction contract。branch_only/open_pr 不提升为 main_integrated。

本轮没有合并、Issue lifecycle 变更、历史重写或其他服务器操作。只做一次最终 YAML/引用/whitespace 一致性检查；冲突主工作树未触碰，PR #267 继续保持 OPEN Draft。

[LightChainr]

Minimal CI repair pushed in 69221b3: restored the three retired-experiment navigation signals that test_canonical_docs_consistency.py intentionally guards (local pivotal N130/N170 replicas, third primitive norm-2 generation, and scalar/free-exponent repeats). The focused 11-test canonical-doc suite passes locally. No science code or production branches were changed.

[LightChainr]

Lifecycle drift note after 7ee897e: Target 1 has now crossed a mechanism boundary on branch-only results.

• analysis/marked-birth-stream-20260829@634040d supplies the common-path pilot and exact source audits.
• @8bef10b proves on the production stream

  Cov(q,J_D)=J_S/2+(p-1/2-<q>)J_D
  

  (and the S/D partner), so the old gamma_D4=Cov(A_top,J_D4)/M' row is a rank-gate contact identity rather than an independent field matrix element.
• @b31581a upgrades this to an all-order finite algebra: every topology-only observer f(q) lies in {1,q,q^2}, and its connected coupling to any root-occupation-independent S/D gate mark closes exactly in gate moments. Tiny axis/Gaussian controls give 8,960 zero-residual checks.
• research/p275-atop-field-identity-pipeline-20260829@36bf91a independently shows the nine-geometry gamma_D4 response is scale zero; all frozen N^-13/8 field-identity bases fail. This falsifies the source map, not the already selected ordinary H4 channel.
• The paired-orientation mean J_D4 row remains live: 634040d gives an exploratory q2 magnitude ratio 0.33085 (effective exponent 1.596, close to 13/8), but a higher-stat three-generation holdout is now being preregistered.

Therefore the N325/N425 common stream remains high value, but qJ should be retained as an exact contact/control row, not the primary continuum-coupling decision. To create a genuine matrix element, the acquisition needs an observer outside the ambient-rank algebra, with same-stream products such as O_ext, O_ext^2, O_ext*J_D4, and O_ext*J_S4. Candidate sectors are a separated local landing/arm observer, bulk Betti/Euler data, or a charged seam/winding row. This is compatible with the Draft's one-block covariance policy and sharpens rather than cancels Target 1.

[LightChainr]

Frozen q2 local-D4 lifecycle result

The preregistered paired-orientation, two-radius mean local landing-D4 selector has completed on Huawei ARM64. N65 and N130 each used 1,000,000 paired replicas, 100 batches, eight threads, disjoint seed/counter domains, and byte-identical runner binaries.

• Frozen target: child/parent = -2^(-13/8) = -0.3242098887.
• Observed raw P4 shell ratio: +0.6742354081.
• After N65-only thermal/H4 training, heldout N130 score: chi2=72.8633/2, p=1.506e-16; rejected at alpha=0.01.
• Exact q-contact controls close below 1.56e-61; q-products were never used as the primary field matrix element.
• Scientific boundary: this rejects the frozen identification of (D_R4-D_R2)/log 2 with the q2 thermal-H4 covector. It does not reject the previously selected global H4 harmonic. More replicas cannot repair the observed sign and amplitude mismatch.

Artifacts, 40x40 delete-one covariance, provenance and hashes: commit eea85c4. The >100 MB sparse per-replica streams remain at the recorded Huawei paths with SHA256; all scoring sufficient statistics are committed. No new PR was opened.

[LightChainr]

Three-generation post-reveal is complete on branch analysis/marked-birth-stream-20260829 at 936cff2 (primary result 507571d). Three Huawei jobs used the same ARM binary SHA 46d8a269...1223, 5m replicas per size, 100 batches, frozen seed/counter range. Full 6x6 aligned-batch covariance and remote SHA provenance are in results/server-20260829/mean-jd4-q2-chain/.

Phase-first result:

• N65->N130 = 0.24609 + 0.47062 i; phase residual from the preregistered negative-real target = -2.0526 +/- 0.0664 rad (z=-30.9), full target chi2=374.0/2.
• held-out N130->N260 = 0.11602 - 0.36618 i; phase residual = +1.8776 +/- 0.2666 rad (z=7.04), chi2=18.60/2.
• N65->N260 is nearly positive-real, but this is cancellation of two rejected single-step phases and does not rescue the frozen q2 mechanism.

Decision: the 20k-pilot phase coincidence did not reproduce. This rejects bare primitive-line mean J_D4 as the proposed scalar q2 H4 carrier; it does not reject the established global H4 response or a new external/local observer coupling.

Post-reveal diagnostic (not a rescue): P4 mean J_S4 magnitudes grow by 1.277 and 1.289, close to thermal 2^(3/8), with alternating phase. If this source remains interesting, the next discriminant must freeze a two-step/matrix transfer and add an independent second coordinate; the q4 coincidence cannot be upgraded to a q2 pass.

The first Zy setup failure is retained and excluded in provenance. Only /workspace/mean-jd4-N65-5m/raw/N65_5m.* is the accepted N65 source; GiB sparse streams remain on their declared server paths.

[LightChainr]

Target 3 now has a strict semantic gate and a frozen missing-cell design at commit 420d2a9 on branch analysis/p267-target3-context-rectangle-20260829.

Result: PR277 U/r2-r6 and P253 A_plus/A_minus share neither source nor readout. The former are global threshold-rank Bernoulli derivatives under cover generation; the latter are fixed-p root-toggle conditional landing-H4 amplitudes under radius transfer. Normalization and batch contracts differ too. Therefore zero existing cross-context pairs are numerically eligible, and no rescaled or renamed pseudo-rectangle is reported.

The actual 2x2 is:

• rows: A_plus ordinary/matching-even and A_minus matching-odd;
• annulus cells: existing P253 N425 R2/R4/R7/R8;
• Gaussian-cover cells: missing, frozen as the exact same fixed-p R2 root-toggle readouts on the two PR277 cover lineages.

N425 is selected with the complete annulus covariance, but its minimum adjacent lambda Mahalanobis squared is only 0.04332, so this is a design freeze rather than a mechanism result. The held-out score compares three fixed shared-generator diagonals lambda={0,1/2,1} against nine fixed context-pair rows, with a composite-null parametric bootstrap and full dependency handling.

The reuse boundary is concrete. The P253 cyclic runner validates all primitive parents at R2; it rejects the nonprimitive Smith-(2,N/2) children. The only missing implementation is a general-period geometry adapter using the existing integer-period backend while preserving P253 event semantics and statistics verbatim. Production remains unauthorized until that adapter, tiny exact equivalence and synthetic scorer recovery pass.

Artifacts: analysis/p267_gaussian_annulus_missing_cells_20260829.yaml, machine result/report, note, checksums, script and 5 tests. No PR was opened or merged.

[LightChainr]

Additional lifecycle update for the navigation map: Issue #122 now has a real production full-shape analysis at 2b9b91b, beyond the synthetic contract on main.

The four finite-N comparisons reject exact standardized shape equality, but norm-2 and norm-5 cover changes from the same parent are nearly rank-one (cosines 0.999995 and 0.994739). The dominant residual is an upper-tail coordinate after median/IQR removal. This adds a reusable shape-flow observable to the Gaussian/context map; it does not supersede the three PR267 priority routes and should share the existing threshold-curve dependency groups. No PR/merge/history change.

[LightChainr]

Target 2 model-development pilot: named non-A_top complex C3 row

A 100k common-field pilot now measures the full primitive-homology automorphic harmonic delta H4=E[1_rank1 u(ell)^4]-H4_Pinson/Arguin(tau) on all three N112 degree-two rho children. This row resolves polarization inside rank one and is not A_top=P2-P0.

• Six-real null: chi2=13.8588/6, p=.0313 (weakly resolved).
• Clean Pell-transported E4/r1: 12.6712/4, p=.0130 (fails).
• E6/r0: 9.1517/4, p=.0574; E4-squared/r2: 9.7172/4, p=.0455.
• Frozen E4+E6 two-support subspace: 4.1844/2, p=.1234.
• Ideal C3 DFT: r0 is individually resolved (p=.0109); r1 and r2 are not.

Interpretation: the first typed complex row is not a clean weight-4/r1 character at this precision and instead nominates a small multi-character or finite-Pell mixture. This is model development, not renewed H4/H8/H12 voting. Full 6x6 common-field covariance, transformed C3 covariance, exact 65,536-mask oracle and provenance are in 8ed805f. An independent 2M/child production discriminator is being preregistered before acquisition.

[LightChainr]

Navigation drift update from current main: the exact quotient frontier has advanced beyond the PR267 text. Commit 4557a76 closes index 10: 86 HNF representatives, 70,690,518 complete filtration paths, 262,901,744 rank-one plateau steps, and zero failures in birth/reflection/rank-sum/reconstruction/line gates. The unrestricted degenerate-quotient theorem remains open.

Commit 83e98fc also adds an exact observer gate relevant to Targets 1/2: a direction-only readout on one C4 orbit has scalar/spin-4 character rank one. Local-H4 external observers must use at least axis+diagonal or retain typed/internal character data. The active Target1/Target2 designs have been told to treat this as an authorization gate.

[LightChainr]

P337/#537 route update from Draft PR #545

The bounded landing-minor falsifier listed here as the next action is now complete.

Decision: FINITE_PURE_THERMAL_RANK_ONE_REJECTED.

• An explicit family exists for every R>=1, L>=2R+5, with equal rest occupation, ordinary four-arm landing, transitions 0->1 / 1->2, and source-midpoint difference -2/L^4.
• The complete axis-L4 aggregate has
  det=-533831111/140737488355328 at p=1/2.
• Sturm certificates show the determinant and T_01+T_12 have no zero in the actual L4 matching-root interval [0.59,0.60].
• After the matching-root Schur choice beta=(A_01+A_12)/(T_01+T_12), the surviving even/H response is exactly

2*(T_01*A_12-T_12*A_01)/(T_01+T_12),

which is +0.0001366476693672144 at the root midpoint.

Validation in #545: 11 exact tests, independent BFS source reconstruction and potential-UF rank reconstruction on all 65,536 L4 states; no sampling or N25 extension.

Routing consequence: do not continue the pure-thermal rank-one cancellation route or assume that all ordinary three-packet landing terms vanish before proving a four-packet remainder. The next canonical object is the signed normalized aggregate residual

2*Psi_4/(T_01+T_12),
Psi_4=T_01*A_12-T_12*A_01,

with kernel, readout and pooled-root terms kept in the same covariance bundle. The explicit witnesses are not an asymptotic lower bound, so #537 remains open for its critical/near-critical rate.

[LightChainr]

2026-09-01 external-review integration update (commit 003391a):

• Consolidated seven supplied review packages into six distinct opinion sets, with snapshot chronology and adopt/correct/defer boundaries.
• Moved Issue [P0 theory] Original Matching-One observable: homology-marked trace, normalization, and identifiable predictions #275 from tooling language to existing-production model elimination. The aligned N112 2M/100-batch covariance excludes normalizer-only, rank-1-mass-only, and independent-real-rescaling parameterizations. Fixed q2 and the naive canonical weight-8 radial transplant are also excluded; Jordan remains compatible but not identified.
• Added the exact restricted source/thermal trace transport contract. The next discriminating objects are actual same-source columns in {s_r, partial_p s_r}, propagated through original q/E and the pooled moving-root normalizer.
• Corrected the [P1 theory] Full original-U thermal/contact asymptotics after the completed one-defect gate #537 cross-size exposure convention: N25 stored P while N65 stored P/N. The old comparison had an extra factor 65, so the specific 5/8 conditional-intensity attribution is withdrawn. The signed tensor, determinant, CONTACT_FUSION_COMPLETION_TRANSMITS decision, and full J65 remain unchanged. The determinant is retained as finite nonseparability with a possible commutator encoding; the unselected response is named complement_of_selected, not a proven nonlocal mechanism.
• README, STATUS, RESEARCH-MAP, ROADMAP, NEXT-TARGETS, and the machine-readable research ledger now reflect these boundaries and priorities.

Only the four focused analysis scripts, YAML/JSON parsing, and git diff --check were run. No new Monte Carlo, GPU/cloud/server work, issue lock/close, merge, rebase, or force-push. This PR remains Draft.

[LightChainr]

Closure update pushed at cb07694.

The overview and machine ledger now absorb the held-out N145 H0 gate, the exact Gaussian unit-rotation real-line theorem, and N145_FULL_T_UNRESOLVED. The previous primitive-H8 interpretation is explicitly superseded as an angle alias. For original-U mechanism identification, current assets are now classified UNIDENTIFIABLE_WITH_CURRENT_ASSETS because vacuum/Ward and thermal-Q4/Jordan do not yet define their same-source restricted-trace thermal columns. The only reopening input is two complete theory columns; no new angle, size, descriptor, certificate or free exponent substitutes for them.

PR remains Draft; no lifecycle changes or new production.

[LightChainr]

Large-scale decision portfolio after Draft PR #546

Commit b2341127 records the repository-level review without launching compute.

Delivered:

• notes/large-scale-decision-portfolio-20260901.md;
• analysis/large_scale_decision_portfolio_20260901.yaml (matching-one.large-scale-decision-portfolio.v1);
• README/NEXT-TARGETS and ledger v61 routing.

Decision: unacquired is not a priority criterion. The immediate P0 remains #275 candidate-specific restricted-trace columns followed by one existing-covariance score. A fresh block is eligible only if the profiled candidate images differ and a rank audit names one indispensable missing coordinate.

The highest new conditional large design is a stabilizer-free generic-modulus full homology-vector transport, which leaves the exact Gaussian unit alias. The most concrete structured control is #156/#159 recast as an actual-modulus Pell/Eisenstein vector Taylor jet, with a smaller power pilot and square-bond control before N418/N780. The N650/N850 Gaussian commuting square remains cost-stopped; transfer-matrix and rigorous-block routes remain behind measured reproduction or power gates; #227 is narrowed to a within-irrep curve-level pilot.

No server, GPU, Monte Carlo, top-up, Issue lifecycle change, merge, rebase or force-push was performed. PR #267 remains Draft.