# v0.13 External Target Search Form Draft Status: **OPERATOR LOCK 2026-05-30.** At lock time no v0.13 runner had been written, no candidate inventory had been computed, no external file had been downloaded for this chapter, and no D5 velocity-fraction value had been scored. This document locks the target-search procedure, eligibility gates, feasibility probes, selection rule, verdict tree, and claim boundary for finding a clean external transfer target after v0.12 blocked on attrition. Reviewed for self-consistency, non-circularity, and integrity: - **The anti-target-shopping firewall is airtight** across four layers: the Integrity Caveat forbidden-list, the D5 probe not recording/using `velocity_fraction`, the signal-blind selection rule, and the Next Chapter Boundary re-lock (a target lock is permission to DRAFT a transfer, never to run or interpret one). - **Feasibility bars correctly encode the v0.12 lesson:** `attrition <= 0.10` AND `Wilson95 upper <= 0.20`, so a target whose CI still plausibly crosses the 0.20 transfer-block gate cannot be locked. Internally consistent (at n=100 a 0.10 point gives Wilson upper ~0.17 < 0.20; 30-row probes are correctly too wide to lock). The 96 h / 24 h-sharded runtime bars would have flagged supp-A (161 h) on speed alone. - **Independence tiers prevent overclaiming:** Tier 0 (supp-A/B / repo-derived) cannot be selected; only Tier 1 -> "near-external," Tier 2/3 -> "independent external." - Verdict tree complete (locked / near / none / access-blocked / operator-probe); claim boundary honestly scoped (no vf-signal claim, no v0.10b / K_facet revision). Two surgical adjustments at lock (both strengthen the firewall or the eventual transfer's validity; neither changes the chapter's substance): 1. Discovery-blindness clause added to the Integrity Caveat forbidden-list: candidate DISCOVERY, not just selection, must be signal-blind. 2. Stability-label commensurability added to the Next Chapter Boundary re-lock list: the external S/U label must denote the same linear (monodromy/Floquet) stability as supp-B, or the transfer is invalid. Implementation note (already mandated by the D5 Feasibility Probe section; flagged because the v0.12 attrition-probe did the opposite): the v0.13 feasibility probe reuses `per_row_pipeline` but MUST NOT emit `velocity_fraction` or `zone_index` to any receipt -- only success/blocked/sanity, runtime, period, mass key, failure reason. Operational note: the inventory/discovery stage is an operator/agent-curated literature and data-store search (it is not algorithmic); the pre-registration firewall binds that search to be signal-blind per adjustment 1, and binds selection to be feature-blind per the Target Selection Rule. Implementation receipt (post-lock, 2026-05-30): additive harness `scripts/v13_external_target_search.py` plus npm aliases `isotrophy:v13:inventory`, `isotrophy:v13:profile`, `isotrophy:v13:probe`, and `isotrophy:v13:select` were added. This is harness work only: no candidate inventory/profile/probe/select command has been run as a v0.13 receipt, no external target has been locked, and no D5 velocity-fraction value has been emitted. The probe path sanitizes `per_row_pipeline` output to omit `velocity_fraction`, `zone_index`, and stability labels from per-row feasibility receipts. Discovery receipt (post-lock, 2026-05-31): a signal-blind literature/data sweep populated `results/isotrophy/k-facet-v13-external-target-search/target_inventory.csv` and `target_discovery_notes.md`, with a tracked snapshot at `docs/isotrophy/kfacet/kfacet_v13_target_inventory_snapshot.md`. This is an inventory only: no profile, D5 probe, selection, target lock, or velocity-fraction computation was run. The sweep used schema/source queries for periodic three-body catalogs with initial conditions, periods, masses, and linear-stability metadata; it did not search for catalogs by known or suspected velocity-fraction/stability behavior. ## Frame The isotrophy reopening now has a four-step ledger: ```text v0.10a: marginal ordered trend registered in-sample v0.10b: mass-marginal held-out predictor failed v0.11: within-m3 conditional velocity-fraction rank passed on supp-B v0.12: same-paper supp-A transfer blocked by D5 attrition ``` v0.12 did not falsify the velocity-fraction hypothesis. It showed that the frozen v0.7 D5 measurement cannot support a supplementary-A transfer: ```text uniform attrition_fraction = 0.3433 Wilson 95% CI = [0.2919, 0.3987] locked block threshold = 0.20 verdict = external_transfer_blocked_by_attrition ``` v0.13 is therefore not a rescue of v0.12. It is a source-selection chapter: > Find and lock an external target catalog where the frozen v0.11 rule could be > tested without leakage and without a v0.12-style D5 attrition wall. If v0.13 succeeds, it selects a target for a later transfer chapter. It does not itself test the velocity-fraction signal. ## Integrity Caveat The danger in a target search is silent target shopping. v0.13 prevents that by separating three activities: ```text allowed in v0.13: source discovery schema/profile checks overlap/leakage checks D5 feasibility probes that report only runtime and attrition target selection by pre-feature source quality and feasibility forbidden in v0.13: inspecting velocity_fraction zone counts by stability computing AUC_cond, J_cond, chi-squared, J-T, or any S/U-vs-feature statistic selecting a target because it appears likely to pass the v0.11 rule searching for, prioritizing, or excluding candidate sources by any known or suspected velocity_fraction-vs-stability behavior -- discovery itself must be signal-blind (you may search for "3-body periodic catalog with linear-stability labels"; you may not search for catalogs known to exhibit the vf/stability link) changing v0.11 cutpoints {0.25, 0.50} relaxing the frozen D5 numerical controls restricting a target post-hoc to the rows that integrate cleanly ``` Any candidate touched by v0.13 must carry a manifest that records what was observed and what was deliberately not observed. ## Frozen Transfer Rule The eventual transfer rule remains v0.11: ```text zone_index: positional-dominant = 0 if velocity_fraction < 0.25 mixed = 1 if 0.25 <= velocity_fraction < 0.50 velocity-heavy = 2 if velocity_fraction >= 0.50 ``` `m3` or the target's exact mass-tuple key is used only as a conditioning stratum. No target is eligible for a v0.11-style transfer unless it supports within-stratum S/U comparisons. ## Candidate Intake Candidate sources may come from: ```text local repo files paper supplementary files public project pages repository releases Zenodo / Figshare / institutional data stores author-provided catalog mirrors ``` Every candidate gets a stable slug: ```text ____ ``` The intake record must include: ```text slug source URL or local path access date download hash, if downloaded license / citation note, if present source family and authors relationship to supp-B and supp-A orbit type and dimensionality body count potential / equations mass parameters initial-condition fields period field stability-label definition row count label counts, if labels are provided in source metadata known transformations needed for integration ``` Downloading or copying a candidate file is allowed only into a clearly named staging path under: ```text docs/isotrophy/external_targets/_staging/ ``` No staged target file is promoted into tracked docs until it passes the source profile gate and the operator approves retaining it. ## Hard Eligibility Gates A candidate is eligible for v0.13 feasibility work only if it passes all hard gates before any D5 computation: ```text three-body Newtonian or explicitly transformable to the same Newtonian substrate periodic-orbit initial conditions sufficient to integrate the orbit period or closure time provided mass tuple or mass parameter provided per row linear stability label or equivalent stable/unstable label provided by source at least 200 rows before overlap quarantine at least 5 candidate conditioning strata before D5 attrition at least 5 S and 5 U rows in at least 5 candidate strata before D5 attrition not supplementary-A not supplementary-B not generated from this repo's v0.3-v0.12 outputs source can be cited and re-fetched or preserved with hash ``` Report-only candidates may be logged if they fail one gate, but they cannot be selected for the next transfer chapter. ## Independence Tiers Independence is scored before feasibility probes: ```text Tier 3: independent source family, independent search or publication, independent catalog construction Tier 2: same broad research lineage or same public site, but different paper, different table, and not derived from supp-A/supp-B Tier 1: same paper/source family but different supplement or representation Tier 0: supp-A, supp-B, duplicate, or repo-derived artifact ``` Tier 0 cannot be selected. Tier 1 can be inventoried for completeness, but v0.13 must prefer any viable Tier 2 or Tier 3 target over a viable Tier 1 target. If only Tier 1 targets are viable, the verdict must say "near-external target locked," not "independent external target locked." ## Overlap Quarantine Every candidate is compared against prior isotrophy evidence sources: ```text docs/isotrophy/supplementary-B_piano-init-condit-3d.txt docs/isotrophy/supplementary-A_periodic-3d_mirror.txt results/isotrophy/k-facet-v03-freeze-supplementary-b-comparison/ results/isotrophy/k-facet-v11-m3-conditional-vf-rank/ results/isotrophy/k-facet-v12-external-frozen-transfer/ ``` Use the strongest identity key the schema supports: ```text preferred: same mass tuple and max_abs_delta(normalized IC fields, T) <= 1e-9 fallback: source-native orbit identifiers plus invariants, marked weaker ``` If the candidate uses the same mirrored ansatz, also test the ansatz reflection image: ```text (z0, vx, vy, vz) -> (-z0, vx, vy, -vz) at same mass tuple and T ``` Overlap policy: ```text overlap rows are excluded from feasibility row counts candidate is leakage-blocked if overlap_fraction > 0.05 candidate is report-only if identity matching is too weak to bound leakage ``` ## D5 Feasibility Probe The D5 measurement is frozen to v0.7/v0.12: ```text integrator: DOP853 rtol: 1e-12 atol: 1e-12 max_step_fraction: 0.02 symplecticity_gate: 1e-4 reciprocal_pair_gate: 1e-4 ``` The feasibility probe may run the D5 pipeline, but it must not expose or use the computed `velocity_fraction` values. It records only: ```text success / integration_blocked / sanity_failed runtime seconds period mass key failure reason ``` It must not tabulate success, failure, or runtime by stability label unless the candidate is already rejected; the selected-target decision must be based on measurement feasibility, not S/U-feature relationships. ### Probe Sampling For each candidate that passes hard eligibility: ```text seed: 20260523 probe rows: 100 uniform rows after overlap quarantine minimum probe: 30 rows if the candidate has fewer than 100 eligible rows stop condition: stage, do not run inline, if projected probe exceeds 10 minutes ``` If a 100-row probe would exceed the inline budget, v0.13 may run a capped 12-row rate probe and stage the full 100-row feasibility command for the operator. The 12-row probe cannot lock a target by itself. ### Feasibility Bars A candidate is D5-viable only if the 100-row uniform probe satisfies: ```text attrition_fraction <= 0.10 Wilson 95% upper bound <= 0.20 projected full transfer wall-clock <= 96 hours single-threaded projected full transfer wall-clock <= 24 hours under the proposed shard count ``` `attrition_fraction` counts both integration-blocked rows and sanity-failed rows. The Wilson upper-bound rule is the direct v0.12 lesson: do not lock a target whose uncertainty still plausibly crosses the 0.20 transfer-block gate. ## Target Selection Rule After all eligible candidates are profiled and feasibility-probed, select the highest-ranked viable target by this ordered rule: 1. Highest independence tier. 2. Lowest Wilson 95% upper bound on attrition. 3. Largest projected primary-domain row count. 4. Most stable source/citation and easiest reproducibility. 5. Shortest projected full transfer runtime. No S/U-vs-feature statistic may enter this ordering. If two candidates remain tied, choose the one with the larger number of conditioning strata. If still tied, choose the lexicographically first slug and record the tie. ## Verdict Tree ```text external_target_locked at least one Tier 2 or Tier 3 candidate passes hard gates, leakage gate, and D5 feasibility bars; the highest-ranked target is locked for a later transfer chapter near_external_target_locked no Tier 2 or Tier 3 candidate is viable, but a Tier 1 non-duplicate candidate passes all gates; it may support a near-external chapter only no_viable_external_target_found inventory completed and no candidate passes the hard gates plus D5 feasibility bars target_search_blocked_by_access candidate sources could not be accessed or preserved well enough to make a reproducible inventory target_search_operator_probe_needed a candidate passes source gates, but the required 100-row feasibility probe exceeds the inline budget and has not yet been run by the operator ``` ## Required Outputs Expected output directory: ```text results/isotrophy/k-facet-v13-external-target-search/ ``` Required files: ```text manifest.json target_inventory.csv source_profiles.json overlap_audits.csv feasibility_probes.csv rejected_targets.csv selection_readback.md operator_commands.md ``` `manifest.json` must include: ```text verdict startedAt completedAt gitCommit candidate_count eligible_candidate_count viable_candidate_count selected_slug, if any selected_independence_tier, if any selected_attrition_fraction, if any selected_wilson95_high, if any selected_projected_primary_rows, if any selected_projected_runtime_hours, if any ``` ## Command Shape After Implementation The implementation should be additive: ```text scripts/v13_external_target_search.py ``` Expected command shapes: ```powershell python scripts/v13_external_target_search.py inventory ` --out results/isotrophy/k-facet-v13-external-target-search python scripts/v13_external_target_search.py profile ` --inventory results/isotrophy/k-facet-v13-external-target-search/target_inventory.csv ` --out results/isotrophy/k-facet-v13-external-target-search python scripts/v13_external_target_search.py probe ` --inventory results/isotrophy/k-facet-v13-external-target-search/target_inventory.csv ` --probe-rows 100 ` --seed 20260523 ` --rtol 1e-12 ` --atol 1e-12 ` --out results/isotrophy/k-facet-v13-external-target-search python scripts/v13_external_target_search.py select ` --out results/isotrophy/k-facet-v13-external-target-search ``` The inventory/profile stages are cheap and may run inline if implemented. The probe stage must obey the repository's long-run rule: if expected wall-clock exceeds about 10 minutes, stage the exact PowerShell command for the operator and stop. ## Claim Boundary If v0.13 locks a target, the allowed claim is: > A pre-feature search identified a reproducible external target that is > independent enough, low-leakage enough, and numerically tractable enough to > support a later frozen v0.11 transfer test. Forbidden upgrades: ```text does not confirm the v0.11 velocity-fraction signal does not overturn v0.10b does not revise the v0.3h K_facet structural null does not make isotrophy theorem-facing does not score any S/U-vs-velocity-fraction relationship does not permit relaxing D5 precision in the later transfer chapter ``` If v0.13 finds no target, the allowed claim is: > The current external-transfer path remains unconfirmed because no reproducible, > low-leakage, D5-tractable target was found under the locked search rules. If v0.13 blocks on operator probe, the allowed claim is: > A candidate target remains possible, but target lock is pending the required > 100-row D5 feasibility probe. ## Next Chapter Boundary Only after `external_target_locked` or `near_external_target_locked` may a later chapter draft a frozen transfer statistic. That later chapter must re-lock: ```text selected target path and hash overlap exclusion list primary conditioning strata stability-label commensurability with supp-B's linear (monodromy/Floquet) stability convention -- the transfer is valid only if the external S/U label denotes the same kind of stability; a non-commensurable label blocks the transfer attrition policy permutation null effect floor claim boundary matching the target's independence tier ``` v0.13 target lock is permission to draft the transfer test, not permission to run or interpret it.