Update CIL Atlas env_clip paper draft

workspace/latex/main.aux

@@ -67,7 +67,6 @@
 \@writefile{lot}{\contentsline {table}{\numberline {8}{\ignorespaces Action-bound audit for the leakage-audited RGB-reference chart DB. The bound is the ManiSkill \texttt  {pd\_ee\_delta\_pose} action space, $[-1,1]^7$. This is an action-representation validity audit, not a collision/contact safety audit.}}{9}{table.8}\protected@file@percent }
 \newlabel{tab:action-bound-audit}{{8}{9}{Action-bound audit for the leakage-audited RGB-reference chart DB. The bound is the ManiSkill \texttt {pd\_ee\_delta\_pose} action space, $[-1,1]^7$. This is an action-representation validity audit, not a collision/contact safety audit}{table.8}{}}
 \@writefile{toc}{\contentsline {section}{\numberline {8}Reproducibility Artifacts}{11}{section.8}\protected@file@percent }
-\@writefile{toc}{\contentsline {section}{\numberline {9}Limitations and Next Steps}{12}{section.9}\protected@file@percent }
 \bibstyle{plain}
 \bibdata{references}
 \bibcite{chen2025robotwin2}{1}
@@ -76,49 +75,56 @@
 \bibcite{kim2024openvla}{4}
 \bibcite{kwok2025robomonkey}{5}
 \bibcite{liu2023libero}{6}
+\@writefile{toc}{\contentsline {section}{\numberline {9}Limitations and Next Steps}{13}{section.9}\protected@file@percent }
+\@writefile{toc}{\contentsline {section}{\numberline {10}Conclusion}{13}{section.10}\protected@file@percent }
 \bibcite{singh2026bokbo}{7}
 \bibcite{tao2024maniskill3}{8}
 \bibcite{zhang2024vlabench}{9}
 \bibcite{zhao2026verispace}{10}
-\@writefile{toc}{\contentsline {section}{\numberline {10}Conclusion}{13}{section.10}\protected@file@percent }
-\@writefile{lot}{\contentsline {table}{\numberline {9}{\ignorespaces No-clipping validation refresh for \texttt  {base\_context\_obs}, K=8. This replays decoded raw actions from restored states and records action-bound validity labels before any clipping. The labels are action-space violations, not collision/contact outcomes.}}{14}{table.9}\protected@file@percent }
-\newlabel{tab:ctt-base-context-obs-val-noclip-rollout}{{9}{14}{No-clipping validation refresh for \texttt {base\_context\_obs}, K=8. This replays decoded raw actions from restored states and records action-bound validity labels before any clipping. The labels are action-space violations, not collision/contact outcomes}{table.9}{}}
-\@writefile{lot}{\contentsline {table}{\numberline {10}{\ignorespaces No-clipping held-out test refresh for \texttt  {base\_context\_obs}, K=8. The proposal oracle remains nonzero, but selected success does not improve over the raw-replay base and almost all known labels indicate action-space bound violations.}}{15}{table.10}\protected@file@percent }
-\newlabel{tab:ctt-base-context-obs-test-noclip-rollout}{{10}{15}{No-clipping held-out test refresh for \texttt {base\_context\_obs}, K=8. The proposal oracle remains nonzero, but selected success does not improve over the raw-replay base and almost all known labels indicate action-space bound violations}{table.10}{}}
-\@writefile{lot}{\contentsline {table}{\numberline {11}{\ignorespaces Scaled raw-action validation refresh for \texttt  {base\_context\_obs}, K=8, with \texttt  {--execution-action-scale 0.215} and clipping disabled. This is a diagnostic convention suggested by the action-bound audit, not a final action representation.}}{16}{table.11}\protected@file@percent }
-\newlabel{tab:ctt-base-context-obs-val-scaled-rollout}{{11}{16}{Scaled raw-action validation refresh for \texttt {base\_context\_obs}, K=8, with \texttt {--execution-action-scale 0.215} and clipping disabled. This is a diagnostic convention suggested by the action-bound audit, not a final action representation}{table.11}{}}
-\@writefile{lot}{\contentsline {table}{\numberline {12}{\ignorespaces Scaled raw-action held-out test refresh for \texttt  {base\_context\_obs}, K=8, with \texttt  {--execution-action-scale 0.215} and clipping disabled. The global scale improves base action-bound validity but does not preserve proposal support.}}{17}{table.12}\protected@file@percent }
-\newlabel{tab:ctt-base-context-obs-test-scaled-rollout}{{12}{17}{Scaled raw-action held-out test refresh for \texttt {base\_context\_obs}, K=8, with \texttt {--execution-action-scale 0.215} and clipping disabled. The global scale improves base action-bound validity but does not preserve proposal support}{table.12}{}}
-\@writefile{lot}{\contentsline {table}{\numberline {13}{\ignorespaces Per-dimension train-max scaled validation refresh for \texttt  {base\_context\_obs}, K=8, with clipping disabled. The scale vector is fit only from the train split action-bound audit.}}{18}{table.13}\protected@file@percent }
-\newlabel{tab:ctt-base-context-obs-val-perdim-rollout}{{13}{18}{Per-dimension train-max scaled validation refresh for \texttt {base\_context\_obs}, K=8, with clipping disabled. The scale vector is fit only from the train split action-bound audit}{table.13}{}}
-\@writefile{lot}{\contentsline {table}{\numberline {14}{\ignorespaces Per-dimension train-max scaled held-out test refresh for \texttt  {base\_context\_obs}, K=8, with clipping disabled. This diagnostic tests whether per-dimension max-fit scaling preserves more support than the global scale.}}{19}{table.14}\protected@file@percent }
-\newlabel{tab:ctt-base-context-obs-test-perdim-rollout}{{14}{19}{Per-dimension train-max scaled held-out test refresh for \texttt {base\_context\_obs}, K=8, with clipping disabled. This diagnostic tests whether per-dimension max-fit scaling preserves more support than the global scale}{table.14}{}}
-\@writefile{lot}{\contentsline {table}{\numberline {15}{\ignorespaces Bounded \texttt  {tanh} validation refresh for \texttt  {base\_context\_obs}, K=8, with clipping disabled. This action convention maps decoded controls into finite action bounds before validity checks.}}{20}{table.15}\protected@file@percent }
-\newlabel{tab:ctt-base-context-obs-val-tanh-rollout}{{15}{20}{Bounded \texttt {tanh} validation refresh for \texttt {base\_context\_obs}, K=8, with clipping disabled. This action convention maps decoded controls into finite action bounds before validity checks}{table.15}{}}
-\@writefile{lot}{\contentsline {table}{\numberline {16}{\ignorespaces Bounded \texttt  {tanh} held-out test refresh for \texttt  {base\_context\_obs}, K=8, with clipping disabled. The convention is action-bound-clean but selected success remains below the tanh base action.}}{21}{table.16}\protected@file@percent }
-\newlabel{tab:ctt-base-context-obs-test-tanh-rollout}{{16}{21}{Bounded \texttt {tanh} held-out test refresh for \texttt {base\_context\_obs}, K=8, with clipping disabled. The convention is action-bound-clean but selected success remains below the tanh base action}{table.16}{}}
-\@writefile{lot}{\contentsline {table}{\numberline {17}{\ignorespaces Measured residual \textsc  {CTT}{} rollout on 69 validation positive-support charts across three train seeds, K=8. Generated candidates are decoded and executed from restored simulator states; these are measured outcome metrics, not \ensuremath  {\mathrm  {PPTC}}{} proxies.}}{22}{table.17}\protected@file@percent }
-\newlabel{tab:ctt-val-rollout}{{17}{22}{Measured residual \ctt {} rollout on 69 validation positive-support charts across three train seeds, K=8. Generated candidates are decoded and executed from restored simulator states; these are measured outcome metrics, not \pptc {} proxies}{table.17}{}}
-\@writefile{lot}{\contentsline {table}{\numberline {18}{\ignorespaces Measured validation rollout for the proxy-positive \texttt  {base\_context\_obs} visual-stat chart token, across three train seeds, K=8. The RGB-stat chart token improves support-side validation metrics, but the selected action still fails to beat the base action.}}{23}{table.18}\protected@file@percent }
-\newlabel{tab:ctt-base-context-obs-val-rollout}{{18}{23}{Measured validation rollout for the proxy-positive \texttt {base\_context\_obs} visual-stat chart token, across three train seeds, K=8. The RGB-stat chart token improves support-side validation metrics, but the selected action still fails to beat the base action}{table.18}{}}
-\@writefile{lot}{\contentsline {table}{\numberline {19}{\ignorespaces Measured validation rollout for the deterministic RGB object-layout chart token, across three train seeds, K=8. The proxy gate passes by mean positive distance, but measured rollout does not improve over the RGB-stat validation row.}}{24}{table.19}\protected@file@percent }
-\newlabel{tab:ctt-base-context-obj-val-rollout}{{19}{24}{Measured validation rollout for the deterministic RGB object-layout chart token, across three train seeds, K=8. The proxy gate passes by mean positive distance, but measured rollout does not improve over the RGB-stat validation row}{table.19}{}}
-\@writefile{lot}{\contentsline {table}{\numberline {20}{\ignorespaces Measured residual \textsc  {CTT}{} rollout on 48 test positive-support charts across three train seeds, K=8. The generated proposal oracle exceeds 50\%, but the selected action fails because the current score/dominance rule chooses poor candidates.}}{25}{table.20}\protected@file@percent }
-\newlabel{tab:ctt-test-rollout}{{20}{25}{Measured residual \ctt {} rollout on 48 test positive-support charts across three train seeds, K=8. The generated proposal oracle exceeds 50\%, but the selected action fails because the current score/dominance rule chooses poor candidates}{table.20}{}}
-\@writefile{lot}{\contentsline {table}{\numberline {21}{\ignorespaces Measured test rollout for the \texttt  {base\_context\_obs} visual-stat chart token, across three train seeds, K=8. Support and score-only selection improve relative to the base-action chart token, but selected success remains below base.}}{26}{table.21}\protected@file@percent }
-\newlabel{tab:ctt-base-context-obs-test-rollout}{{21}{26}{Measured test rollout for the \texttt {base\_context\_obs} visual-stat chart token, across three train seeds, K=8. Support and score-only selection improve relative to the base-action chart token, but selected success remains below base}{table.21}{}}
-\@writefile{lot}{\contentsline {table}{\numberline {22}{\ignorespaces Validation-calibrated dominance fallback evaluated on the measured test rollout. The threshold and conformal residual quantile are fit on validation rows only; test outcomes are used only for reporting. The fallback reduces coverage but does not yet repair selector transfer.}}{26}{table.22}\protected@file@percent }
-\newlabel{tab:ctt-dominance}{{22}{26}{Validation-calibrated dominance fallback evaluated on the measured test rollout. The threshold and conformal residual quantile are fit on validation rows only; test outcomes are used only for reporting. The fallback reduces coverage but does not yet repair selector transfer}{table.22}{}}
-\@writefile{lot}{\contentsline {table}{\numberline {23}{\ignorespaces Learned dominance fallback trained on validation measured rows and evaluated on held-out test rows. Features are deployment-visible candidate features: utility-energy scores, score margins to base, rank, and tangent norms. This improves over the base test success but remains a partial selector diagnostic.}}{26}{table.23}\protected@file@percent }
-\newlabel{tab:ctt-learned-dominance}{{23}{26}{Learned dominance fallback trained on validation measured rows and evaluated on held-out test rows. Features are deployment-visible candidate features: utility-energy scores, score margins to base, rank, and tangent norms. This improves over the base test success but remains a partial selector diagnostic}{table.23}{}}
-\@writefile{lot}{\contentsline {table}{\numberline {24}{\ignorespaces Best clipped-convention validation-calibrated dominance diagnostic: learned context dominance over the measured \texttt  {base\_context\_obs} visual-stat rollout rows. The calibrator is fit on validation measured rows only and evaluated on held-out test rows.}}{26}{table.24}\protected@file@percent }
-\newlabel{tab:ctt-base-context-obs-learned-dominance}{{24}{26}{Best clipped-convention validation-calibrated dominance diagnostic: learned context dominance over the measured \texttt {base\_context\_obs} visual-stat rollout rows. The calibrator is fit on validation measured rows only and evaluated on held-out test rows}{table.24}{}}
-\@writefile{lot}{\contentsline {table}{\numberline {25}{\ignorespaces Bounded-tanh selector diagnostic over already measured candidates. The learned context+tangent selector is fit on validation tanh rows and evaluated once on held-out test tanh rows. It is action-bound-clean, but not a train-clean deployment selector.}}{27}{table.25}\protected@file@percent }
-\newlabel{tab:ctt-base-context-obs-tanh-learned-dominance}{{25}{27}{Bounded-tanh selector diagnostic over already measured candidates. The learned context+tangent selector is fit on validation tanh rows and evaluated once on held-out test tanh rows. It is action-bound-clean, but not a train-clean deployment selector}{table.25}{}}
-\@writefile{lot}{\contentsline {table}{\numberline {26}{\ignorespaces Train-calibrated bounded-tanh selector diagnostic. Calibration uses train-split tanh measured rollout rows with same-chart and same-state source retrieval excluded, then evaluates once on held-out test tanh rows.}}{27}{table.26}\protected@file@percent }
-\newlabel{tab:ctt-base-context-obs-tanh-learned-train-dominance}{{26}{27}{Train-calibrated bounded-tanh selector diagnostic. Calibration uses train-split tanh measured rollout rows with same-chart and same-state source retrieval excluded, then evaluates once on held-out test tanh rows}{table.26}{}}
-\@writefile{lot}{\contentsline {table}{\numberline {27}{\ignorespaces Train-calibrated learned dominance evaluated on the same held-out test rollout rows. Calibration uses only train-split measured generated rollouts with same-chart and same-state source retrieval excluded. This is a cleaner selector diagnostic than validation calibration, but it does not beat the validation-calibrated best row and still fails the deployment gate.}}{27}{table.27}\protected@file@percent }
-\newlabel{tab:ctt-base-context-obs-learned-train-dominance}{{27}{27}{Train-calibrated learned dominance evaluated on the same held-out test rollout rows. Calibration uses only train-split measured generated rollouts with same-chart and same-state source retrieval excluded. This is a cleaner selector diagnostic than validation calibration, but it does not beat the validation-calibrated best row and still fails the deployment gate}{table.27}{}}
-\@writefile{lot}{\contentsline {table}{\numberline {28}{\ignorespaces Nonlinear train-calibrated selector diagnostic. The model and threshold are selected only on held-out train-calibration rows, then evaluated once on the held-out test rollout rows. The best nonlinear row does not beat Table\nobreakspace  {}\ref  {tab:ctt-base-context-obs-learned-train-dominance}, so the current bottleneck is not just linear separability in the dominance selector.}}{27}{table.28}\protected@file@percent }
-\newlabel{tab:ctt-base-context-obs-nonlinear-train-dominance}{{28}{27}{Nonlinear train-calibrated selector diagnostic. The model and threshold are selected only on held-out train-calibration rows, then evaluated once on the held-out test rollout rows. The best nonlinear row does not beat Table~\ref {tab:ctt-base-context-obs-learned-train-dominance}, so the current bottleneck is not just linear separability in the dominance selector}{table.28}{}}
-\gdef \@abspage@last{27}
+\@writefile{lot}{\contentsline {table}{\numberline {9}{\ignorespaces No-clipping validation refresh for \texttt  {base\_context\_obs}, K=8. This replays decoded raw actions from restored states and records action-bound validity labels before any clipping. The labels are action-space violations, not collision/contact outcomes.}}{15}{table.9}\protected@file@percent }
+\newlabel{tab:ctt-base-context-obs-val-noclip-rollout}{{9}{15}{No-clipping validation refresh for \texttt {base\_context\_obs}, K=8. This replays decoded raw actions from restored states and records action-bound validity labels before any clipping. The labels are action-space violations, not collision/contact outcomes}{table.9}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {10}{\ignorespaces No-clipping held-out test refresh for \texttt  {base\_context\_obs}, K=8. The proposal oracle remains nonzero, but selected success does not improve over the raw-replay base and almost all known labels indicate action-space bound violations.}}{16}{table.10}\protected@file@percent }
+\newlabel{tab:ctt-base-context-obs-test-noclip-rollout}{{10}{16}{No-clipping held-out test refresh for \texttt {base\_context\_obs}, K=8. The proposal oracle remains nonzero, but selected success does not improve over the raw-replay base and almost all known labels indicate action-space bound violations}{table.10}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {11}{\ignorespaces Scaled raw-action validation refresh for \texttt  {base\_context\_obs}, K=8, with \texttt  {--execution-action-scale 0.215} and clipping disabled. This is a diagnostic convention suggested by the action-bound audit, not a final action representation.}}{17}{table.11}\protected@file@percent }
+\newlabel{tab:ctt-base-context-obs-val-scaled-rollout}{{11}{17}{Scaled raw-action validation refresh for \texttt {base\_context\_obs}, K=8, with \texttt {--execution-action-scale 0.215} and clipping disabled. This is a diagnostic convention suggested by the action-bound audit, not a final action representation}{table.11}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {12}{\ignorespaces Scaled raw-action held-out test refresh for \texttt  {base\_context\_obs}, K=8, with \texttt  {--execution-action-scale 0.215} and clipping disabled. The global scale improves base action-bound validity but does not preserve proposal support.}}{18}{table.12}\protected@file@percent }
+\newlabel{tab:ctt-base-context-obs-test-scaled-rollout}{{12}{18}{Scaled raw-action held-out test refresh for \texttt {base\_context\_obs}, K=8, with \texttt {--execution-action-scale 0.215} and clipping disabled. The global scale improves base action-bound validity but does not preserve proposal support}{table.12}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {13}{\ignorespaces Per-dimension train-max scaled validation refresh for \texttt  {base\_context\_obs}, K=8, with clipping disabled. The scale vector is fit only from the train split action-bound audit.}}{19}{table.13}\protected@file@percent }
+\newlabel{tab:ctt-base-context-obs-val-perdim-rollout}{{13}{19}{Per-dimension train-max scaled validation refresh for \texttt {base\_context\_obs}, K=8, with clipping disabled. The scale vector is fit only from the train split action-bound audit}{table.13}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {14}{\ignorespaces Per-dimension train-max scaled held-out test refresh for \texttt  {base\_context\_obs}, K=8, with clipping disabled. This diagnostic tests whether per-dimension max-fit scaling preserves more support than the global scale.}}{20}{table.14}\protected@file@percent }
+\newlabel{tab:ctt-base-context-obs-test-perdim-rollout}{{14}{20}{Per-dimension train-max scaled held-out test refresh for \texttt {base\_context\_obs}, K=8, with clipping disabled. This diagnostic tests whether per-dimension max-fit scaling preserves more support than the global scale}{table.14}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {15}{\ignorespaces Explicit \texttt  {env\_clip} validation refresh for \texttt  {base\_context\_obs}, K=8, with clipping disabled. The declared decoder convention clips decoded controls to action-space bounds before validity checks, so action-bound labels measure the declared convention rather than silent simulator clipping.}}{21}{table.15}\protected@file@percent }
+\newlabel{tab:ctt-base-context-obs-val-envclip-rollout}{{15}{21}{Explicit \texttt {env\_clip} validation refresh for \texttt {base\_context\_obs}, K=8, with clipping disabled. The declared decoder convention clips decoded controls to action-space bounds before validity checks, so action-bound labels measure the declared convention rather than silent simulator clipping}{table.15}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {16}{\ignorespaces Explicit \texttt  {env\_clip} held-out test refresh for \texttt  {base\_context\_obs}, K=8, with clipping disabled. This convention is action-bound-clean and preserves more proposal support than bounded \texttt  {tanh}, but score-only selection remains below base.}}{22}{table.16}\protected@file@percent }
+\newlabel{tab:ctt-base-context-obs-test-envclip-rollout}{{16}{22}{Explicit \texttt {env\_clip} held-out test refresh for \texttt {base\_context\_obs}, K=8, with clipping disabled. This convention is action-bound-clean and preserves more proposal support than bounded \texttt {tanh}, but score-only selection remains below base}{table.16}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {17}{\ignorespaces Bounded \texttt  {tanh} validation refresh for \texttt  {base\_context\_obs}, K=8, with clipping disabled. This action convention maps decoded controls into finite action bounds before validity checks.}}{23}{table.17}\protected@file@percent }
+\newlabel{tab:ctt-base-context-obs-val-tanh-rollout}{{17}{23}{Bounded \texttt {tanh} validation refresh for \texttt {base\_context\_obs}, K=8, with clipping disabled. This action convention maps decoded controls into finite action bounds before validity checks}{table.17}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {18}{\ignorespaces Bounded \texttt  {tanh} held-out test refresh for \texttt  {base\_context\_obs}, K=8, with clipping disabled. The convention is action-bound-clean but selected success remains below the tanh base action.}}{24}{table.18}\protected@file@percent }
+\newlabel{tab:ctt-base-context-obs-test-tanh-rollout}{{18}{24}{Bounded \texttt {tanh} held-out test refresh for \texttt {base\_context\_obs}, K=8, with clipping disabled. The convention is action-bound-clean but selected success remains below the tanh base action}{table.18}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {19}{\ignorespaces Measured residual \textsc  {CTT}{} rollout on 69 validation positive-support charts across three train seeds, K=8. Generated candidates are decoded and executed from restored simulator states; these are measured outcome metrics, not \ensuremath  {\mathrm  {PPTC}}{} proxies.}}{25}{table.19}\protected@file@percent }
+\newlabel{tab:ctt-val-rollout}{{19}{25}{Measured residual \ctt {} rollout on 69 validation positive-support charts across three train seeds, K=8. Generated candidates are decoded and executed from restored simulator states; these are measured outcome metrics, not \pptc {} proxies}{table.19}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {20}{\ignorespaces Measured validation rollout for the proxy-positive \texttt  {base\_context\_obs} visual-stat chart token, across three train seeds, K=8. The RGB-stat chart token improves support-side validation metrics, but the selected action still fails to beat the base action.}}{26}{table.20}\protected@file@percent }
+\newlabel{tab:ctt-base-context-obs-val-rollout}{{20}{26}{Measured validation rollout for the proxy-positive \texttt {base\_context\_obs} visual-stat chart token, across three train seeds, K=8. The RGB-stat chart token improves support-side validation metrics, but the selected action still fails to beat the base action}{table.20}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {21}{\ignorespaces Measured validation rollout for the deterministic RGB object-layout chart token, across three train seeds, K=8. The proxy gate passes by mean positive distance, but measured rollout does not improve over the RGB-stat validation row.}}{27}{table.21}\protected@file@percent }
+\newlabel{tab:ctt-base-context-obj-val-rollout}{{21}{27}{Measured validation rollout for the deterministic RGB object-layout chart token, across three train seeds, K=8. The proxy gate passes by mean positive distance, but measured rollout does not improve over the RGB-stat validation row}{table.21}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {22}{\ignorespaces Measured residual \textsc  {CTT}{} rollout on 48 test positive-support charts across three train seeds, K=8. The generated proposal oracle exceeds 50\%, but the selected action fails because the current score/dominance rule chooses poor candidates.}}{28}{table.22}\protected@file@percent }
+\newlabel{tab:ctt-test-rollout}{{22}{28}{Measured residual \ctt {} rollout on 48 test positive-support charts across three train seeds, K=8. The generated proposal oracle exceeds 50\%, but the selected action fails because the current score/dominance rule chooses poor candidates}{table.22}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {23}{\ignorespaces Measured test rollout for the \texttt  {base\_context\_obs} visual-stat chart token, across three train seeds, K=8. Support and score-only selection improve relative to the base-action chart token, but selected success remains below base.}}{29}{table.23}\protected@file@percent }
+\newlabel{tab:ctt-base-context-obs-test-rollout}{{23}{29}{Measured test rollout for the \texttt {base\_context\_obs} visual-stat chart token, across three train seeds, K=8. Support and score-only selection improve relative to the base-action chart token, but selected success remains below base}{table.23}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {24}{\ignorespaces Validation-calibrated dominance fallback evaluated on the measured test rollout. The threshold and conformal residual quantile are fit on validation rows only; test outcomes are used only for reporting. The fallback reduces coverage but does not yet repair selector transfer.}}{29}{table.24}\protected@file@percent }
+\newlabel{tab:ctt-dominance}{{24}{29}{Validation-calibrated dominance fallback evaluated on the measured test rollout. The threshold and conformal residual quantile are fit on validation rows only; test outcomes are used only for reporting. The fallback reduces coverage but does not yet repair selector transfer}{table.24}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {25}{\ignorespaces Learned dominance fallback trained on validation measured rows and evaluated on held-out test rows. Features are deployment-visible candidate features: utility-energy scores, score margins to base, rank, and tangent norms. This improves over the base test success but remains a partial selector diagnostic.}}{29}{table.25}\protected@file@percent }
+\newlabel{tab:ctt-learned-dominance}{{25}{29}{Learned dominance fallback trained on validation measured rows and evaluated on held-out test rows. Features are deployment-visible candidate features: utility-energy scores, score margins to base, rank, and tangent norms. This improves over the base test success but remains a partial selector diagnostic}{table.25}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {26}{\ignorespaces Best clipped-convention validation-calibrated dominance diagnostic: learned context dominance over the measured \texttt  {base\_context\_obs} visual-stat rollout rows. The calibrator is fit on validation measured rows only and evaluated on held-out test rows.}}{29}{table.26}\protected@file@percent }
+\newlabel{tab:ctt-base-context-obs-learned-dominance}{{26}{29}{Best clipped-convention validation-calibrated dominance diagnostic: learned context dominance over the measured \texttt {base\_context\_obs} visual-stat rollout rows. The calibrator is fit on validation measured rows only and evaluated on held-out test rows}{table.26}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {27}{\ignorespaces Bounded-tanh selector diagnostic over already measured candidates. The learned context+tangent selector is fit on validation tanh rows and evaluated once on held-out test tanh rows. It is action-bound-clean, but not a train-clean deployment selector.}}{30}{table.27}\protected@file@percent }
+\newlabel{tab:ctt-base-context-obs-tanh-learned-dominance}{{27}{30}{Bounded-tanh selector diagnostic over already measured candidates. The learned context+tangent selector is fit on validation tanh rows and evaluated once on held-out test tanh rows. It is action-bound-clean, but not a train-clean deployment selector}{table.27}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {28}{\ignorespaces Train-calibrated bounded-tanh selector diagnostic. Calibration uses train-split tanh measured rollout rows with same-chart and same-state source retrieval excluded, then evaluates once on held-out test tanh rows.}}{30}{table.28}\protected@file@percent }
+\newlabel{tab:ctt-base-context-obs-tanh-learned-train-dominance}{{28}{30}{Train-calibrated bounded-tanh selector diagnostic. Calibration uses train-split tanh measured rollout rows with same-chart and same-state source retrieval excluded, then evaluates once on held-out test tanh rows}{table.28}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {29}{\ignorespaces Train-calibrated \texttt  {env\_clip} selector diagnostic. Calibration uses train-split \texttt  {env\_clip} measured rollout rows with same-chart and same-state source retrieval excluded, then evaluates once on held-out test \texttt  {env\_clip} rows.}}{30}{table.29}\protected@file@percent }
+\newlabel{tab:ctt-base-context-obs-envclip-learned-train-dominance}{{29}{30}{Train-calibrated \texttt {env\_clip} selector diagnostic. Calibration uses train-split \texttt {env\_clip} measured rollout rows with same-chart and same-state source retrieval excluded, then evaluates once on held-out test \texttt {env\_clip} rows}{table.29}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {30}{\ignorespaces Train-calibrated learned dominance evaluated on the same held-out test rollout rows. Calibration uses only train-split measured generated rollouts with same-chart and same-state source retrieval excluded. This is a cleaner selector diagnostic than validation calibration, but it does not beat the validation-calibrated best row and still fails the deployment gate.}}{30}{table.30}\protected@file@percent }
+\newlabel{tab:ctt-base-context-obs-learned-train-dominance}{{30}{30}{Train-calibrated learned dominance evaluated on the same held-out test rollout rows. Calibration uses only train-split measured generated rollouts with same-chart and same-state source retrieval excluded. This is a cleaner selector diagnostic than validation calibration, but it does not beat the validation-calibrated best row and still fails the deployment gate}{table.30}{}}
+\@writefile{lot}{\contentsline {table}{\numberline {31}{\ignorespaces Nonlinear train-calibrated selector diagnostic. The model and threshold are selected only on held-out train-calibration rows, then evaluated once on the held-out test rollout rows. The best nonlinear row does not beat Table\nobreakspace  {}\ref  {tab:ctt-base-context-obs-learned-train-dominance}, so the current bottleneck is not just linear separability in the dominance selector.}}{30}{table.31}\protected@file@percent }
+\newlabel{tab:ctt-base-context-obs-nonlinear-train-dominance}{{31}{30}{Nonlinear train-calibrated selector diagnostic. The model and threshold are selected only on held-out train-calibration rows, then evaluated once on the held-out test rollout rows. The best nonlinear row does not beat Table~\ref {tab:ctt-base-context-obs-learned-train-dominance}, so the current bottleneck is not just linear separability in the dominance selector}{table.31}{}}
+\gdef \@abspage@last{30}

workspace/latex/main.fdb_latexmk

@@ -1,26 +1,29 @@
 # Fdb version 4
-["bibtex main"] 1783118880 "main.aux" "main.bbl" "main" 1783118881 0
+["bibtex main"] 1783121632 "main.aux" "main.bbl" "main" 1783121633 0
   "./references.bib" 1783037215 3572 027cd2cf27b344368db784320d235a8c ""
   "/cvmfs/soft.computecanada.ca/gentoo/2023/x86-64-v3/usr/share/texmf-dist/bibtex/bst/base/plain.bst" 1292289607 20613 bd3fbfa9f64872b81ac57a0dd2ed855f ""
-  "main.aux" 1783118880 23182 fe2d042142dd9e79639f0bb6637ddf0d "pdflatex"
+  "main.aux" 1783121632 25408 adcf1f2c6cb43a2116bcae59ed042598 "pdflatex"
   (generated)
   "main.bbl"
   "main.blg"
   (rewritten before read)
-["pdflatex"] 1783118880 "main.tex" "main.pdf" "main" 1783118881 0
+["pdflatex"] 1783121632 "main.tex" "main.pdf" "main" 1783121633 0
   "../paper/sections/theory.tex" 1783037322 4023 e8c0485606fdfea0020faae044bf1189 ""
   "../runs/action_bound_audit_rgb_refs/table.tex" 1783102548 379 5a7d42a0f3c58a06c4eeeaa280615fb3 ""
   "../runs/ctt_base_context_obj_val_rollout_comparison/table.tex" 1783103591 2172 16a82c2b9b79cbd8d2afdfca5503a296 ""
+  "../runs/ctt_base_context_obs_learned_dominance_basic_envclip_train_to_test/table.tex" 1783121047 342 4ae8602b8d1e8ca3eb0baa82d0311b8d ""
   "../runs/ctt_base_context_obs_learned_dominance_context_success_tanh_train_to_test/table.tex" 1783107116 342 9bf813f887041aededfb26367a196a15 ""
   "../runs/ctt_base_context_obs_learned_dominance_context_tangent_success_tanh_val_to_test/table.tex" 1783105123 342 dd9a1e5b4a221f68ef356cf5d2587225 ""
   "../runs/ctt_base_context_obs_learned_dominance_context_val_to_test/table.tex" 1783094939 342 289b441acb02d293402c3de06a7e2f7a ""
   "../runs/ctt_base_context_obs_learned_dominance_train_to_test/table.tex" 1783096583 342 20aaeaf8942b48fd162703dc2da041d7 ""
   "../runs/ctt_base_context_obs_nonlinear_dominance_basic_positive_train_to_test/table.tex" 1783097340 344 b41a694c7be179a8b2e13d9326613f99 ""
+  "../runs/ctt_base_context_obs_test_envclip_rollout_comparison/table.tex" 1783120980 2516 76a70f818600a600a645ca511f8631a0 ""
   "../runs/ctt_base_context_obs_test_noclip_rollout_comparison/table.tex" 1783103591 2514 53eb0178d1f721f8d1d39db3b88e715d ""
   "../runs/ctt_base_context_obs_test_perdim_trainmax_rollout_comparison/table.tex" 1783118633 2523 901a0e931d1b515e9dc956b081b93b78 ""
   "../runs/ctt_base_context_obs_test_rollout_comparison/table.tex" 1783103578 2170 0ea3761f3f17352fa51b5f8b4979fb70 ""
   "../runs/ctt_base_context_obs_test_scaled0215_rollout_comparison/table.tex" 1783103317 2523 835503f4c8c1c02ad3e6761d733eef19 ""
   "../runs/ctt_base_context_obs_test_tanh_rollout_comparison/table.tex" 1783104884 2518 7c061ede8cc8e34656555253887a5e0c ""
+  "../runs/ctt_base_context_obs_val_envclip_rollout_comparison/table.tex" 1783120978 2516 59fd79135a4ada5d2783b6937fb1b494 ""
   "../runs/ctt_base_context_obs_val_noclip_rollout_comparison/table.tex" 1783103592 2516 efab11865e5c18448639a1adb45fc9f7 ""
   "../runs/ctt_base_context_obs_val_perdim_trainmax_rollout_comparison/table.tex" 1783118631 2523 ab8f8d646407958b1d62fa9cac540d97 ""
   "../runs/ctt_base_context_obs_val_rollout_comparison/table.tex" 1783103576 2170 ada38b0197c9b24996df6ca37ebf8ce6 ""
@@ -152,10 +155,10 @@
   "/cvmfs/soft.computecanada.ca/gentoo/2023/x86-64-v3/usr/share/texmf-dist/web2c/texmf.cnf" 1692823885 39639 818dd8e5a0dce3a72597a7e6bad45d55 ""
   "/cvmfs/soft.computecanada.ca/gentoo/2023/x86-64-v3/var/lib/texmf/fonts/map/pdftex/updmap/pdftex.map" 1692821535 5031293 07973d5e761f645996ac32ebf7b8d1f3 ""
   "/cvmfs/soft.computecanada.ca/gentoo/2023/x86-64-v3/var/lib/texmf/web2c/pdftex/pdflatex.fmt" 1692821530 1386016 abd9ba6eb53c47bc5de8d672b6ac87d3 ""
-  "main.aux" 1783118880 23182 fe2d042142dd9e79639f0bb6637ddf0d "pdflatex"
-  "main.bbl" 1783118880 3120 92555a174bae4ac684748bbe3fae0b8c "bibtex main"
-  "main.out" 1783118880 2647 6c337a545f287573528c7cb52648bc95 "pdflatex"
-  "main.tex" 1783118688 53729 761f67f13c5ab1c7f0f3d35aa19fe554 ""
+  "main.aux" 1783121632 25408 adcf1f2c6cb43a2116bcae59ed042598 "pdflatex"
+  "main.bbl" 1783121632 3120 92555a174bae4ac684748bbe3fae0b8c "bibtex main"
+  "main.out" 1783121632 2647 6c337a545f287573528c7cb52648bc95 "pdflatex"
+  "main.tex" 1783121374 56782 f93c784a2769b370635a88a96f187542 ""
   "tables/car_decomposition.tex" 1783032219 662 bac24c0014c2ba573860ce199b776934 ""
   "tables/main_results.tex" 1783004655 1061 19bfdcc844614aadefc921dae15c700d ""
   "tables/selector_calibration.tex" 1783011198 788 1b54a8eb88519c50862f4acf718798ab ""

workspace/latex/main.fls

@@ -790,6 +790,28 @@ INPUT ../runs/ctt_base_context_obs_test_perdim_trainmax_rollout_comparison/table
 INPUT ../runs/ctt_base_context_obs_test_perdim_trainmax_rollout_comparison/table.tex
 INPUT ../runs/ctt_base_context_obs_test_perdim_trainmax_rollout_comparison/table.tex
 INPUT ../runs/ctt_base_context_obs_test_perdim_trainmax_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_val_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_val_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_val_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_val_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_val_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_val_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_val_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_val_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_val_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_val_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_val_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_test_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_test_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_test_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_test_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_test_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_test_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_test_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_test_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_test_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_test_envclip_rollout_comparison/table.tex
+INPUT ../runs/ctt_base_context_obs_test_envclip_rollout_comparison/table.tex
 INPUT ../runs/ctt_base_context_obs_val_tanh_rollout_comparison/table.tex
 INPUT ../runs/ctt_base_context_obs_val_tanh_rollout_comparison/table.tex
 INPUT ../runs/ctt_base_context_obs_val_tanh_rollout_comparison/table.tex
@@ -922,6 +944,17 @@ INPUT ../runs/ctt_base_context_obs_learned_dominance_context_success_tanh_train_
 INPUT ../runs/ctt_base_context_obs_learned_dominance_context_success_tanh_train_to_test/table.tex
 INPUT ../runs/ctt_base_context_obs_learned_dominance_context_success_tanh_train_to_test/table.tex
 INPUT ../runs/ctt_base_context_obs_learned_dominance_context_success_tanh_train_to_test/table.tex
+INPUT ../runs/ctt_base_context_obs_learned_dominance_basic_envclip_train_to_test/table.tex
+INPUT ../runs/ctt_base_context_obs_learned_dominance_basic_envclip_train_to_test/table.tex
+INPUT ../runs/ctt_base_context_obs_learned_dominance_basic_envclip_train_to_test/table.tex
+INPUT ../runs/ctt_base_context_obs_learned_dominance_basic_envclip_train_to_test/table.tex
+INPUT ../runs/ctt_base_context_obs_learned_dominance_basic_envclip_train_to_test/table.tex
+INPUT ../runs/ctt_base_context_obs_learned_dominance_basic_envclip_train_to_test/table.tex
+INPUT ../runs/ctt_base_context_obs_learned_dominance_basic_envclip_train_to_test/table.tex
+INPUT ../runs/ctt_base_context_obs_learned_dominance_basic_envclip_train_to_test/table.tex
+INPUT ../runs/ctt_base_context_obs_learned_dominance_basic_envclip_train_to_test/table.tex
+INPUT ../runs/ctt_base_context_obs_learned_dominance_basic_envclip_train_to_test/table.tex
+INPUT ../runs/ctt_base_context_obs_learned_dominance_basic_envclip_train_to_test/table.tex
 INPUT ../runs/ctt_base_context_obs_learned_dominance_train_to_test/table.tex
 INPUT ../runs/ctt_base_context_obs_learned_dominance_train_to_test/table.tex
 INPUT ../runs/ctt_base_context_obs_learned_dominance_train_to_test/table.tex

workspace/latex/main.log

@@ -1,4 +1,4 @@
-This is pdfTeX, Version 3.141592653-2.6-1.40.22 (TeX Live 2021 Gentoo Linux) (preloaded format=pdflatex 2023.8.23)  3 JUL 2026 18:48
+This is pdfTeX, Version 3.141592653-2.6-1.40.22 (TeX Live 2021 Gentoo Linux) (preloaded format=pdflatex 2023.8.23)  3 JUL 2026 19:33
 entering extended mode
  restricted \write18 enabled.
  %&-line parsing enabled.
@@ -502,16 +502,16 @@ File: umsb.fd 2013/01/14 v3.01 AMS symbols B
 tex/updmap/pdftex.map}] [2]
 (../runs/data_accounting/table.tex)
 LaTeX Font Info:    Trying to load font information for OMS+cmtt on input line 
-209.
-LaTeX Font Info:    No file OMScmtt.fd. on input line 209.
+212.
+LaTeX Font Info:    No file OMScmtt.fd. on input line 212.
 
 
 LaTeX Font Warning: Font shape `OMS/cmtt/m/n' undefined
 (Font)              using `OMS/cmsy/m/n' instead
-(Font)              for symbol `textbraceleft' on input line 209.
+(Font)              for symbol `textbraceleft' on input line 212.
 
 [3]
-Overfull \hbox (61.1175pt too wide) in paragraph at lines 275--283
+Overfull \hbox (61.1175pt too wide) in paragraph at lines 278--286
  []\OT1/cmr/bx/n/10 Safety cov-er-age.[] \OT1/cmr/m/n/10 Mea-sured roll-out row
 s may in-clude \OT1/cmtt/m/n/10 safety[]violation \OT1/cmr/m/n/10 in-side \OT1/
 cmtt/m/n/10 base[]outcome \OT1/cmr/m/n/10 and \OT1/cmtt/m/n/10 candidate[]outco
@@ -519,7 +519,7 @@ mes\OT1/cmr/m/n/10 .
  []
 
 
-Overfull \hbox (30.30605pt too wide) in paragraph at lines 275--283
+Overfull \hbox (30.30605pt too wide) in paragraph at lines 278--286
 \OT1/cmr/m/n/10 side the Man-iSkill ac-tion space be-fore op-tional clip-ping, 
 the row records \OT1/cmtt/m/n/10 safety[]violation[]source=action[]bounds\OT1/c
 mr/m/n/10 .
@@ -536,14 +536,14 @@ sets vary Lipschitz-continuously
 (./tables/source_score_sweep.tex) (./tables/selector_calibration.tex)
 (../runs/ctt_residual_smoke_proxy/table.tex)
 (../runs/ctt_val_proxy_comparison/table.tex) [7]
-Overfull \hbox (21.77454pt too wide) in paragraph at lines 458--465
+Overfull \hbox (21.77454pt too wide) in paragraph at lines 461--468
 \OT1/cmr/m/n/10 base plus gen-er-ated can-di-dates in Man-iSkill, and writes me
 a-sured rows con-sum-able by \OT1/cmtt/m/n/10 scripts/eval[]metrics.py\OT1/cmr/
 m/n/10 .
  []
 
 
-Overfull \hbox (131.08893pt too wide) in paragraph at lines 466--472
+Overfull \hbox (131.08893pt too wide) in paragraph at lines 469--475
 []\OT1/cmr/m/n/10 A one-chart smoke run at K=2 ver-i-fies the mea-sured path an
 d re-store ac-cu-racy (\OT1/cmtt/m/n/10 runs/ctt[]residual[]rollout[]direct[]sm
 oke[]seed0[]v3\OT1/cmr/m/n/10 ),
@@ -552,19 +552,19 @@ oke[]seed0[]v3\OT1/cmr/m/n/10 ),
 (../runs/action_bound_audit_rgb_refs/table.tex) [8]
 (../runs/ctt_base_context_obs_val_noclip_rollout_comparison/table.tex)
 (../runs/ctt_base_context_obs_test_noclip_rollout_comparison/table.tex)
-Underfull \hbox (badness 10000) in paragraph at lines 555--555
+Underfull \hbox (badness 10000) in paragraph at lines 558--558
 []\OT1/cmr/m/n/10 Table 11: []Scaled raw-action val-i-da-tion re-fresh for \OT1
 /cmtt/m/n/10 base[]context[]obs\OT1/cmr/m/n/10 , K=8, with
  []
 
 (../runs/ctt_base_context_obs_val_scaled0215_rollout_comparison/table.tex)
-Underfull \hbox (badness 10000) in paragraph at lines 566--566
+Underfull \hbox (badness 10000) in paragraph at lines 569--569
 []\OT1/cmr/m/n/10 Table 12: []Scaled raw-action held-out test re-fresh for \OT1
 /cmtt/m/n/10 base[]context[]obs\OT1/cmr/m/n/10 , K=8, with
  []
 
 
-Underfull \hbox (badness 1082) in paragraph at lines 566--566
+Underfull \hbox (badness 1082) in paragraph at lines 569--569
 \OT1/cmtt/m/n/10 --execution-action-scale 0.215 \OT1/cmr/m/n/10 and clip-ping d
 is-abled. The global scale im-proves base action-
  []
@@ -573,98 +573,102 @@ is-abled. The global scale im-proves base action-
 (../runs/ctt_base_context_obs_val_perdim_trainmax_rollout_comparison/table.tex)
 
 (../runs/ctt_base_context_obs_test_perdim_trainmax_rollout_comparison/table.tex
-) (../runs/ctt_base_context_obs_val_tanh_rollout_comparison/table.tex)
-(../runs/ctt_base_context_obs_test_tanh_rollout_comparison/table.tex)
-(../runs/ctt_val_rollout_comparison/table.tex) [9]
+) (../runs/ctt_base_context_obs_val_envclip_rollout_comparison/table.tex)
+(../runs/ctt_base_context_obs_test_envclip_rollout_comparison/table.tex)
+(../runs/ctt_base_context_obs_val_tanh_rollout_comparison/table.tex)
+(../runs/ctt_base_context_obs_test_tanh_rollout_comparison/table.tex) [9]
+(../runs/ctt_val_rollout_comparison/table.tex)
 (../runs/ctt_base_context_obs_val_rollout_comparison/table.tex)
 (../runs/ctt_base_context_obj_val_rollout_comparison/table.tex)
 (../runs/ctt_test_rollout_comparison/table.tex)
 (../runs/ctt_base_context_obs_test_rollout_comparison/table.tex)
 (../runs/ctt_dominance_val_to_test/table.tex)
-(../runs/ctt_learned_dominance_val_to_test/table.tex)
+(../runs/ctt_learned_dominance_val_to_test/table.tex) [10]
 (../runs/ctt_base_context_obs_learned_dominance_context_val_to_test/table.tex)
 (../runs/ctt_base_context_obs_learned_dominance_context_tangent_success_tanh_va
-l_to_test/table.tex) [10]
+l_to_test/table.tex)
 (../runs/ctt_base_context_obs_learned_dominance_context_success_tanh_train_to_t
 est/table.tex)
-(../runs/ctt_base_context_obs_learned_dominance_train_to_test/table.tex)
+(../runs/ctt_base_context_obs_learned_dominance_basic_envclip_train_to_test/tab
+le.tex) (../runs/ctt_base_context_obs_learned_dominance_train_to_test/table.tex
+)
 (../runs/ctt_base_context_obs_nonlinear_dominance_basic_positive_train_to_test/
 table.tex)
-Overfull \hbox (31.94885pt too wide) in paragraph at lines 925--929
+Overfull \hbox (31.94885pt too wide) in paragraph at lines 983--987
 []\OT1/cmtt/m/n/10 scripts/export[]cil[]charts.py\OT1/cmr/m/n/10 , \OT1/cmtt/m/
 n/10 scripts/build[]data[]accounting.py\OT1/cmr/m/n/10 , and \OT1/cmtt/m/n/10 s
 cripts/audit[]cil[]charts.py
  []
 
-
-Overfull \hbox (127.11275pt too wide) in paragraph at lines 931--935
+[11]
+Overfull \hbox (127.11275pt too wide) in paragraph at lines 989--993
 []\OT1/cmtt/m/n/10 cil/chart[]features.py \OT1/cmr/m/n/10 cen-tral-izes deploym
 ent-visible chart fea-ture con-struc-tion, and \OT1/cmtt/m/n/10 scripts/audit[]
 chart[]feature[]sources.py
  []
 
 
-Overfull \hbox (104.78587pt too wide) in paragraph at lines 935--938
+Overfull \hbox (104.78587pt too wide) in paragraph at lines 993--996
 []\OT1/cmtt/m/n/10 scripts/slurm/render[]six[]task[]chart[]observations.sbatch 
 \OT1/cmr/m/n/10 and \OT1/cmtt/m/n/10 scripts/slurm/reexport[]rgb[]ref[]cil[]cha
 rts.sbatch
  []
 
 
-Overfull \hbox (23.72661pt too wide) in paragraph at lines 938--943
+Overfull \hbox (23.72661pt too wide) in paragraph at lines 996--1001
 []\OT1/cmtt/m/n/10 scripts/export[]chart[]observation[]embeddings.py \OT1/cmr/m
 /n/10 and \OT1/cmtt/m/n/10 scripts/export[]chart[]object[]embeddings.py
  []
 
 
-Overfull \hbox (129.91359pt too wide) in paragraph at lines 938--943
+Overfull \hbox (129.91359pt too wide) in paragraph at lines 996--1001
 \OT1/cmr/m/n/10 cre-ate de-ter-min-is-tic 32D RGB-stat and 64D RGB object-layou
 t em-bed-dings, and \OT1/cmtt/m/n/10 scripts/slurm/train[]ctt[]feature[]proxy.s
 batch
  []
 
 
-Overfull \hbox (7.97675pt too wide) in paragraph at lines 943--952
+Overfull \hbox (7.97675pt too wide) in paragraph at lines 1001--1010
 []\OT1/cmtt/m/n/10 scripts/eval[]ctt[]generated[]rollout.py \OT1/cmr/m/n/10 and
  \OT1/cmtt/m/n/10 scripts/slurm/eval[]ctt[]generated[]rollout.sbatch
  []
 
 
-Overfull \hbox (20.61057pt too wide) in paragraph at lines 943--952
+Overfull \hbox (20.61057pt too wide) in paragraph at lines 1001--1010
 \OT1/cmr/m/n/10 train-split cal-i-bra-tion and meta-data load-ing for deploymen
 t-visible chart fea-tures such as \OT1/cmtt/m/n/10 base[]context[]obs\OT1/cmr/m
 /n/10 .
  []
 
 
-Overfull \hbox (49.05014pt too wide) in paragraph at lines 943--952
+Overfull \hbox (49.05014pt too wide) in paragraph at lines 1001--1010
 \OT1/cmr/m/n/10 They also record action-bound va-lid-ity la-bels and sup-port r
 aw-action re-play through \OT1/cmtt/m/n/10 --disable-action-clipping
  []
 
-[11]
-Overfull \hbox (11.7557pt too wide) in paragraph at lines 952--955
+
+Overfull \hbox (11.7557pt too wide) in paragraph at lines 1010--1013
 []\OT1/cmtt/m/n/10 scripts/audit[]action[]bounds.py \OT1/cmr/m/n/10 pro-duces \
 OT1/cmtt/m/n/10 runs/action[]bound[]audit[]rgb[]refs\OT1/cmr/m/n/10 , which au-
 dits whether
  []
 
 
-Overfull \hbox (6.0625pt too wide) in paragraph at lines 971--974
+Overfull \hbox (6.0625pt too wide) in paragraph at lines 1029--1032
 []\OT1/cmtt/m/n/10 scripts/eval[]chart[]positive[]memory[]proxy.py \OT1/cmr/m/n
 /10 and \OT1/cmtt/m/n/10 scripts/build[]ctt[]proxy[]comparison.py \OT1/cmr/m/n/
 10 gen-
  []
 
 
-Overfull \hbox (20.3325pt too wide) in paragraph at lines 976--979
+Overfull \hbox (20.3325pt too wide) in paragraph at lines 1034--1037
 []\OT1/cmtt/m/n/10 scripts/train[]utility[]energy.py \OT1/cmr/m/n/10 and \OT1/c
 mtt/m/n/10 scripts/calibrate[]dominance.py \OT1/cmr/m/n/10 im-ple-ment the util
 -ity/scoring
  []
 
-[12] (./main.bbl) [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23]
-[24] [25] [26] [27] (./main.aux)
+[12] (./main.bbl [13]) [14] [15] [16] [17] [18] [19] [20] [21] [22] [23]
+[24] [25] [26] [27] [28] [29] [30] (./main.aux)
 
 LaTeX Font Warning: Some font shapes were not available, defaults substituted.
 
@@ -672,13 +676,13 @@ Package rerunfilecheck Info: File `main.out' has not changed.
 (rerunfilecheck)             Checksum: 6C337A545F287573528C7CB52648BC95;2647.
  ) 
 Here is how much of TeX's memory you used:
- 10227 strings out of 480884
- 156066 string characters out of 5900692
- 657059 words of memory out of 5000000
- 27141 multiletter control sequences out of 15000+600000
+ 10248 strings out of 480884
+ 157127 string characters out of 5900692
+ 692905 words of memory out of 5000000
+ 27147 multiletter control sequences out of 15000+600000
  414764 words of font info for 71 fonts, out of 8000000 for 9000
  36 hyphenation exceptions out of 8191
- 71i,10n,74p,608b,415s stack positions out of 5000i,500n,10000p,200000b,80000s
+ 71i,10n,74p,608b,446s stack positions out of 5000i,500n,10000p,200000b,80000s
 {/cvmfs/soft.computecanada.ca/gentoo/2023/x86-64-v3/usr/share/texmf-dist/font
 s/enc/dvips/cm-super/cm-super-ts1.enc}</cvmfs/soft.computecanada.ca/gentoo/2023
 /x86-64-v3/usr/share/texmf-dist/fonts/type1/public/amsfonts/cm/cmbx10.pfb></cvm
@@ -719,10 +723,10 @@ cm/cmtt9.pfb></cvmfs/soft.computecanada.ca/gentoo/2023/x86-64-v3/usr/share/texm
 f-dist/fonts/type1/public/amsfonts/symbols/msbm10.pfb></cvmfs/soft.computecanad
 a.ca/gentoo/2023/x86-64-v3/usr/share/texmf-dist/fonts/type1/public/cm-super/sfr
 m1000.pfb>
-Output written on main.pdf (27 pages, 405832 bytes).
+Output written on main.pdf (30 pages, 416341 bytes).
 PDF statistics:
- 409 PDF objects out of 1000 (max. 8388607)
- 349 compressed objects within 4 object streams
- 100 named destinations out of 1000 (max. 500000)
+ 426 PDF objects out of 1000 (max. 8388607)
+ 363 compressed objects within 4 object streams
+ 106 named destinations out of 1000 (max. 500000)
  137 words of extra memory for PDF output out of 10000 (max. 10000000)
 

workspace/latex/main.pdf

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2018c89d5b0d1c3eda8a63b0cbd612cb7fb9bab88b77f6d3ce38240c1eb307e5
-size 405832
+oid sha256:121e6bd969e8c09d89d1a759624f5cfd1c240cda9e7e2c73924df99251c703e2
+size 416341

workspace/latex/main.tex

@@ -61,7 +61,10 @@ selector reaches 31.25\% test success at 25.69\% coverage. A bounded
 train-calibrated context selector reaches 38.19\% selected test success, matching
 the tanh proposal oracle in micro success. The row-level selector-gap metric is
 still 12.50 points, and the proposal support itself is lower than the clipped
-visual-stat row. We keep the metric boundary explicit:
+visual-stat row. An explicit \texttt{env\_clip} diagnostic is action-bound-clean
+and preserves more proposal support (47.92\% on held-out test), but its best
+train-calibrated selector reaches only 31.94\% selected success. We keep the
+metric boundary explicit:
 \outcomeptr{} is reported only after generated candidates are rolled out, while
 distance-only support diagnostics are \pptc{}. These runs are therefore
 diagnostic evidence for the Atlas thesis and the next selector/generator gate,
@@ -609,6 +612,40 @@ success is 22.22\%, selected success is 11.81\%, and \outcomeptr{}@8 is
 36.11\%. This rejects another simple post-processing convention and leaves the
 decoder/action representation itself as the blocker.
 
+\begin{table}[t]
+\centering
+\caption{Explicit \texttt{env\_clip} validation refresh for
+\texttt{base\_context\_obs}, K=8, with clipping disabled. The declared decoder
+convention clips decoded controls to action-space bounds before validity checks,
+so action-bound labels measure the declared convention rather than silent
+simulator clipping.}
+\label{tab:ctt-base-context-obs-val-envclip-rollout}
+\scriptsize
+\resizebox{\linewidth}{!}{\input{../runs/ctt_base_context_obs_val_envclip_rollout_comparison/table}}
+\end{table}
+
+\begin{table}[t]
+\centering
+\caption{Explicit \texttt{env\_clip} held-out test refresh for
+\texttt{base\_context\_obs}, K=8, with clipping disabled. This convention is
+action-bound-clean and preserves more proposal support than bounded
+\texttt{tanh}, but score-only selection remains below base.}
+\label{tab:ctt-base-context-obs-test-envclip-rollout}
+\scriptsize
+\resizebox{\linewidth}{!}{\input{../runs/ctt_base_context_obs_test_envclip_rollout_comparison/table}}
+\end{table}
+
+The explicit \texttt{env\_clip} refresh is the strongest bounded-action support
+diagnostic so far. Generated, selected, and base action-bound violation rates
+are all 0.00\% with 100\% known labels. On validation, \outcomeptr{}@8 is
+50.24\%, proposal-oracle success is 41.55\%, selected success is 22.22\%, and
+base success is 27.54\%. On held-out test, \outcomeptr{}@8 is 50.69\%,
+proposal-oracle success is 47.92\%, selected success is 23.61\%, and base
+success is 29.17\%. Thus \texttt{env\_clip} largely removes the action-bound
+validity confound without the severe support collapse of per-dimension scaling
+or bounded \texttt{tanh}, but it exposes the remaining selector/dominance
+failure.
+
 \begin{table}[t]
 \centering
 \caption{Bounded \texttt{tanh} validation refresh for
@@ -859,6 +896,27 @@ success selector gap remains 12.50 points and CTT no longer generates enough
 successful proposals. Support, with residual selection error, remains the main
 blocker for this convention.
 
+\begin{table}[t]
+\centering
+\caption{Train-calibrated \texttt{env\_clip} selector diagnostic. Calibration
+uses train-split \texttt{env\_clip} measured rollout rows with same-chart and
+same-state source retrieval excluded, then evaluates once on held-out test
+\texttt{env\_clip} rows.}
+\label{tab:ctt-base-context-obs-envclip-learned-train-dominance}
+\scriptsize
+\resizebox{\linewidth}{!}{\input{../runs/ctt_base_context_obs_learned_dominance_basic_envclip_train_to_test/table}}
+\end{table}
+
+Table~\ref{tab:ctt-base-context-obs-envclip-learned-train-dominance} tests
+whether the better bounded proposal support translates through train-clean
+selection. It does not yet. The best current env-clip selector is the basic
+utility-margin ridge row: selected held-out test success reaches 31.94\% at
+22.22\% coverage, above the env-clip base action at 29.17\% but far below the
+47.92\% proposal oracle. The success selector gap remains 21.53 points. The
+method evidence therefore separates the action-convention fix from the
+remaining dominance problem: \texttt{env\_clip} preserves bounded proposal
+support, but current selector features do not exploit it.
+
 The feature-source audits turn this negative result into a concrete engineering
 target. In the original \texttt{data/cil\_charts} indexes, scene ids and
 instructions are present, but observation embeddings and raw observation