Add Luna 6 prompt-development charts and tables; tighten Qwen chart headers

SHA256SUMS

@@ -6872,8 +6872,8 @@ cb3c0dd86b1b7be2383765711ac69d541dfcd0f549180a1921826cd7c7209bdc  media/frames/e
 dd16282f2437a28f85e7ba5d1f7867ff3fc51e483db798ce3028d515637a1d45  media/frames/evaluation/ego_view_pad_res256_freq20/part-0004.tar
 45f167743c8e7151229d2fd60f14710fcedc39361f1e67166812b2e726c02698  media/frames/prompt_development/ego_view_bg_crop_pad_res256_freq20/part-0000.tar
 2175258e1b78aacd0613774599a503e9b481e74c1e7d2454209e8436920cb8db  media/frames/prompt_development/ego_view_pad_res256_freq20/part-0000.tar
-ccb32ee42ca233a99bdbc18cb3b977af6babbf235ffece5b9a810885c2d6ce39  paper/README.md
-579dad342031f182ae7b44240562a791f7513b1a444a17f0517db950340889b5  paper/SHA256SUMS
+f9fb5fa24c912521a86442d026f0aa93fe7477dc41d92746a017d6365ddb82b5  paper/README.md
+bda01827de98bde33a4f9918ba770db68a72ce0f4e0782273e54721d8dc46c48  paper/SHA256SUMS
 c9c4a01691083d984b752349f1c0a307b828aa6b7e8c2229d469aa87bc1de040  paper/data/episodes.csv
 a462c2f6848c89b2346317f9d09a21939dd8f861123274a955a1b57490f9a1b2  paper/figures/c01_gemma4.pdf
 04add1e52c310ebade9a90b88a7ec9b8f008a31b8fa77a087b89b59decfc86f3  paper/figures/c01_gemma4.png
@@ -6921,22 +6921,34 @@ d358e7709a1ad626c0fdce59090fa55bb6a72174ca2807b6595f72db2e8d594a  paper/figures/
 08102587c7299b2f32dcdf67d33e9f1c42e5b2b54e0ed725a018300d0a7cd5fd  paper/figures/f10_timing.png
 660d3a9ba63753eb30ad44f6c0b7d236a3a002d406815daba45f8a13133b3c54  paper/figures/f10_timing_side.pdf
 d73632def8217848f18d6d04f3d7aa08b6827ae3cb15d2ea1437ec91e278d790  paper/figures/f10_timing_side.png
-a719701f31d5676fc7e907263425acc506678beea52cef68d58f041c08a527ed  paper/index.html
+27abfd9b5428bd58c3a29cbb5ca9d0cbdd1da78734520b434ca73343578e83f2  paper/index.html
 b987cb6438d89a8416e0779135bcec899941e5f0fcf753f517d75e3054b4d640  paper/manifest.json
 37c54ae2ede45681497e06b806386ff6837d460b9cf9120961b09f0d3c1e9125  paper/prompt_development/data/episodes.csv
-de69766eaa16c0c111bc49d97ae04f984dcefcb3ff032d26d9c9e51b384a0f94  paper/prompt_development/figures/p01_qwen38.pdf
-08c3bddc790aa3ad09a1d3ffc2054fce54501c5de13ed153905eedfc5ffba5f1  paper/prompt_development/figures/p01_qwen38.png
-878dcc258318382d8950b1224296bb9a17a86bc8900da535d33a499eb17455ca  paper/prompt_development/figures/p01_qwen38_side.pdf
-71ff520e87330759ae6c363be46dee57827e5235bd6198f703a0ac187e4a67a3  paper/prompt_development/figures/p01_qwen38_side.png
-ea41392e51e7ff3d2698b193fe19e31ec27d7bc5c93b65ac70e74d429f9bfb1c  paper/prompt_development/index.html
-62620f5917310fb374e5d55fc28c2257908c217611a3ae80ba8ec4c92885c2d8  paper/prompt_development/manifest.json
+8e27f0832e09535b9abc606aa4ec605f434e18bf170fd4f46c4c11a1882cf9ec  paper/prompt_development/figures/p01_qwen38.pdf
+ebd6235b4c7a253b5e64a668d5036a70d95a4e8264f92a155edbd090647fa5be  paper/prompt_development/figures/p01_qwen38.png
+aaa7ac11c0717d38a08d7c4086e54b0eeb5a93bbbd77c48c603233850668750b  paper/prompt_development/figures/p01_qwen38_side.pdf
+57c6fb9753b57abf0bd28c29c213c096dac423dc2e4bc5d3b97c5a562a39cdc6  paper/prompt_development/figures/p01_qwen38_side.png
+fcd823655804af4ee76b42f4c037ef2da18e1a31b3eefece14b7d58a1e676e8c  paper/prompt_development/figures/p05_luna6.pdf
+eb459330a943259b5ccd7180fa390135cfd89277a1910370774c93a48b4cdbbc  paper/prompt_development/figures/p05_luna6.png
+cfb56f577104b201d57f7ba66545f52212dec998047c873743cd7ceebeeb989e  paper/prompt_development/figures/p05_luna6_side.pdf
+12b256aa22354524e3d722b24d251225237067fb32baf963d079403a7c8ee587  paper/prompt_development/figures/p05_luna6_side.png
+738ab43f39c43d151c8972f79355e5e30c62ee954b94a1c4f8b16dddee0f14b2  paper/prompt_development/index.html
+0c2fbd44e78fd3d208555cd70be432292f7fc9715e46259e17a7cc26cd77f4fc  paper/prompt_development/manifest.json
 bd3a24a285d0c74fe3f039adec0d523fd7af44605855cd091fe26d6df0ed26e2  paper/prompt_development/tables/p02_detail.csv
-114c0c7c89b152915adc0ac45e8a4e1ac282fa88a5bc49f620b3d8f8184064b9  paper/prompt_development/tables/p02_detail.html
+0189a47a4cc74b6380a7d2d42a6dc19ab360e4e8237614a9b2c28f99db9b52f5  paper/prompt_development/tables/p02_detail.html
 4afc0babb397400e994af4da9ecd06a21d22e7efcef6822b0446c95ef8791ffa  paper/prompt_development/tables/p03_counts.csv
-9b76a85a28026178a9807cbaf267dff7974f463d5411f0cd9f145c72a74b4d9c  paper/prompt_development/tables/p03_counts.html
+e3caac8a5541af6521f96cb2409ef128b87f6617051633e2e619668d94cf633a  paper/prompt_development/tables/p03_counts.html
+9cd4a401692c7ec1b514300807a09203ee51029887fb7dba0a4b70d455219307  paper/prompt_development/tables/p03_counts.tex
 764338885df8515b3e0fc5bdbb6cfd5a1ec27d1b6563eb4eb408cea4ef19ab06  paper/prompt_development/tables/p04_coverage.csv
-b2ecec928c4763107de3e40b7c318a70e930bf2cedb51dc1f9d3cf66a020d4e5  paper/prompt_development/tables/p04_coverage.html
-016cf87890c2cfb65dae4b5e5de8ee183422a11a07728a520e523a4d579cdc12  paper/release_provenance.json
+ebde0f06dfe8c6a60694cc48415359ad94922ceee49cd727d0d5e6e0a1845b9d  paper/prompt_development/tables/p04_coverage.html
+4042e07a4b6ecd82c17dd3acf191af8ee1aa81fd6b6572aa6ce40933f59f320e  paper/prompt_development/tables/p06_luna6_detail.csv
+ae8941b164164093bb46ddc031ef731fbfea4cc26c3713cb4660497459592ceb  paper/prompt_development/tables/p06_luna6_detail.html
+e1779e4e5f36cdf977bdab437a90f880a6b090e37761f1990f59c9d2fe31ed16  paper/prompt_development/tables/p07_luna6_counts.csv
+13954aea4756a20631f7b700a2d5903d246b4fce6810b256a3e5210cfab5cea8  paper/prompt_development/tables/p07_luna6_counts.html
+d4266a083040824e15d66ff10b37cb292996fadb258d94249214f02ee697e48b  paper/prompt_development/tables/p07_luna6_counts.tex
+9f9cde19f12521ff2eece2bd9f1e441078ef5fecf07f494bfba9cdbed6e63699  paper/prompt_development/tables/p08_luna6_coverage.csv
+6f19ba6da25f1bd7e1bbb26d2c135671cffc8c7dcc757cc09210c0cfcc8163fe  paper/prompt_development/tables/p08_luna6_coverage.html
+b076f066124150cfc41cbcd54fab5f842e6b9eb8023441321801f84817e951a8  paper/release_provenance.json
 2a7f650d3227736b9e98c66b05d9827419f39018b08a0ea1f6b8c145c43c9c41  paper/tables/d01_corpus.csv
 9d2cf35b01e05ae461934345caf8188788a08b9434d13340a810dd48a304f878  paper/tables/d01_corpus.html
 e107bacf986ab3debc793be5f83582f22095743555635a522dd5d19d40540b80  paper/tables/d02_endpoint.csv
@@ -7012,6 +7024,7 @@ bf3f48dc850aa9742a97978e379675adc13130bcc6c1343d0096e9faf62a6dfa  provenance/eva
 8c7083caee3919914015904bbafb5db5eef96c4b465e5c9b75bfac29b8d34622  provenance/huggingface_offline_validation.json
 bad51eb4f8619962f007feea60a43481798fbbb2fc9ed7ce00ec7d578957ee50  provenance/ordinal_reconciliation.json
 f422c7c34af12b8dba9b7182efe9aa1a7c029205c18b1708bef29ffa2150bab0  provenance/privacy_transformations.json
+3226a080b944bcebda509293d3fd5e83991e643518a79b21b7cf59a0a3cc0b14  provenance/prompt_development.json
 718bc37d536110d252ba3666f75f950dd704376ea1ec4b0906b0ae165288ce79  provenance/prompt_development_episodes.txt
 6d757625ac9b00ec2b5cc6757d3f444963cb6c1bd764dc82b287a4d419266a56  provenance/prompts/6d757625ac9b00ec2b5cc6757d3f444963cb6c1bd764dc82b287a4d419266a56.txt
 75adbbaf49c5f7ce77a73229d97d44716e750c48f257f67309dc1805b4c1da50  provenance/prompts/75adbbaf49c5f7ce77a73229d97d44716e750c48f257f67309dc1805b4c1da50.txt
@@ -7026,7 +7039,7 @@ f66f8c05bf1a018fd610739e4dde7950cefb68265f90df0ec4ef884174e2f898  provenance/pro
 0095f9a15bc96390db0148a556a406cbb959cb0c4ff40c9a2fa2bd7beb0a2f8b  provenance/quality_issues.json
 ad15ba0bc241c7ab97352cecbcfdf88c1fcc001850f753ef091a435ee2af9524  provenance/source_files.jsonl
 5854fec1a03f0dc7cfcb3a2c2658fc91665da85d3bdff52bf2235afd3f073373  provenance/summary_exports.json
-c478733fe9eb04c760d7304550b75b872556d4bc61d0c9786e9b3448ff4064a6  release_manifest.json
+378b7e58713b5bd1b554f12e8dd9c55874a534dd721d89f3651dcf67fef7b743  release_manifest.json
 7d49c306861805f49b5cf723b1ff09cfb342de2758342de707a16e1d11fa8e34  summaries/btaudit2_all_episodes_visualization_observations.csv
 2c2712b31d6f3c8606f3e26cbb9ac12d709b04dc0e342cd09c2c4e1fffe70b49  summaries/btaudit2_all_episodes_visualization_rows.csv
 8f1a6fda3147b6ab5ff59633993e42cf9c1a2928e368a6e2654ef80c2709df7d  summaries/btaudit2_episodes_2_21_visualization_observations.csv
@@ -7041,4 +7054,4 @@ a9d43e836a7132f6ab835789f1e83d1e3a1e5e2179ca772eeb99ff48438a671d  summaries/btau
 ffb1261942f6fd51663800ec6fe595477386108ff64fcf92a7ee0ae8a8dafaba  summaries/sheet0_progress_review_dry_run_20260805.csv
 5418806012f49f445b9643ae764a2dde36eb5240abbb0b408468b4ec5e76ac1f  summaries/sheet0_progress_review_report_20260805.csv
 28dea2bbcef09f25b2aa2dfba992f156c6dee13253077b695f18047c2f96fc3a  validation.json
-91e48050b8d22fb137ba2222b69bc6b009ecc576a8d22257c606984feafc5761  vibtarve_paper.zip
+152cace4bb53001b3fa2601d50cd2bf0bd6c42817360a7a11285c1cb62148cd6  vibtarve_paper.zip

paper/README.md

@@ -6,7 +6,7 @@ Open [index.html](index.html) for titles, descriptions, previews, and downloads.
 - [Catalog and provenance](manifest.json).
 - `figures/`: short-labeled vector PDFs and 300 dpi PNGs, 7.2 inches wide, minimum 9.5 pt text.
 - `*_side.pdf` / `*_side.png`: side-by-side alternatives for every two-/three-panel chart; 14.4-inch masters with minimum 19 pt text (9.5 pt at a 7.2-inch print width).
-- `prompt_development/`, when explicitly exported: separate Qwen3.8 v1--v5 episode-1 development comparison and its own source datasheet; never pooled into primary performance.
+- `prompt_development/`, when explicitly exported: separate Qwen3.8 and Luna 6 v1--v5 episode-1 development comparison and its own source datasheet; never pooled into primary performance.
 - `tables/t*.tex`: five compact, captioned manuscript tables. Include `\usepackage{booktabs}`; use `\input{tables/t02_mission.tex}` in LaTeX. `table*` spans both columns; no shrink-to-fit wrapper is used.
 - All tables also have exact CSV and titled HTML versions. Detailed `d*` and evidence `e*` tables are supporting data, not oversized manuscript tables.
 

paper/SHA256SUMS

@@ -1,4 +1,4 @@
-ccb32ee42ca233a99bdbc18cb3b977af6babbf235ffece5b9a810885c2d6ce39  README.md
+f9fb5fa24c912521a86442d026f0aa93fe7477dc41d92746a017d6365ddb82b5  README.md
 c9c4a01691083d984b752349f1c0a307b828aa6b7e8c2229d469aa87bc1de040  data/episodes.csv
 a462c2f6848c89b2346317f9d09a21939dd8f861123274a955a1b57490f9a1b2  figures/c01_gemma4.pdf
 04add1e52c310ebade9a90b88a7ec9b8f008a31b8fa77a087b89b59decfc86f3  figures/c01_gemma4.png
@@ -46,22 +46,34 @@ d358e7709a1ad626c0fdce59090fa55bb6a72174ca2807b6595f72db2e8d594a  figures/f10_ti
 08102587c7299b2f32dcdf67d33e9f1c42e5b2b54e0ed725a018300d0a7cd5fd  figures/f10_timing.png
 660d3a9ba63753eb30ad44f6c0b7d236a3a002d406815daba45f8a13133b3c54  figures/f10_timing_side.pdf
 d73632def8217848f18d6d04f3d7aa08b6827ae3cb15d2ea1437ec91e278d790  figures/f10_timing_side.png
-a719701f31d5676fc7e907263425acc506678beea52cef68d58f041c08a527ed  index.html
+27abfd9b5428bd58c3a29cbb5ca9d0cbdd1da78734520b434ca73343578e83f2  index.html
 b987cb6438d89a8416e0779135bcec899941e5f0fcf753f517d75e3054b4d640  manifest.json
 37c54ae2ede45681497e06b806386ff6837d460b9cf9120961b09f0d3c1e9125  prompt_development/data/episodes.csv
-de69766eaa16c0c111bc49d97ae04f984dcefcb3ff032d26d9c9e51b384a0f94  prompt_development/figures/p01_qwen38.pdf
-08c3bddc790aa3ad09a1d3ffc2054fce54501c5de13ed153905eedfc5ffba5f1  prompt_development/figures/p01_qwen38.png
-878dcc258318382d8950b1224296bb9a17a86bc8900da535d33a499eb17455ca  prompt_development/figures/p01_qwen38_side.pdf
-71ff520e87330759ae6c363be46dee57827e5235bd6198f703a0ac187e4a67a3  prompt_development/figures/p01_qwen38_side.png
-ea41392e51e7ff3d2698b193fe19e31ec27d7bc5c93b65ac70e74d429f9bfb1c  prompt_development/index.html
-62620f5917310fb374e5d55fc28c2257908c217611a3ae80ba8ec4c92885c2d8  prompt_development/manifest.json
+8e27f0832e09535b9abc606aa4ec605f434e18bf170fd4f46c4c11a1882cf9ec  prompt_development/figures/p01_qwen38.pdf
+ebd6235b4c7a253b5e64a668d5036a70d95a4e8264f92a155edbd090647fa5be  prompt_development/figures/p01_qwen38.png
+aaa7ac11c0717d38a08d7c4086e54b0eeb5a93bbbd77c48c603233850668750b  prompt_development/figures/p01_qwen38_side.pdf
+57c6fb9753b57abf0bd28c29c213c096dac423dc2e4bc5d3b97c5a562a39cdc6  prompt_development/figures/p01_qwen38_side.png
+fcd823655804af4ee76b42f4c037ef2da18e1a31b3eefece14b7d58a1e676e8c  prompt_development/figures/p05_luna6.pdf
+eb459330a943259b5ccd7180fa390135cfd89277a1910370774c93a48b4cdbbc  prompt_development/figures/p05_luna6.png
+cfb56f577104b201d57f7ba66545f52212dec998047c873743cd7ceebeeb989e  prompt_development/figures/p05_luna6_side.pdf
+12b256aa22354524e3d722b24d251225237067fb32baf963d079403a7c8ee587  prompt_development/figures/p05_luna6_side.png
+738ab43f39c43d151c8972f79355e5e30c62ee954b94a1c4f8b16dddee0f14b2  prompt_development/index.html
+0c2fbd44e78fd3d208555cd70be432292f7fc9715e46259e17a7cc26cd77f4fc  prompt_development/manifest.json
 bd3a24a285d0c74fe3f039adec0d523fd7af44605855cd091fe26d6df0ed26e2  prompt_development/tables/p02_detail.csv
-114c0c7c89b152915adc0ac45e8a4e1ac282fa88a5bc49f620b3d8f8184064b9  prompt_development/tables/p02_detail.html
+0189a47a4cc74b6380a7d2d42a6dc19ab360e4e8237614a9b2c28f99db9b52f5  prompt_development/tables/p02_detail.html
 4afc0babb397400e994af4da9ecd06a21d22e7efcef6822b0446c95ef8791ffa  prompt_development/tables/p03_counts.csv
-9b76a85a28026178a9807cbaf267dff7974f463d5411f0cd9f145c72a74b4d9c  prompt_development/tables/p03_counts.html
+e3caac8a5541af6521f96cb2409ef128b87f6617051633e2e619668d94cf633a  prompt_development/tables/p03_counts.html
+9cd4a401692c7ec1b514300807a09203ee51029887fb7dba0a4b70d455219307  prompt_development/tables/p03_counts.tex
 764338885df8515b3e0fc5bdbb6cfd5a1ec27d1b6563eb4eb408cea4ef19ab06  prompt_development/tables/p04_coverage.csv
-b2ecec928c4763107de3e40b7c318a70e930bf2cedb51dc1f9d3cf66a020d4e5  prompt_development/tables/p04_coverage.html
-016cf87890c2cfb65dae4b5e5de8ee183422a11a07728a520e523a4d579cdc12  release_provenance.json
+ebde0f06dfe8c6a60694cc48415359ad94922ceee49cd727d0d5e6e0a1845b9d  prompt_development/tables/p04_coverage.html
+4042e07a4b6ecd82c17dd3acf191af8ee1aa81fd6b6572aa6ce40933f59f320e  prompt_development/tables/p06_luna6_detail.csv
+ae8941b164164093bb46ddc031ef731fbfea4cc26c3713cb4660497459592ceb  prompt_development/tables/p06_luna6_detail.html
+e1779e4e5f36cdf977bdab437a90f880a6b090e37761f1990f59c9d2fe31ed16  prompt_development/tables/p07_luna6_counts.csv
+13954aea4756a20631f7b700a2d5903d246b4fce6810b256a3e5210cfab5cea8  prompt_development/tables/p07_luna6_counts.html
+d4266a083040824e15d66ff10b37cb292996fadb258d94249214f02ee697e48b  prompt_development/tables/p07_luna6_counts.tex
+9f9cde19f12521ff2eece2bd9f1e441078ef5fecf07f494bfba9cdbed6e63699  prompt_development/tables/p08_luna6_coverage.csv
+6f19ba6da25f1bd7e1bbb26d2c135671cffc8c7dcc757cc09210c0cfcc8163fe  prompt_development/tables/p08_luna6_coverage.html
+b076f066124150cfc41cbcd54fab5f842e6b9eb8023441321801f84817e951a8  release_provenance.json
 2a7f650d3227736b9e98c66b05d9827419f39018b08a0ea1f6b8c145c43c9c41  tables/d01_corpus.csv
 9d2cf35b01e05ae461934345caf8188788a08b9434d13340a810dd48a304f878  tables/d01_corpus.html
 e107bacf986ab3debc793be5f83582f22095743555635a522dd5d19d40540b80  tables/d02_endpoint.csv

paper/index.html

@@ -11,7 +11,7 @@ code{overflow-wrap:anywhere}caption{text-align:left;font-weight:bold;padding:.5r
 <p><a href='data/episodes.csv'>Portable episode CSV</a> | <a href='manifest.json'>Machine-readable catalog</a></p>
 <p>All three stages exported from the same source CSV.</p><p class='warning'>Node comparisons use ordinal positions, not semantic BT equivalence. Only shared positions are scored; extra positions are not false incomplete. Whole-mission agreement excludes unequal node counts. Original IDs and binding provenance are retained; some Qwen3 episode bindings use explicitly inferred frame-zero response-prefix order with recorded full BT counts. Original Qwen3-VL uses a historical source-clock convention. Reconcile clocks before publishing cross-auditor timing claims; use current timing tables rather than the earlier manuscript&#x27;s exact-ID examples. Luna 6 uses ten separate images per full window (3.6 s first-to-last); historical Qwen3.8 uses eight-frame contact sheets (2.8 s first-to-last). These are not controlled input-matched comparisons, and local/cloud latency is not a controlled hardware benchmark. Luna timing is calibrated from saved frame indices and measured video duration, rounded as in the replay writer; original workbook times remain in stored_video_time fields. Calibration uses the existing duration-ratio convention, not new visual event annotations. Simulator success, human planned-path completion, and auditor judgments remain separate signals.</p>
 <p><small>Generated 2026-10-25T05:05:20.333990+00:00. CSV SHA256: <code>c9c4a01691083d984b752349f1c0a307b828aa6b7e8c2229d469aa87bc1de040</code></small></p>
-<p><a href='prompt_development/index.html'>Separate Qwen3.8 v1--v5 prompt-development supplement</a> (not primary evaluation performance).</p>
+<p><a href='prompt_development/index.html'>Separate Qwen3.8 and Luna 6 v1--v5 prompt-development supplement</a> (not primary evaluation performance).</p>
 <h2>01. Corpus and Stage 1</h2>
 <article id='t01_corpus'><h3>Evaluation corpus and episode-specific BT structure</h3><p>Episode counts, simulator outcomes, required-node totals, and BT-size summaries. Simulator rates use known outcomes.</p><p><a href='tables/t01_corpus.csv'>CSV</a> | <a href='tables/t01_corpus.html'>HTML</a> | <a href='tables/t01_corpus.tex'>TEX</a></p></article>
 <article id='f01_corpus'><h3>Corpus and BT structure</h3><p>Evaluation and reviewed episode counts, plus within-task distributions of required BT nodes.</p><p><a href='figures/f01_corpus.pdf'>PDF</a> | <a href='figures/f01_corpus.png'>PNG</a> | <a href='figures/f01_corpus_side.pdf'>SIDE PDF</a> | <a href='figures/f01_corpus_side.png'>SIDE PNG</a></p><img src='figures/f01_corpus.png' alt='Corpus and BT structure'><details><summary>Side-by-side version</summary><img src='figures/f01_corpus_side.png' alt='Side-by-side: Corpus and BT structure'></details></article>

paper/prompt_development/figures/p01_qwen38.pdf

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:de69766eaa16c0c111bc49d97ae04f984dcefcb3ff032d26d9c9e51b384a0f94
-size 15476
+oid sha256:8e27f0832e09535b9abc606aa4ec605f434e18bf170fd4f46c4c11a1882cf9ec
+size 15511

paper/prompt_development/figures/p01_qwen38_side.pdf

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:878dcc258318382d8950b1224296bb9a17a86bc8900da535d33a499eb17455ca
-size 15173
+oid sha256:aaa7ac11c0717d38a08d7c4086e54b0eeb5a93bbbd77c48c603233850668750b
+size 15206

paper/prompt_development/figures/p05_luna6.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fcd823655804af4ee76b42f4c037ef2da18e1a31b3eefece14b7d58a1e676e8c
+size 14847

paper/prompt_development/figures/p05_luna6_side.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cfb56f577104b201d57f7ba66545f52212dec998047c873743cd7ceebeeb989e
+size 14461

paper/prompt_development/index.html

@@ -1,5 +1,5 @@
-<!doctype html><html lang='en'><meta charset='utf-8'><title>Qwen3.8 prompt development</title><style>body{font:17px/1.5 system-ui,sans-serif;max-width:1150px;margin:2rem auto;padding:0 1rem;color:#17212b}
+<!doctype html><html lang='en'><meta charset='utf-8'><title>Prompt development: Qwen3.8 and Luna 6</title><style>body{font:17px/1.5 system-ui,sans-serif;max-width:1150px;margin:2rem auto;padding:0 1rem;color:#17212b}
 h1,h2,h3{line-height:1.2}a{color:#005b96}figure,article{margin:1.5rem 0;padding:1rem;border:1px solid #b8c3cc;border-radius:6px}
 img{display:block;width:100%;height:auto}table{border-collapse:collapse;font-size:15px}th,td{border-bottom:1px solid #ccd4db;padding:.5rem;text-align:left;vertical-align:top}
 th{background:#edf3f7;position:sticky;top:0}.scroll{overflow:auto;max-height:75vh}.warning{border-left:5px solid #b45309;padding:1rem;background:#fff5df}
-code{overflow-wrap:anywhere}caption{text-align:left;font-weight:bold;padding:.5rem}small{color:#435566}</style><h1>Qwen3.8 v1--v5: prompt development</h1><p class='warning'>Episode-1 prompt development only, not held-out evaluation. Each revision&#x27;s Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. The v1 budget was 256 tokens for 22 episodes and 1024 for two successful retries. V4 was selected for evaluation, where it is reported simply as Qwen3.8.</p><p><a href='figures/p01_qwen38.pdf'>PDF</a> | <a href='figures/p01_qwen38.png'>PNG</a> | <a href='figures/p01_qwen38_side.pdf'>SIDE PDF</a> | <a href='figures/p01_qwen38_side.png'>SIDE PNG</a></p><img src='figures/p01_qwen38_side.png' alt='Five Qwen3.8 development revisions'><p><a href='tables/p02_detail.html'>Prompt-development detail</a> | <a href='tables/p03_counts.html'>Prompt-development counts</a> | <a href='tables/p04_coverage.html'>Prompt-development coverage</a> | <a href='data/episodes.csv'>Full development datasheet</a> | <a href='manifest.json'>Development provenance</a></p></html>
+code{overflow-wrap:anywhere}caption{text-align:left;font-weight:bold;padding:.5rem}small{color:#435566}</style><h1>Prompt development: Qwen3.8 and Luna 6</h1><p class='warning'>Episode-1 prompt development only, not held-out evaluation. Each revision&#x27;s Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. </p><p>Use vector PDFs at 7.2-inch journal width. Stacked masters have text at least 9.5 pt; 14.4-inch side-by-side masters have text at least 19 pt (9.5 pt at 7.2-inch print width). Do not shrink to a single column. Exact counts and eligibility are retained in CSV, HTML and LaTeX tables.</p><h2>Qwen3.8 v1-v5</h2><p>Episode-1 prompt development only, not held-out evaluation. Each revision&#x27;s Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. Qwen3.8 v1 used 256 output tokens for 22 episodes and 1024 for two successful retries. V4 was selected for evaluation. Historical Qwen3.8 uses eight-frame contact sheets.</p><p><a href='figures/p01_qwen38.pdf'>PDF</a> | <a href='figures/p01_qwen38.png'>PNG</a> | <a href='figures/p01_qwen38_side.pdf'>SIDE PDF</a> | <a href='figures/p01_qwen38_side.png'>SIDE PNG</a></p><img src='figures/p01_qwen38_side.png' alt='Qwen3.8: five prompt revisions'><p><a href='tables/p02_detail.html'>Qwen3.8 development: episode-node decisions</a> | <a href='tables/p03_counts.html'>Qwen3.8 development: sequential completion agreement</a> | <a href='tables/p04_coverage.html'>Qwen3.8 development: eligibility and coverage</a></p><h2>Luna 6 v1-v5</h2><p>Episode-1 prompt development only, not held-out evaluation. Each revision&#x27;s Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. Luna 6 v5 was selected for evaluation. Luna uses ten separate images per full window; these input conditions differ from Qwen3.8 and are not a controlled input-matched comparison. These tables compare completion decisions, not source-clock timing accuracy.</p><p><a href='figures/p05_luna6.pdf'>PDF</a> | <a href='figures/p05_luna6.png'>PNG</a> | <a href='figures/p05_luna6_side.pdf'>SIDE PDF</a> | <a href='figures/p05_luna6_side.png'>SIDE PNG</a></p><img src='figures/p05_luna6_side.png' alt='Luna 6: five prompt revisions'><p><a href='tables/p06_luna6_detail.html'>Luna 6 development: episode-node decisions</a> | <a href='tables/p07_luna6_counts.html'>Luna 6 development: sequential completion agreement</a> | <a href='tables/p08_luna6_coverage.html'>Luna 6 development: eligibility and coverage</a></p><p><a href='data/episodes.csv'>Full development datasheet</a> | <a href='manifest.json'>Development provenance</a></p></html>

paper/prompt_development/manifest.json

@@ -6,13 +6,17 @@
     "episodes": 24,
     "original_sha256": "498f70eacf34d0698adff7a510b2344393165887a87ade39b3090069544070c9"
   },
-  "description": "Episode-1 prompt development only, not held-out evaluation. Each revision's Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. The v1 budget was 256 tokens for 22 episodes and 1024 for two successful retries. V4 was selected for evaluation, where it is reported simply as Qwen3.8.",
+  "description": "Episode-1 prompt development only, not held-out evaluation. Each revision's Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. ",
+  "families": {
+    "qwen38": "Episode-1 prompt development only, not held-out evaluation. Each revision's Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. Qwen3.8 v1 used 256 output tokens for 22 episodes and 1024 for two successful retries. V4 was selected for evaluation. Historical Qwen3.8 uses eight-frame contact sheets.",
+    "luna6": "Episode-1 prompt development only, not held-out evaluation. Each revision's Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. Luna 6 v5 was selected for evaluation. Luna uses ten separate images per full window; these input conditions differ from Qwen3.8 and are not a controlled input-matched comparison. These tables compare completion decisions, not source-clock timing accuracy."
+  },
   "entries": [
     {
       "stage": "development",
       "label": "p01_qwen38",
-      "title": "Qwen3.8 prompt development: v1--v5",
-      "description": "Episode-1 prompt development only, not held-out evaluation. Each revision's Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. The v1 budget was 256 tokens for 22 episodes and 1024 for two successful retries. V4 was selected for evaluation, where it is reported simply as Qwen3.8.",
+      "title": "Qwen3.8 prompt development (v1-v5)",
+      "description": "Episode-1 prompt development only, not held-out evaluation. Each revision's Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. Qwen3.8 v1 used 256 output tokens for 22 episodes and 1024 for two successful retries. V4 was selected for evaluation. Historical Qwen3.8 uses eight-frame contact sheets.",
       "kind": "figure",
       "width_in": 7.2,
       "minimum_font_pt": 9.5,
@@ -31,8 +35,8 @@
     {
       "stage": "development",
       "label": "p02_detail",
-      "title": "Prompt-development detail",
-      "description": "Episode-1 prompt development only, not held-out evaluation. Each revision's Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. The v1 budget was 256 tokens for 22 episodes and 1024 for two successful retries. V4 was selected for evaluation, where it is reported simply as Qwen3.8.",
+      "title": "Qwen3.8 development: episode-node decisions",
+      "description": "Episode-1 prompt development only, not held-out evaluation. Each revision's Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. Qwen3.8 v1 used 256 output tokens for 22 episodes and 1024 for two successful retries. V4 was selected for evaluation. Historical Qwen3.8 uses eight-frame contact sheets.",
       "kind": "supporting table",
       "rows": 360,
       "files": {
@@ -43,26 +47,84 @@
     {
       "stage": "development",
       "label": "p03_counts",
-      "title": "Prompt-development counts",
-      "description": "Episode-1 prompt development only, not held-out evaluation. Each revision's Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. The v1 budget was 256 tokens for 22 episodes and 1024 for two successful retries. V4 was selected for evaluation, where it is reported simply as Qwen3.8.",
-      "kind": "supporting table",
+      "title": "Qwen3.8 development: sequential completion agreement",
+      "description": "Episode-1 prompt development only, not held-out evaluation. Each revision's Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. Qwen3.8 v1 used 256 output tokens for 22 episodes and 1024 for two successful retries. V4 was selected for evaluation. Historical Qwen3.8 uses eight-frame contact sheets.",
+      "kind": "manuscript table",
       "rows": 15,
       "files": {
         "csv": "tables/p03_counts.csv",
-        "html": "tables/p03_counts.html"
+        "html": "tables/p03_counts.html",
+        "tex": "tables/p03_counts.tex"
       }
     },
     {
       "stage": "development",
       "label": "p04_coverage",
-      "title": "Prompt-development coverage",
-      "description": "Episode-1 prompt development only, not held-out evaluation. Each revision's Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. The v1 budget was 256 tokens for 22 episodes and 1024 for two successful retries. V4 was selected for evaluation, where it is reported simply as Qwen3.8.",
+      "title": "Qwen3.8 development: eligibility and coverage",
+      "description": "Episode-1 prompt development only, not held-out evaluation. Each revision's Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. Qwen3.8 v1 used 256 output tokens for 22 episodes and 1024 for two successful retries. V4 was selected for evaluation. Historical Qwen3.8 uses eight-frame contact sheets.",
       "kind": "supporting table",
       "rows": 15,
       "files": {
         "csv": "tables/p04_coverage.csv",
         "html": "tables/p04_coverage.html"
       }
+    },
+    {
+      "stage": "development",
+      "label": "p05_luna6",
+      "title": "Luna 6 prompt development (v1-v5)",
+      "description": "Episode-1 prompt development only, not held-out evaluation. Each revision's Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. Luna 6 v5 was selected for evaluation. Luna uses ten separate images per full window; these input conditions differ from Qwen3.8 and are not a controlled input-matched comparison. These tables compare completion decisions, not source-clock timing accuracy.",
+      "kind": "figure",
+      "width_in": 7.2,
+      "minimum_font_pt": 9.5,
+      "side_by_side": {
+        "width_in": 14.4,
+        "minimum_font_pt": 19,
+        "intended_print_width_in": 7.2
+      },
+      "files": {
+        "pdf": "figures/p05_luna6.pdf",
+        "png": "figures/p05_luna6.png",
+        "side_pdf": "figures/p05_luna6_side.pdf",
+        "side_png": "figures/p05_luna6_side.png"
+      }
+    },
+    {
+      "stage": "development",
+      "label": "p06_luna6_detail",
+      "title": "Luna 6 development: episode-node decisions",
+      "description": "Episode-1 prompt development only, not held-out evaluation. Each revision's Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. Luna 6 v5 was selected for evaluation. Luna uses ten separate images per full window; these input conditions differ from Qwen3.8 and are not a controlled input-matched comparison. These tables compare completion decisions, not source-clock timing accuracy.",
+      "kind": "supporting table",
+      "rows": 360,
+      "files": {
+        "csv": "tables/p06_luna6_detail.csv",
+        "html": "tables/p06_luna6_detail.html"
+      }
+    },
+    {
+      "stage": "development",
+      "label": "p07_luna6_counts",
+      "title": "Luna 6 development: sequential completion agreement",
+      "description": "Episode-1 prompt development only, not held-out evaluation. Each revision's Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. Luna 6 v5 was selected for evaluation. Luna uses ten separate images per full window; these input conditions differ from Qwen3.8 and are not a controlled input-matched comparison. These tables compare completion decisions, not source-clock timing accuracy.",
+      "kind": "manuscript table",
+      "rows": 15,
+      "files": {
+        "csv": "tables/p07_luna6_counts.csv",
+        "html": "tables/p07_luna6_counts.html",
+        "tex": "tables/p07_luna6_counts.tex"
+      }
+    },
+    {
+      "stage": "development",
+      "label": "p08_luna6_coverage",
+      "title": "Luna 6 development: eligibility and coverage",
+      "description": "Episode-1 prompt development only, not held-out evaluation. Each revision's Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. Luna 6 v5 was selected for evaluation. Luna uses ten separate images per full window; these input conditions differ from Qwen3.8 and are not a controlled input-matched comparison. These tables compare completion decisions, not source-clock timing accuracy.",
+      "kind": "supporting table",
+      "rows": 15,
+      "files": {
+        "csv": "tables/p08_luna6_coverage.csv",
+        "html": "tables/p08_luna6_coverage.html"
+      }
     }
   ]
 }

paper/prompt_development/tables/p02_detail.html

@@ -1,8 +1,8 @@
-<!doctype html><html lang='en'><meta charset='utf-8'><title>Prompt-development detail</title><style>body{font:17px/1.5 system-ui,sans-serif;max-width:1150px;margin:2rem auto;padding:0 1rem;color:#17212b}
+<!doctype html><html lang='en'><meta charset='utf-8'><title>Qwen3.8 development: episode-node decisions</title><style>body{font:17px/1.5 system-ui,sans-serif;max-width:1150px;margin:2rem auto;padding:0 1rem;color:#17212b}
 h1,h2,h3{line-height:1.2}a{color:#005b96}figure,article{margin:1.5rem 0;padding:1rem;border:1px solid #b8c3cc;border-radius:6px}
 img{display:block;width:100%;height:auto}table{border-collapse:collapse;font-size:15px}th,td{border-bottom:1px solid #ccd4db;padding:.5rem;text-align:left;vertical-align:top}
 th{background:#edf3f7;position:sticky;top:0}.scroll{overflow:auto;max-height:75vh}.warning{border-left:5px solid #b45309;padding:1rem;background:#fff5df}
-code{overflow-wrap:anywhere}caption{text-align:left;font-weight:bold;padding:.5rem}small{color:#435566}</style><h1>Prompt-development detail</h1><p>Episode-1 prompt development only, not held-out evaluation. Each revision&#x27;s Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. The v1 budget was 256 tokens for 22 episodes and 1024 for two successful retries. V4 was selected for evaluation, where it is reported simply as Qwen3.8.</p><p>360 rows. CSV retains full numeric precision. <a href='../index.html'>Bundle index</a> | <a href='p02_detail.csv'>CSV</a></p><div class='scroll'><table class="dataframe">
+code{overflow-wrap:anywhere}caption{text-align:left;font-weight:bold;padding:.5rem}small{color:#435566}</style><h1>Qwen3.8 development: episode-node decisions</h1><p>Episode-1 prompt development only, not held-out evaluation. Each revision&#x27;s Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. Qwen3.8 v1 used 256 output tokens for 22 episodes and 1024 for two successful retries. V4 was selected for evaluation. Historical Qwen3.8 uses eight-frame contact sheets.</p><p>360 rows. CSV retains full numeric precision. <a href='../index.html'>Bundle index</a> | <a href='p02_detail.csv'>CSV</a></p><div class='scroll'><table class="dataframe">
   <thead>
     <tr style="text-align: right;">
       <th>episode_dir</th>

paper/prompt_development/tables/p03_counts.html

@@ -1,17 +1,17 @@
-<!doctype html><html lang='en'><meta charset='utf-8'><title>Prompt-development counts</title><style>body{font:17px/1.5 system-ui,sans-serif;max-width:1150px;margin:2rem auto;padding:0 1rem;color:#17212b}
+<!doctype html><html lang='en'><meta charset='utf-8'><title>Qwen3.8 development: sequential completion agreement</title><style>body{font:17px/1.5 system-ui,sans-serif;max-width:1150px;margin:2rem auto;padding:0 1rem;color:#17212b}
 h1,h2,h3{line-height:1.2}a{color:#005b96}figure,article{margin:1.5rem 0;padding:1rem;border:1px solid #b8c3cc;border-radius:6px}
 img{display:block;width:100%;height:auto}table{border-collapse:collapse;font-size:15px}th,td{border-bottom:1px solid #ccd4db;padding:.5rem;text-align:left;vertical-align:top}
 th{background:#edf3f7;position:sticky;top:0}.scroll{overflow:auto;max-height:75vh}.warning{border-left:5px solid #b45309;padding:1rem;background:#fff5df}
-code{overflow-wrap:anywhere}caption{text-align:left;font-weight:bold;padding:.5rem}small{color:#435566}</style><h1>Prompt-development counts</h1><p>Episode-1 prompt development only, not held-out evaluation. Each revision&#x27;s Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. The v1 budget was 256 tokens for 22 episodes and 1024 for two successful retries. V4 was selected for evaluation, where it is reported simply as Qwen3.8.</p><p>15 rows. CSV retains full numeric precision. <a href='../index.html'>Bundle index</a> | <a href='p03_counts.csv'>CSV</a></p><div class='scroll'><table class="dataframe">
+code{overflow-wrap:anywhere}caption{text-align:left;font-weight:bold;padding:.5rem}small{color:#435566}</style><h1>Qwen3.8 development: sequential completion agreement</h1><p>Episode-1 prompt development only, not held-out evaluation. Each revision&#x27;s Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. Qwen3.8 v1 used 256 output tokens for 22 episodes and 1024 for two successful retries. V4 was selected for evaluation. Historical Qwen3.8 uses eight-frame contact sheets.</p><p>15 rows. CSV retains full numeric precision. <a href='../index.html'>Bundle index</a> | <a href='p03_counts.csv'>CSV</a></p><div class='scroll'><table class="dataframe">
   <thead>
     <tr style="text-align: right;">
-      <th>model</th>
-      <th>node_slot</th>
+      <th>Prompt revision</th>
+      <th>Node</th>
       <th>Correct complete</th>
       <th>Correct incomplete</th>
       <th>False complete</th>
       <th>False incomplete</th>
-      <th>eligible_count</th>
+      <th>Eligible pairs</th>
     </tr>
   </thead>
   <tbody>

paper/prompt_development/tables/p03_counts.tex

@@ -0,0 +1,28 @@
+\begin{table*}[t]
+\centering
+\small
+\setlength{\tabcolsep}{3pt}
+\caption{Qwen3.8 development: sequential completion agreement. Episode-1 prompt development only, not held-out evaluation. Each revision's Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. Qwen3.8 v1 used 256 output tokens for 22 episodes and 1024 for two successful retries. V4 was selected for evaluation. Historical Qwen3.8 uses eight-frame contact sheets.}
+\label{tab:vibtarve_p03_counts}
+\begin{tabular}{lrrrrrr}
+\toprule
+\shortstack{Prompt\\revision} & Node & \shortstack{Correct\\complete} & \shortstack{Correct\\incomplete} & \shortstack{False\\complete} & \shortstack{False\\incomplete} & \shortstack{Eligible\\pairs} \\
+\midrule
+Qwen3.8 v1 & 1 & 15 & 0 & 9 & 0 & 24 \\
+Qwen3.8 v1 & 2 & 14 & 2 & 7 & 1 & 24 \\
+Qwen3.8 v1 & 3 & 1 & 1 & 9 & 0 & 11 \\
+Qwen3.8 v2 & 1 & 13 & 4 & 5 & 2 & 24 \\
+Qwen3.8 v2 & 2 & 9 & 3 & 2 & 4 & 18 \\
+Qwen3.8 v2 & 3 & 0 & 4 & 1 & 1 & 6 \\
+Qwen3.8 v3 & 1 & 12 & 8 & 1 & 3 & 24 \\
+Qwen3.8 v3 & 2 & 8 & 1 & 0 & 4 & 13 \\
+Qwen3.8 v3 & 3 & 0 & 3 & 1 & 0 & 4 \\
+Qwen3.8 v4 & 1 & 15 & 5 & 4 & 0 & 24 \\
+Qwen3.8 v4 & 2 & 12 & 2 & 2 & 3 & 19 \\
+Qwen3.8 v4 & 3 & 0 & 6 & 1 & 1 & 8 \\
+Qwen3.8 v5 & 1 & 15 & 3 & 6 & 0 & 24 \\
+Qwen3.8 v5 & 2 & 12 & 5 & 1 & 3 & 21 \\
+Qwen3.8 v5 & 3 & 0 & 5 & 1 & 1 & 7 \\
+\bottomrule
+\end{tabular}
+\end{table*}

paper/prompt_development/tables/p04_coverage.html

@@ -1,8 +1,8 @@
-<!doctype html><html lang='en'><meta charset='utf-8'><title>Prompt-development coverage</title><style>body{font:17px/1.5 system-ui,sans-serif;max-width:1150px;margin:2rem auto;padding:0 1rem;color:#17212b}
+<!doctype html><html lang='en'><meta charset='utf-8'><title>Qwen3.8 development: eligibility and coverage</title><style>body{font:17px/1.5 system-ui,sans-serif;max-width:1150px;margin:2rem auto;padding:0 1rem;color:#17212b}
 h1,h2,h3{line-height:1.2}a{color:#005b96}figure,article{margin:1.5rem 0;padding:1rem;border:1px solid #b8c3cc;border-radius:6px}
 img{display:block;width:100%;height:auto}table{border-collapse:collapse;font-size:15px}th,td{border-bottom:1px solid #ccd4db;padding:.5rem;text-align:left;vertical-align:top}
 th{background:#edf3f7;position:sticky;top:0}.scroll{overflow:auto;max-height:75vh}.warning{border-left:5px solid #b45309;padding:1rem;background:#fff5df}
-code{overflow-wrap:anywhere}caption{text-align:left;font-weight:bold;padding:.5rem}small{color:#435566}</style><h1>Prompt-development coverage</h1><p>Episode-1 prompt development only, not held-out evaluation. Each revision&#x27;s Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. The v1 budget was 256 tokens for 22 episodes and 1024 for two successful retries. V4 was selected for evaluation, where it is reported simply as Qwen3.8.</p><p>15 rows. CSV retains full numeric precision. <a href='../index.html'>Bundle index</a> | <a href='p04_coverage.csv'>CSV</a></p><div class='scroll'><table class="dataframe">
+code{overflow-wrap:anywhere}caption{text-align:left;font-weight:bold;padding:.5rem}small{color:#435566}</style><h1>Qwen3.8 development: eligibility and coverage</h1><p>Episode-1 prompt development only, not held-out evaluation. Each revision&#x27;s Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. Qwen3.8 v1 used 256 output tokens for 22 episodes and 1024 for two successful retries. V4 was selected for evaluation. Historical Qwen3.8 uses eight-frame contact sheets.</p><p>15 rows. CSV retains full numeric precision. <a href='../index.html'>Bundle index</a> | <a href='p04_coverage.csv'>CSV</a></p><div class='scroll'><table class="dataframe">
   <thead>
     <tr style="text-align: right;">
       <th>model</th>

paper/prompt_development/tables/p06_luna6_detail.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4042e07a4b6ecd82c17dd3acf191af8ee1aa81fd6b6572aa6ce40933f59f320e
+size 96535

paper/prompt_development/tables/p07_luna6_counts.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e1779e4e5f36cdf977bdab437a90f880a6b090e37761f1990f59c9d2fe31ed16
+size 440

paper/prompt_development/tables/p07_luna6_counts.html

@@ -0,0 +1,154 @@
+<!doctype html><html lang='en'><meta charset='utf-8'><title>Luna 6 development: sequential completion agreement</title><style>body{font:17px/1.5 system-ui,sans-serif;max-width:1150px;margin:2rem auto;padding:0 1rem;color:#17212b}
+h1,h2,h3{line-height:1.2}a{color:#005b96}figure,article{margin:1.5rem 0;padding:1rem;border:1px solid #b8c3cc;border-radius:6px}
+img{display:block;width:100%;height:auto}table{border-collapse:collapse;font-size:15px}th,td{border-bottom:1px solid #ccd4db;padding:.5rem;text-align:left;vertical-align:top}
+th{background:#edf3f7;position:sticky;top:0}.scroll{overflow:auto;max-height:75vh}.warning{border-left:5px solid #b45309;padding:1rem;background:#fff5df}
+code{overflow-wrap:anywhere}caption{text-align:left;font-weight:bold;padding:.5rem}small{color:#435566}</style><h1>Luna 6 development: sequential completion agreement</h1><p>Episode-1 prompt development only, not held-out evaluation. Each revision&#x27;s Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. Luna 6 v5 was selected for evaluation. Luna uses ten separate images per full window; these input conditions differ from Qwen3.8 and are not a controlled input-matched comparison. These tables compare completion decisions, not source-clock timing accuracy.</p><p>15 rows. CSV retains full numeric precision. <a href='../index.html'>Bundle index</a> | <a href='p07_luna6_counts.csv'>CSV</a></p><div class='scroll'><table class="dataframe">
+  <thead>
+    <tr style="text-align: right;">
+      <th>Prompt revision</th>
+      <th>Node</th>
+      <th>Correct complete</th>
+      <th>Correct incomplete</th>
+      <th>False complete</th>
+      <th>False incomplete</th>
+      <th>Eligible pairs</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td>Luna 6 v1</td>
+      <td>1</td>
+      <td>2</td>
+      <td>9</td>
+      <td>0</td>
+      <td>13</td>
+      <td>24</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v1</td>
+      <td>2</td>
+      <td>2</td>
+      <td>0</td>
+      <td>0</td>
+      <td>0</td>
+      <td>2</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v1</td>
+      <td>3</td>
+      <td>0</td>
+      <td>1</td>
+      <td>0</td>
+      <td>0</td>
+      <td>1</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v2</td>
+      <td>1</td>
+      <td>9</td>
+      <td>7</td>
+      <td>2</td>
+      <td>6</td>
+      <td>24</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v2</td>
+      <td>2</td>
+      <td>1</td>
+      <td>2</td>
+      <td>0</td>
+      <td>8</td>
+      <td>11</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v2</td>
+      <td>3</td>
+      <td>0</td>
+      <td>0</td>
+      <td>0</td>
+      <td>0</td>
+      <td>0</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v3</td>
+      <td>1</td>
+      <td>8</td>
+      <td>8</td>
+      <td>1</td>
+      <td>7</td>
+      <td>24</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v3</td>
+      <td>2</td>
+      <td>4</td>
+      <td>1</td>
+      <td>0</td>
+      <td>4</td>
+      <td>9</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v3</td>
+      <td>3</td>
+      <td>0</td>
+      <td>2</td>
+      <td>1</td>
+      <td>0</td>
+      <td>3</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v4</td>
+      <td>1</td>
+      <td>13</td>
+      <td>7</td>
+      <td>2</td>
+      <td>2</td>
+      <td>24</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v4</td>
+      <td>2</td>
+      <td>5</td>
+      <td>2</td>
+      <td>0</td>
+      <td>8</td>
+      <td>15</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v4</td>
+      <td>3</td>
+      <td>0</td>
+      <td>2</td>
+      <td>1</td>
+      <td>0</td>
+      <td>3</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v5</td>
+      <td>1</td>
+      <td>12</td>
+      <td>6</td>
+      <td>3</td>
+      <td>3</td>
+      <td>24</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v5</td>
+      <td>2</td>
+      <td>7</td>
+      <td>3</td>
+      <td>0</td>
+      <td>5</td>
+      <td>15</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v5</td>
+      <td>3</td>
+      <td>0</td>
+      <td>3</td>
+      <td>1</td>
+      <td>0</td>
+      <td>4</td>
+    </tr>
+  </tbody>
+</table></div></html>

paper/prompt_development/tables/p07_luna6_counts.tex

@@ -0,0 +1,28 @@
+\begin{table*}[t]
+\centering
+\small
+\setlength{\tabcolsep}{3pt}
+\caption{Luna 6 development: sequential completion agreement. Episode-1 prompt development only, not held-out evaluation. Each revision's Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. Luna 6 v5 was selected for evaluation. Luna uses ten separate images per full window; these input conditions differ from Qwen3.8 and are not a controlled input-matched comparison. These tables compare completion decisions, not source-clock timing accuracy.}
+\label{tab:vibtarve_p07_luna6_counts}
+\begin{tabular}{lrrrrrr}
+\toprule
+\shortstack{Prompt\\revision} & Node & \shortstack{Correct\\complete} & \shortstack{Correct\\incomplete} & \shortstack{False\\complete} & \shortstack{False\\incomplete} & \shortstack{Eligible\\pairs} \\
+\midrule
+Luna 6 v1 & 1 & 2 & 9 & 0 & 13 & 24 \\
+Luna 6 v1 & 2 & 2 & 0 & 0 & 0 & 2 \\
+Luna 6 v1 & 3 & 0 & 1 & 0 & 0 & 1 \\
+Luna 6 v2 & 1 & 9 & 7 & 2 & 6 & 24 \\
+Luna 6 v2 & 2 & 1 & 2 & 0 & 8 & 11 \\
+Luna 6 v2 & 3 & 0 & 0 & 0 & 0 & 0 \\
+Luna 6 v3 & 1 & 8 & 8 & 1 & 7 & 24 \\
+Luna 6 v3 & 2 & 4 & 1 & 0 & 4 & 9 \\
+Luna 6 v3 & 3 & 0 & 2 & 1 & 0 & 3 \\
+Luna 6 v4 & 1 & 13 & 7 & 2 & 2 & 24 \\
+Luna 6 v4 & 2 & 5 & 2 & 0 & 8 & 15 \\
+Luna 6 v4 & 3 & 0 & 2 & 1 & 0 & 3 \\
+Luna 6 v5 & 1 & 12 & 6 & 3 & 3 & 24 \\
+Luna 6 v5 & 2 & 7 & 3 & 0 & 5 & 15 \\
+Luna 6 v5 & 3 & 0 & 3 & 1 & 0 & 4 \\
+\bottomrule
+\end{tabular}
+\end{table*}

paper/prompt_development/tables/p08_luna6_coverage.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f9cde19f12521ff2eece2bd9f1e441078ef5fecf07f494bfba9cdbed6e63699
+size 513

paper/prompt_development/tables/p08_luna6_coverage.html

@@ -0,0 +1,170 @@
+<!doctype html><html lang='en'><meta charset='utf-8'><title>Luna 6 development: eligibility and coverage</title><style>body{font:17px/1.5 system-ui,sans-serif;max-width:1150px;margin:2rem auto;padding:0 1rem;color:#17212b}
+h1,h2,h3{line-height:1.2}a{color:#005b96}figure,article{margin:1.5rem 0;padding:1rem;border:1px solid #b8c3cc;border-radius:6px}
+img{display:block;width:100%;height:auto}table{border-collapse:collapse;font-size:15px}th,td{border-bottom:1px solid #ccd4db;padding:.5rem;text-align:left;vertical-align:top}
+th{background:#edf3f7;position:sticky;top:0}.scroll{overflow:auto;max-height:75vh}.warning{border-left:5px solid #b45309;padding:1rem;background:#fff5df}
+code{overflow-wrap:anywhere}caption{text-align:left;font-weight:bold;padding:.5rem}small{color:#435566}</style><h1>Luna 6 development: eligibility and coverage</h1><p>Episode-1 prompt development only, not held-out evaluation. Each revision&#x27;s Node 2/3 cohort requires its preceding completion decisions; blank human times mean incomplete in this development review. Different revisions therefore have different later-node denominators. Luna 6 v5 was selected for evaluation. Luna uses ten separate images per full window; these input conditions differ from Qwen3.8 and are not a controlled input-matched comparison. These tables compare completion decisions, not source-clock timing accuracy.</p><p>15 rows. CSV retains full numeric precision. <a href='../index.html'>Bundle index</a> | <a href='p08_luna6_coverage.csv'>CSV</a></p><div class='scroll'><table class="dataframe">
+  <thead>
+    <tr style="text-align: right;">
+      <th>model</th>
+      <th>node_slot</th>
+      <th>episode_count</th>
+      <th>eligible_count</th>
+      <th>missing_model</th>
+      <th>missing_human_review</th>
+      <th>missing_node</th>
+      <th>previous_node_incomplete</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <td>Luna 6 v1</td>
+      <td>1</td>
+      <td>24</td>
+      <td>24</td>
+      <td>0</td>
+      <td>0</td>
+      <td>0</td>
+      <td>0</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v1</td>
+      <td>2</td>
+      <td>24</td>
+      <td>2</td>
+      <td>0</td>
+      <td>0</td>
+      <td>0</td>
+      <td>22</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v1</td>
+      <td>3</td>
+      <td>24</td>
+      <td>1</td>
+      <td>0</td>
+      <td>0</td>
+      <td>11</td>
+      <td>12</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v2</td>
+      <td>1</td>
+      <td>24</td>
+      <td>24</td>
+      <td>0</td>
+      <td>0</td>
+      <td>0</td>
+      <td>0</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v2</td>
+      <td>2</td>
+      <td>24</td>
+      <td>11</td>
+      <td>0</td>
+      <td>0</td>
+      <td>0</td>
+      <td>13</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v2</td>
+      <td>3</td>
+      <td>24</td>
+      <td>0</td>
+      <td>0</td>
+      <td>0</td>
+      <td>11</td>
+      <td>13</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v3</td>
+      <td>1</td>
+      <td>24</td>
+      <td>24</td>
+      <td>0</td>
+      <td>0</td>
+      <td>0</td>
+      <td>0</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v3</td>
+      <td>2</td>
+      <td>24</td>
+      <td>9</td>
+      <td>0</td>
+      <td>0</td>
+      <td>0</td>
+      <td>15</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v3</td>
+      <td>3</td>
+      <td>24</td>
+      <td>3</td>
+      <td>0</td>
+      <td>0</td>
+      <td>11</td>
+      <td>10</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v4</td>
+      <td>1</td>
+      <td>24</td>
+      <td>24</td>
+      <td>0</td>
+      <td>0</td>
+      <td>0</td>
+      <td>0</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v4</td>
+      <td>2</td>
+      <td>24</td>
+      <td>15</td>
+      <td>0</td>
+      <td>0</td>
+      <td>0</td>
+      <td>9</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v4</td>
+      <td>3</td>
+      <td>24</td>
+      <td>3</td>
+      <td>0</td>
+      <td>0</td>
+      <td>11</td>
+      <td>10</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v5</td>
+      <td>1</td>
+      <td>24</td>
+      <td>24</td>
+      <td>0</td>
+      <td>0</td>
+      <td>0</td>
+      <td>0</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v5</td>
+      <td>2</td>
+      <td>24</td>
+      <td>15</td>
+      <td>0</td>
+      <td>0</td>
+      <td>0</td>
+      <td>9</td>
+    </tr>
+    <tr>
+      <td>Luna 6 v5</td>
+      <td>3</td>
+      <td>24</td>
+      <td>4</td>
+      <td>0</td>
+      <td>0</td>
+      <td>11</td>
+      <td>9</td>
+    </tr>
+  </tbody>
+</table></div></html>

paper/release_provenance.json

@@ -6,13 +6,13 @@
     "dataset_path": 8706,
     "summary_path": 1
   },
-  "source_files_verified_unchanged": 132,
+  "source_files_verified_unchanged": 144,
   "files": [
     {
       "path": "README.md",
-      "sha256": "7c513659727072420c512f149ead7100f43d0c6641a557566014066dc103fba5",
-      "bytes": 3158,
-      "released_sha256": "ccb32ee42ca233a99bdbc18cb3b977af6babbf235ffece5b9a810885c2d6ce39"
+      "sha256": "23d876c1afea52bd4b443ac78b95f3aaa931e6e6e9924643450d3f1db02e4723",
+      "bytes": 3169,
+      "released_sha256": "f9fb5fa24c912521a86442d026f0aa93fe7477dc41d92746a017d6365ddb82b5"
     },
     {
       "path": "data/episodes.csv",
@@ -298,9 +298,9 @@
     },
     {
       "path": "index.html",
-      "sha256": "86f594c945e7fca19ee817d9ece189c1eb4db9609b7bcf01343c40e0e760030c",
-      "bytes": 28915,
-      "released_sha256": "a719701f31d5676fc7e907263425acc506678beea52cef68d58f041c08a527ed"
+      "sha256": "81c64c76ce051af67ca98f4c225b3f86f3a407c1ec09576656634689f6978415",
+      "bytes": 28926,
+      "released_sha256": "27abfd9b5428bd58c3a29cbb5ca9d0cbdd1da78734520b434ca73343578e83f2"
     },
     {
       "path": "manifest.json",
@@ -316,39 +316,63 @@
     },
     {
       "path": "prompt_development/figures/p01_qwen38.pdf",
-      "sha256": "de69766eaa16c0c111bc49d97ae04f984dcefcb3ff032d26d9c9e51b384a0f94",
-      "bytes": 15476,
-      "released_sha256": "de69766eaa16c0c111bc49d97ae04f984dcefcb3ff032d26d9c9e51b384a0f94"
+      "sha256": "8e27f0832e09535b9abc606aa4ec605f434e18bf170fd4f46c4c11a1882cf9ec",
+      "bytes": 15511,
+      "released_sha256": "8e27f0832e09535b9abc606aa4ec605f434e18bf170fd4f46c4c11a1882cf9ec"
     },
     {
       "path": "prompt_development/figures/p01_qwen38.png",
-      "sha256": "08c3bddc790aa3ad09a1d3ffc2054fce54501c5de13ed153905eedfc5ffba5f1",
-      "bytes": 317047,
-      "released_sha256": "08c3bddc790aa3ad09a1d3ffc2054fce54501c5de13ed153905eedfc5ffba5f1"
+      "sha256": "ebd6235b4c7a253b5e64a668d5036a70d95a4e8264f92a155edbd090647fa5be",
+      "bytes": 315895,
+      "released_sha256": "ebd6235b4c7a253b5e64a668d5036a70d95a4e8264f92a155edbd090647fa5be"
     },
     {
       "path": "prompt_development/figures/p01_qwen38_side.pdf",
-      "sha256": "878dcc258318382d8950b1224296bb9a17a86bc8900da535d33a499eb17455ca",
-      "bytes": 15173,
-      "released_sha256": "878dcc258318382d8950b1224296bb9a17a86bc8900da535d33a499eb17455ca"
+      "sha256": "aaa7ac11c0717d38a08d7c4086e54b0eeb5a93bbbd77c48c603233850668750b",
+      "bytes": 15206,
+      "released_sha256": "aaa7ac11c0717d38a08d7c4086e54b0eeb5a93bbbd77c48c603233850668750b"
     },
     {
       "path": "prompt_development/figures/p01_qwen38_side.png",
-      "sha256": "71ff520e87330759ae6c363be46dee57827e5235bd6198f703a0ac187e4a67a3",
-      "bytes": 296951,
-      "released_sha256": "71ff520e87330759ae6c363be46dee57827e5235bd6198f703a0ac187e4a67a3"
+      "sha256": "57c6fb9753b57abf0bd28c29c213c096dac423dc2e4bc5d3b97c5a562a39cdc6",
+      "bytes": 295802,
+      "released_sha256": "57c6fb9753b57abf0bd28c29c213c096dac423dc2e4bc5d3b97c5a562a39cdc6"
+    },
+    {
+      "path": "prompt_development/figures/p05_luna6.pdf",
+      "sha256": "fcd823655804af4ee76b42f4c037ef2da18e1a31b3eefece14b7d58a1e676e8c",
+      "bytes": 14847,
+      "released_sha256": "fcd823655804af4ee76b42f4c037ef2da18e1a31b3eefece14b7d58a1e676e8c"
+    },
+    {
+      "path": "prompt_development/figures/p05_luna6.png",
+      "sha256": "eb459330a943259b5ccd7180fa390135cfd89277a1910370774c93a48b4cdbbc",
+      "bytes": 266225,
+      "released_sha256": "eb459330a943259b5ccd7180fa390135cfd89277a1910370774c93a48b4cdbbc"
+    },
+    {
+      "path": "prompt_development/figures/p05_luna6_side.pdf",
+      "sha256": "cfb56f577104b201d57f7ba66545f52212dec998047c873743cd7ceebeeb989e",
+      "bytes": 14461,
+      "released_sha256": "cfb56f577104b201d57f7ba66545f52212dec998047c873743cd7ceebeeb989e"
+    },
+    {
+      "path": "prompt_development/figures/p05_luna6_side.png",
+      "sha256": "12b256aa22354524e3d722b24d251225237067fb32baf963d079403a7c8ee587",
+      "bytes": 283725,
+      "released_sha256": "12b256aa22354524e3d722b24d251225237067fb32baf963d079403a7c8ee587"
     },
     {
       "path": "prompt_development/index.html",
-      "sha256": "ea41392e51e7ff3d2698b193fe19e31ec27d7bc5c93b65ac70e74d429f9bfb1c",
-      "bytes": 1797,
-      "released_sha256": "ea41392e51e7ff3d2698b193fe19e31ec27d7bc5c93b65ac70e74d429f9bfb1c"
+      "sha256": "738ab43f39c43d151c8972f79355e5e30c62ee954b94a1c4f8b16dddee0f14b2",
+      "bytes": 3660,
+      "released_sha256": "738ab43f39c43d151c8972f79355e5e30c62ee954b94a1c4f8b16dddee0f14b2"
     },
     {
       "path": "prompt_development/manifest.json",
-      "sha256": "8f8d3d4da49bfc240d4e8cb1e7a678d4241280a509704b418a024888d114d853",
-      "bytes": 3463,
-      "released_sha256": "62620f5917310fb374e5d55fc28c2257908c217611a3ae80ba8ec4c92885c2d8"
+      "sha256": "d4bb3a12bf81d82552a4f33761d71597be11a0d3314d25eba6a6dce068b3b196",
+      "bytes": 8574,
+      "released_sha256": "0c2fbd44e78fd3d208555cd70be432292f7fc9715e46259e17a7cc26cd77f4fc"
     },
     {
       "path": "prompt_development/tables/p02_detail.csv",
@@ -358,9 +382,9 @@
     },
     {
       "path": "prompt_development/tables/p02_detail.html",
-      "sha256": "114c0c7c89b152915adc0ac45e8a4e1ac282fa88a5bc49f620b3d8f8184064b9",
-      "bytes": 193430,
-      "released_sha256": "114c0c7c89b152915adc0ac45e8a4e1ac282fa88a5bc49f620b3d8f8184064b9"
+      "sha256": "0189a47a4cc74b6380a7d2d42a6dc19ab360e4e8237614a9b2c28f99db9b52f5",
+      "bytes": 193553,
+      "released_sha256": "0189a47a4cc74b6380a7d2d42a6dc19ab360e4e8237614a9b2c28f99db9b52f5"
     },
     {
       "path": "prompt_development/tables/p03_counts.csv",
@@ -370,9 +394,15 @@
     },
     {
       "path": "prompt_development/tables/p03_counts.html",
-      "sha256": "9b76a85a28026178a9807cbaf267dff7974f463d5411f0cd9f145c72a74b4d9c",
-      "bytes": 3881,
-      "released_sha256": "9b76a85a28026178a9807cbaf267dff7974f463d5411f0cd9f145c72a74b4d9c"
+      "sha256": "e3caac8a5541af6521f96cb2409ef128b87f6617051633e2e619668d94cf633a",
+      "bytes": 4027,
+      "released_sha256": "e3caac8a5541af6521f96cb2409ef128b87f6617051633e2e619668d94cf633a"
+    },
+    {
+      "path": "prompt_development/tables/p03_counts.tex",
+      "sha256": "9cd4a401692c7ec1b514300807a09203ee51029887fb7dba0a4b70d455219307",
+      "bytes": 1474,
+      "released_sha256": "9cd4a401692c7ec1b514300807a09203ee51029887fb7dba0a4b70d455219307"
     },
     {
       "path": "prompt_development/tables/p04_coverage.csv",
@@ -382,9 +412,51 @@
     },
     {
       "path": "prompt_development/tables/p04_coverage.html",
-      "sha256": "b2ecec928c4763107de3e40b7c318a70e930bf2cedb51dc1f9d3cf66a020d4e5",
-      "bytes": 4189,
-      "released_sha256": "b2ecec928c4763107de3e40b7c318a70e930bf2cedb51dc1f9d3cf66a020d4e5"
+      "sha256": "ebde0f06dfe8c6a60694cc48415359ad94922ceee49cd727d0d5e6e0a1845b9d",
+      "bytes": 4312,
+      "released_sha256": "ebde0f06dfe8c6a60694cc48415359ad94922ceee49cd727d0d5e6e0a1845b9d"
+    },
+    {
+      "path": "prompt_development/tables/p06_luna6_detail.csv",
+      "sha256": "4042e07a4b6ecd82c17dd3acf191af8ee1aa81fd6b6572aa6ce40933f59f320e",
+      "bytes": 96535,
+      "released_sha256": "4042e07a4b6ecd82c17dd3acf191af8ee1aa81fd6b6572aa6ce40933f59f320e"
+    },
+    {
+      "path": "prompt_development/tables/p06_luna6_detail.html",
+      "sha256": "ae8941b164164093bb46ddc031ef731fbfea4cc26c3713cb4660497459592ceb",
+      "bytes": 193613,
+      "released_sha256": "ae8941b164164093bb46ddc031ef731fbfea4cc26c3713cb4660497459592ceb"
+    },
+    {
+      "path": "prompt_development/tables/p07_luna6_counts.csv",
+      "sha256": "e1779e4e5f36cdf977bdab437a90f880a6b090e37761f1990f59c9d2fe31ed16",
+      "bytes": 440,
+      "released_sha256": "e1779e4e5f36cdf977bdab437a90f880a6b090e37761f1990f59c9d2fe31ed16"
+    },
+    {
+      "path": "prompt_development/tables/p07_luna6_counts.html",
+      "sha256": "13954aea4756a20631f7b700a2d5903d246b4fce6810b256a3e5210cfab5cea8",
+      "bytes": 4094,
+      "released_sha256": "13954aea4756a20631f7b700a2d5903d246b4fce6810b256a3e5210cfab5cea8"
+    },
+    {
+      "path": "prompt_development/tables/p07_luna6_counts.tex",
+      "sha256": "d4266a083040824e15d66ff10b37cb292996fadb258d94249214f02ee697e48b",
+      "bytes": 1542,
+      "released_sha256": "d4266a083040824e15d66ff10b37cb292996fadb258d94249214f02ee697e48b"
+    },
+    {
+      "path": "prompt_development/tables/p08_luna6_coverage.csv",
+      "sha256": "9f9cde19f12521ff2eece2bd9f1e441078ef5fecf07f494bfba9cdbed6e63699",
+      "bytes": 513,
+      "released_sha256": "9f9cde19f12521ff2eece2bd9f1e441078ef5fecf07f494bfba9cdbed6e63699"
+    },
+    {
+      "path": "prompt_development/tables/p08_luna6_coverage.html",
+      "sha256": "6f19ba6da25f1bd7e1bbb26d2c135671cffc8c7dcc757cc09210c0cfcc8163fe",
+      "bytes": 4390,
+      "released_sha256": "6f19ba6da25f1bd7e1bbb26d2c135671cffc8c7dcc757cc09210c0cfcc8163fe"
     },
     {
       "path": "tables/d01_corpus.csv",

provenance/prompt_development.json

@@ -0,0 +1,45 @@
+{
+  "generated_utc": "2026-10-27T13:22:18.798766+00:00",
+  "code_commit": "42a65c902721c06c4b3b86f806b3df7f689c3d26",
+  "code_sha256": "667e30eb4315ba5a16edfe8685b9ea5d8e05eda9f334b8709f41d45fa53e6a50",
+  "previous_revision": "28b6157f7a47e3d9ea5b6cb4ee27ba0e5723742d",
+  "scope": "Paper prompt-development assets only",
+  "families": {
+    "qwen38": {
+      "revisions": [
+        1,
+        2,
+        3,
+        4,
+        5
+      ],
+      "selected": 4
+    },
+    "luna6": {
+      "revisions": [
+        1,
+        2,
+        3,
+        4,
+        5
+      ],
+      "selected": 5
+    }
+  },
+  "development_recordings": 24,
+  "paper_bundle": {
+    "directory": "paper",
+    "archive": "vibtarve_paper.zip",
+    "source_file_count": 144,
+    "file_count": 146,
+    "archive_sha256": "152cace4bb53001b3fa2601d50cd2bf0bd6c42817360a7a11285c1cb62148cd6"
+  },
+  "validation": {
+    "local_tests_passed": 122,
+    "minimum_font_pt_at_7_2_inch_width": 9.5,
+    "main_title_to_panel_title_gap_inches_at_print_scale": 0.1,
+    "qwen_counts_and_decisions_unchanged": true,
+    "evaluation_files_unchanged": true,
+    "dataset_card_unchanged_sha256": "ca29d54e353fce155aad7014d0f0cdcf0be1fe437dd037ec0c2cfadb71ff3de2"
+  }
+}

release_manifest.json

@@ -237,9 +237,9 @@
   "paper_bundle": {
     "directory": "paper",
     "archive": "vibtarve_paper.zip",
-    "source_file_count": 132,
-    "file_count": 134,
-    "archive_sha256": "91e48050b8d22fb137ba2222b69bc6b009ecc576a8d22257c606984feafc5761"
+    "source_file_count": 144,
+    "file_count": 146,
+    "archive_sha256": "152cace4bb53001b3fa2601d50cd2bf0bd6c42817360a7a11285c1cb62148cd6"
   },
   "derived_revision": {
     "generated_utc": "2026-10-25T05:06:21.239328+00:00",
@@ -331,5 +331,32 @@
     "code_commit": "d27950199f6185a17b4112175a1c80c851065303",
     "provenance": "provenance/episode_access.json",
     "review_hyperlinks": "Local targets rebased relative to workbook; omitted bt.svg diagrams link to published bt.json instead"
+  },
+  "prompt_development_revision": {
+    "generated_utc": "2026-10-27T13:22:18.798766+00:00",
+    "code_commit": "42a65c902721c06c4b3b86f806b3df7f689c3d26",
+    "previous_revision": "28b6157f7a47e3d9ea5b6cb4ee27ba0e5723742d",
+    "families": {
+      "qwen38": {
+        "revisions": [
+          1,
+          2,
+          3,
+          4,
+          5
+        ],
+        "selected": 4
+      },
+      "luna6": {
+        "revisions": [
+          1,
+          2,
+          3,
+          4,
+          5
+        ],
+        "selected": 5
+      }
+    }
   }
 }

vibtarve_paper.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:91e48050b8d22fb137ba2222b69bc6b009ecc576a8d22257c606984feafc5761
-size 24517247
+oid sha256:152cace4bb53001b3fa2601d50cd2bf0bd6c42817360a7a11285c1cb62148cd6
+size 25007793