blog: Part 2 complete + Part 1 reconciliation#28
Merged
Conversation
Part 2 (post + 4 figures + captions + renderer output): the agentic staircase with a real forecast anatomy; agents on the same scoreboard incl. the war/quiet decomposition and the divergence-sentinel finding (exploratory, n=24); Track-2 scenarios + judge with the right-call-wrong-reason set piece; the honest limit; adaptive pre/post reported as an instructive null with the same-session-gate finding (verified exhaustive: all 28 hypotheses opened_on == confirmed_on); coda. Part 1: h=1 eval digits reconciled post calendar-fix (0.00497/0.00501), fig3/fig4 re-rendered (fig4 lightgbm_cov h=21 gap closed), fig5 byte-identical, EDITORS-NOTES harmonized, _blogdata gains BLOG_WS_DATA_ROOT/BLOG_YF_CACHE env overrides (defaults unchanged). All figure numbers independently recomputed from stores with 1e-5 assertions. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…res, width decomposition, verdict table, QA fixes - fig5: complete 21-method board across horizons (h=5 LLM-family sweep now told); fig6: war/quiet split + paired agency dumbbells (title corrected in QA from editorializing against its own data) - §2: width decomposition of the divergence — the alarm is the width, not the center (rho=0.52 vs 0.20); closes on conventional-for-nominal-periods + agent-as-sentinel framing - §3: four-verdict score table; §5: hand-written-discipline finding sharpened - QA fixes: caption/paragraph break, bottom-third -> 14th and 17th precision, quiet-week penalty synced to fully-resolved n=24 cut (+22%, n=14) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…olumn width Root cause: figures were authored on a ~2000px+ canvas with wide aspect ratios (footnote overflow + bbox_inches=tight inflated widths to 3.3-3.9:1), so text downscaled to ~8-10px in the ~1090px blog column. Retuned all six Part-2 figures to aspect <=2.0 with fonts sized for the actual display width; QA-verified at 1090px. Two dataviz fixes: fig3 router bars and fig4 pre/post bars now use zoomed/broken axes so the (small but real) differences are visible instead of flattened to identical-looking bars from a zero baseline. Layout-only changes; every in-script data assertion preserved, no plotted number changed. Part-1 nits: fig2 em-dash, fig3 stale n/a footer removed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The label floated at y~22 while its dashed line sits at ~17.2, reading as detached (Ethan spotted). The panel center is filled by the pre/post bars, so no in-panel label fits near the line without colliding — moved it to a dashed-line entry in the top legend instead. Layout-only; asserts unchanged. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The 4-across legend ran the LightGBM+cov reference entry into the WHAT GRADUATED card column at native resolution (Ethan caught it in the rendered page; my 1090px downscale hid the near-collision). Switched to a 2x2 legend confined to the left panel block and lowered the panels slightly for headroom. Verified at native resolution, not just downscaled. Layout-only; asserts unchanged. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Full two-part series ready for review. Part 2 sections 1-6 with all numbers final and sourced; Part 1 digits/figures reconciled under the covariate calendar fix. Self-contained HTML for both parts under blog/dist/. NOT for merge until Ethan reviews.
🤖 Generated with Claude Code