Skip to content

Commit efd2985

Browse files
committed
replace multiplication sign with ascii x
1 parent 2b75c59 commit efd2985

27 files changed

Lines changed: 82 additions & 82 deletions

OVERVIEW_ANIMATION.md

Lines changed: 5 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -61,7 +61,7 @@ flowchart LR
6161
S8 -.- DOC
6262
6363
G1{{"gate: symbolic perf heuristic"}}:::gate
64-
G2{{"gate: max_speedup ≥ 1.05× on Daytona"}}:::gate
64+
G2{{"gate: max_speedup ≥ 1.05x on Daytona"}}:::gate
6565
S2 --> G1 --> S3
6666
S7 --> G2 --> S8
6767
@@ -424,12 +424,12 @@ internally; we batch-submit PRs and let Harbor parallelize.
424424
HuggingFace, re-push their Docker images to DockerHub, mark
425425
`pull_requests.published_at`.
426426

427-
**Why the 1.05× threshold.** The paper's stage 4 statistical test
427+
**Why the 1.05x threshold.** The paper's stage 4 statistical test
428428
(Mann–Whitney U at p<0.002) is the _correctness_ test — does the
429-
speedup exist statistically? The 1.05× threshold is an _interestingness_
429+
speedup exist statistically? The 1.05x threshold is an _interestingness_
430430
filter — the paper's premise is that LLMs should be evaluated on
431431
_meaningful_ optimizations, not noise-floor improvements. A PR that
432-
passes Mann–Whitney with a 1.001× speedup is real but uninteresting;
432+
passes Mann–Whitney with a 1.001x speedup is real but uninteresting;
433433
including it would inflate dataset size while diluting the benchmark.
434434

435435
**Why publish is fail-closed.** If Harbor never ran successfully on
@@ -540,7 +540,7 @@ Important caveats so a reader doesn't take the diagram too literally:
540540
containers, get speedups — is one operation.
541541
- **The paper's "stage 4" (statistical validation) is _inside_ our
542542
stage 7, not our stage 8.** Mann–Whitney U is what Harbor's reward
543-
computation is doing under the hood; the 1.05× filter in publish is
543+
computation is doing under the hood; the 1.05x filter in publish is
544544
on top of that statistical test, not in place of it.
545545

546546
If you need any of those details, fall back to Variation C

PARITY_VERIFICATION.md

Lines changed: 6 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -16,7 +16,7 @@ Legend: ✅ = matches paper · ⚠️ = partial / known divergence · ❌ = wron
1616
| F4 | Long-tail repository performance | Table 3 (Repo popularity quintiles) || Per-task stars sourced from `task.metadata["pr_base_stargazers_count"]`; 39/40 cells populated. Q2 best / Q4 weak matches paper. |
1717
| F5 | Cost efficiency | Figure 4 (Cost-Performance Pareto) || Frontier line + halos; Claude at expensive-end of Pareto matches. |
1818
| F6 | Multi-workload tradeoffs | Figure 5 (Multi-workload tradeoff) || Re-exported from plain `multi_objective_analysis.ipynb` (excludes failed agents). 226 rows = 188 agent + 38 expert. |
19-
| F7 | Temporal generalization | Table 4 (Temporal analysis) | ⚠️ | Schema now matches paper Table 4's 6-bin layout (`6+ mo before``6+ mo after`) and Qwen 3 Coder is excluded — 3 models × 6 bins. **11 of 18 cells populated** vs paper's 18/18; the 7 nulls reflect the upstream task set being sparser per bin than the paper's run. Will fill in as more tasks ingest. |
19+
| F7 | Temporal generalization | Table 4 (Temporal analysis) | ⚠️ | Schema now matches paper Table 4's 6-bin layout (`6+ mo before``6+ mo after`) and Qwen 3 Coder is excluded — 3 models x 6 bins. **11 of 18 cells populated** vs paper's 18/18; the 7 nulls reflect the upstream task set being sparser per bin than the paper's run. Will fill in as more tasks ingest. |
2020

2121
Open upstream issue tracking the divergences: [`formula-code/fc-eval#19`](https://github.com/formula-code/fc-eval/issues/19).
2222

@@ -32,9 +32,9 @@ Open upstream issue tracking the divergences: [`formula-code/fc-eval#19`](https:
3232
| API table | `findings_global_leaderboard` |
3333
| Website component | `src/components/sections/findings/F1_GlobalLeaderboard.svelte` |
3434
| Cached JSON | `src/data/findings/f1_leaderboard.json` |
35-
| Render style | Heatmap table — RP rank, agent, model, advantage (diverging RdBu @ 0), speedup geomean (diverging RdBu @ ) |
35+
| Render style | Heatmap table — RP rank, agent, model, advantage (diverging RdBu @ 0), speedup geomean (diverging RdBu @ 1x) |
3636

37-
**Parity check:** Baseline row's `speedup_geomean` now shows **1.1193×** (was previously `null`), so the diverging color scale on the Speedup column has a center to read against. The RP ordering is what `analysis.task.compute_leaderboard` returns — that's the paper's methodology (Ranked Pairs voting), not advantage-sorted. The earlier-flagged "Claude at rp_rank=1" expectation came from a stale scaffold that had been sorted on advantage rather than computed via RP.
37+
**Parity check:** Baseline row's `speedup_geomean` now shows **1.1193x** (was previously `null`), so the diverging color scale on the Speedup column has a center to read against. The RP ordering is what `analysis.task.compute_leaderboard` returns — that's the paper's methodology (Ranked Pairs voting), not advantage-sorted. The earlier-flagged "Claude at rp_rank=1" expectation came from a stale scaffold that had been sorted on advantage rather than computed via RP.
3838

3939
### F2 — Figure 3 (Stratified advantage)
4040

@@ -62,7 +62,7 @@ Open upstream issue tracking the divergences: [`formula-code/fc-eval#19`](https:
6262
| API table | `findings_tag_advantage` |
6363
| Website component | `src/components/sections/findings/F3_StrategyAdvantage.svelte` |
6464
| Cached JSON | `src/data/findings/f3_tags.json` |
65-
| Render style | Heatmap table — agent × tag matrix, diverging RdBu @ 0 |
65+
| Render style | Heatmap table — agent x tag matrix, diverging RdBu @ 0 |
6666

6767
**Parity check:** The exporter now reads from `task.metadata["classification"]` (the same per-PR enum `analysis/table_8_tags.ipynb` uses), so the website table carries 9 tag keys — `parallelization, batching, caching, algorithmic, data_structure, reduce_work, higher_level, micro, lower_level`. The five paper tags missing from the API (`approximation, scale, db, io, uncategorized`) have zero tasks classified into them in `filtered_formulacode-verified.parquet`; that matches Table 2's actual sparsity. The website renders those columns as `` and otherwise mirrors the paper.
6868

@@ -76,7 +76,7 @@ Open upstream issue tracking the divergences: [`formula-code/fc-eval#19`](https:
7676
| API table | `findings_repo_quintiles` |
7777
| Website component | `src/components/sections/findings/F4_RepoQuintiles.svelte` |
7878
| Cached JSON | `src/data/findings/f4_longtail.json` |
79-
| Render style | Heatmap table — agent × Q1–Q5 matrix, diverging RdBu @ 0 |
79+
| Render style | Heatmap table — agent x Q1–Q5 matrix, diverging RdBu @ 0 |
8080

8181
**Visual sanity:** Q2 row mostly positive (best quintile), Q3 / Q4 rows mostly negative (worst quintiles). Matches paper claim "performance varies dramatically by repository popularity; worst in Q4, best in Q2."
8282

@@ -118,7 +118,7 @@ Open upstream issue tracking the divergences: [`formula-code/fc-eval#19`](https:
118118
| API table | `findings_temporal_generalization` |
119119
| Website component | `src/components/sections/findings/F7_TemporalGeneralization.svelte` |
120120
| Cached JSON | `src/data/findings/f7_temporal.json` |
121-
| Render style | Heatmap table — model × 6 paper-Table-4 bins (`6+ mo before` / `3–6 mo before` / `0–3 mo before` / `0–3 mo after` / `3–6 mo after` / `6+ mo after`), sequential blues on speedup |
121+
| Render style | Heatmap table — model x 6 paper-Table-4 bins (`6+ mo before` / `3–6 mo before` / `0–3 mo before` / `0–3 mo after` / `3–6 mo after` / `6+ mo after`), sequential blues on speedup |
122122

123123
**Parity check:** Upstream re-shipped `findings_temporal_generalization` with the paper Table 4 binning — six 3-month-wide windows (`pre6plus`, `pre3to6`, `pre0to3`, `post0to3`, `post3to6`, `post6plus`), three models (Claude / GPT-5 / Gemini, Qwen excluded per `figure_1_temporal.ipynb` cell 5 popping its cutoff). Cell values are mean `agent/nop` within each bin.
124124

docs/_app/immutable/chunks/B7dnQVhM.js

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -2357,7 +2357,7 @@ var _ = {},
23572357
],
23582358
"±": u.MO.BIN4,
23592359
"·": u.MO.BIN4,
2360-
"×": u.MO.BIN4,
2360+
"x": u.MO.BIN4,
23612361
"÷": u.MO.BIN4,
23622362
ʹ: u.MO.ORD,
23632363
"̀": u.MO.ACCENT,

docs/_app/immutable/chunks/BlNhWkxT.js

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -8859,9 +8859,9 @@ new it.CommandMap(
88598859
new it.CharacterMap("Physics-vector-mo", Dn.default.mathchar0mo, {
88608860
dotproduct: ["⋅", { mathvariant: kt.TexConstant.Variant.BOLD }],
88618861
vdot: ["⋅", { mathvariant: kt.TexConstant.Variant.BOLD }],
8862-
crossproduct: "×",
8863-
cross: "×",
8864-
cp: "×",
8862+
crossproduct: "x",
8863+
cross: "x",
8864+
cp: "x",
88658865
gradientnabla: ["∇", { mathvariant: kt.TexConstant.Variant.BOLD }]
88668866
});
88678867
new it.CharacterMap("Physics-vector-mi", Dn.default.mathchar0mi, {
@@ -9838,7 +9838,7 @@ new nl.CommandMap(
98389838
textsurd: ["Insert", "√"],
98399839
textthreequarters: ["Insert", "¾"],
98409840
textthreesuperior: ["Insert", "³"],
9841-
texttimes: ["Insert", "×"],
9841+
texttimes: ["Insert", "x"],
98429842
texttwosuperior: ["Insert", "²"],
98439843
textuparrow: ["Insert", "↑"],
98449844
textborn: ["Insert", "*"],

docs/_app/immutable/chunks/m924b1c9.js

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -688,7 +688,7 @@ It.asyncLoad = ar;
688688
theta: "θ",
689689
thetav: "ϑ",
690690
tilde: "˜",
691-
times: "×",
691+
times: "x",
692692
triangle: "▵",
693693
triangleq: "≜",
694694
upsi: "υ",
@@ -6264,7 +6264,7 @@ new at.CharacterMap("mathchar0mo", ct.default.mathchar0mo, {
62646264
setminus: "∖",
62656265
cdot: "⋅",
62666266
ast: "∗",
6267-
times: "×",
6267+
times: "x",
62686268
star: "⋆",
62696269
propto: "∝",
62706270
sqsubseteq: "⊑",

docs/_app/immutable/nodes/7.Ck3xZhtG.js

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -122,7 +122,7 @@ Level 2 (Function): algorithms.Sorting.time_sort_int.*</code></pre></div></arti
122122
averaged across groups:</p> <div class="formula svelte-s2t5c8">Adv<sup>(ℓ)</sup> = (1/|G<sup>ℓ</sup>|) Σ Adv<sub>g</sub></div> <p class="svelte-s2t5c8">This captures whether improvements are spread evenly or concentrated
123123
in specific subsystems.</p></div> <div class="metric-def svelte-s2t5c8"><h2 class="svelte-s2t5c8">Normalized Advantage</h2> <p class="svelte-s2t5c8">Signal-to-noise ratio of the advantage, penalizing erratic performance
124124
across workloads:</p> <div class="formula svelte-s2t5c8">Ã = Adv / √(σ²<sub>agent</sub> + σ²<sub>expert</sub>)</div> <p class="svelte-s2t5c8">Rewards agents that deliver steady improvements across many workloads
125-
rather than volatile, concentrated gains.</p></div> <div class="metric-def svelte-s2t5c8"><h2 class="svelte-s2t5c8">Cost-Weighted Advantage</h2> <p class="svelte-s2t5c8">Advantage divided by total inference cost (input + output tokens ×
125+
rather than volatile, concentrated gains.</p></div> <div class="metric-def svelte-s2t5c8"><h2 class="svelte-s2t5c8">Cost-Weighted Advantage</h2> <p class="svelte-s2t5c8">Advantage divided by total inference cost (input + output tokens x
126126
per-token price):</p> <div class="formula svelte-s2t5c8">Cost(Adv) = Adv / C<sub>agent</sub></div> <p class="svelte-s2t5c8">Captures the human-relative improvement obtained per unit of inference
127127
budget.</p></div></article> <article id="submitting" class="doc-section svelte-s2t5c8"><h1 class="svelte-s2t5c8">Submitting Results</h1> <p class="svelte-s2t5c8">After running your evaluation, you can submit results to be included on
128128
the <a href="/leaderboard/" class="svelte-s2t5c8">public leaderboard</a>:</p> <ol class="steps-list svelte-s2t5c8"><li class="svelte-s2t5c8"><strong class="svelte-s2t5c8">Run the full evaluation</strong> on the FormulaCode dataset with

docs/docs/index.html

Lines changed: 9 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -64,7 +64,7 @@
6464
averaged across groups:</p> <div class="formula svelte-s2t5c8">Adv<sup>(ℓ)</sup> = (1/|G<sup></sup>|) Σ Adv<sub>g</sub></div> <p class="svelte-s2t5c8">This captures whether improvements are spread evenly or concentrated
6565
in specific subsystems.</p></div> <div class="metric-def svelte-s2t5c8"><h2 class="svelte-s2t5c8">Normalized Advantage</h2> <p class="svelte-s2t5c8">Signal-to-noise ratio of the advantage, penalizing erratic performance
6666
across workloads:</p> <div class="formula svelte-s2t5c8">Ã = Adv / √(σ²<sub>agent</sub> + σ²<sub>expert</sub>)</div> <p class="svelte-s2t5c8">Rewards agents that deliver steady improvements across many workloads
67-
rather than volatile, concentrated gains.</p></div> <div class="metric-def svelte-s2t5c8"><h2 class="svelte-s2t5c8">Cost-Weighted Advantage</h2> <p class="svelte-s2t5c8">Advantage divided by total inference cost (input + output tokens ×
67+
rather than volatile, concentrated gains.</p></div> <div class="metric-def svelte-s2t5c8"><h2 class="svelte-s2t5c8">Cost-Weighted Advantage</h2> <p class="svelte-s2t5c8">Advantage divided by total inference cost (input + output tokens x
6868
per-token price):</p> <div class="formula svelte-s2t5c8">Cost(Adv) = Adv / C<sub>agent</sub></div> <p class="svelte-s2t5c8">Captures the human-relative improvement obtained per unit of inference
6969
budget.</p></div></article> <article id="submitting" class="doc-section svelte-s2t5c8"><h1 class="svelte-s2t5c8">Submitting Results</h1> <p class="svelte-s2t5c8">After running your evaluation, you can submit results to be included on
7070
the <a href="/leaderboard/" class="svelte-s2t5c8">public leaderboard</a>:</p> <ol class="steps-list svelte-s2t5c8"><li class="svelte-s2t5c8"><strong class="svelte-s2t5c8">Run the full evaluation</strong> on the FormulaCode dataset with
@@ -74,14 +74,14 @@
7474
configuration details.</li> <li class="svelte-s2t5c8"><strong class="svelte-s2t5c8">We&#39;ll review and publish</strong> your results on the leaderboard
7575
within one update cycle (monthly).</li></ol> <div class="info-box svelte-s2t5c8"><strong class="svelte-s2t5c8">Questions?</strong> Open an issue on
7676
<a href="https://github.com/formula-code/" target="_blank" rel="noopener" class="svelte-s2t5c8">GitHub</a>
77-
or reach out to the authors listed on the <a href="/" class="svelte-s2t5c8">overview page</a>.</div></article></main> </section></main> <footer class="svelte-126cfes"><div class="c svelte-126cfes"> <div class="paper-footer-sections svelte-126cfes"> <section class="paper-section citation-section svelte-126cfes"><h3 class="paper-section-title svelte-126cfes">Citation</h3> <div class="bibtex-wrapper svelte-126cfes"><pre class="bibtex-code svelte-126cfes"><code class="svelte-126cfes">@misc{sehgal2025formulacode,
78-
title={FormulaCode: Evaluating Agentic Optimization on Large Codebases},
79-
author={Atharva Sehgal and James Hou and Akanksha Sarkar and Ishaan Mantripragada and Swarat Chaudhuri and Jennifer J. Sun and Yisong Yue},
80-
year={2025},
81-
eprint={2409.XXXXX},
82-
archivePrefix={arXiv},
83-
primaryClass={cs.LG},
84-
url={https://arxiv.org/abs/2409.XXXXX},
77+
or reach out to the authors listed on the <a href="/" class="svelte-s2t5c8">overview page</a>.</div></article></main> </section></main> <footer class="svelte-126cfes"><div class="c svelte-126cfes"> <div class="paper-footer-sections svelte-126cfes"> <section class="paper-section citation-section svelte-126cfes"><h3 class="paper-section-title svelte-126cfes">Citation</h3> <div class="bibtex-wrapper svelte-126cfes"><pre class="bibtex-code svelte-126cfes"><code class="svelte-126cfes">@misc{sehgal2025formulacode,
78+
title={FormulaCode: Evaluating Agentic Optimization on Large Codebases},
79+
author={Atharva Sehgal and James Hou and Akanksha Sarkar and Ishaan Mantripragada and Swarat Chaudhuri and Jennifer J. Sun and Yisong Yue},
80+
year={2025},
81+
eprint={2409.XXXXX},
82+
archivePrefix={arXiv},
83+
primaryClass={cs.LG},
84+
url={https://arxiv.org/abs/2409.XXXXX},
8585
}</code></pre></div></section> <section class="paper-section funding-section svelte-126cfes"><h3 class="paper-section-title svelte-126cfes">Acknowledgements</h3> <p class="paper-section-text svelte-126cfes"><!-- HTML_TAG_START -->This project was made possible by a Slingshot Award from the <a href=https://www.laude.org/>Laude Institute</a> and an <a href=https://www.neurosymbolic.org/>NSF Expeditions</a> grant.<!-- HTML_TAG_END --></p></section></div> <div class="footer-acknowledgements svelte-126cfes"><p class="svelte-126cfes"><!-- HTML_TAG_START -->The website design heavily uses the template developed by <a href=https://pudding.cool/author/fox-meyer/>Fox Meyer</a> and <a href=https://pudding.cool/author/jan-diehm/>Jan Diehm</a> for their interactive article in <a href='https://pudding.cool/'>Pudding.cool</a> on <a href='https://github.com/the-pudding/wine-animals'>The Pour-ing of species</a> that is distributed under an MIT license. The code itself is based on <a href='https://github.com/the-pudding/svelte-starter' target='_blank'>The Pudding's SvelteKit starter template</a>. The Visualizations use <a href='https://layercake.graphics/' target='_blank'>LayerCake</a> and <a href='https://d3js.org/' target='_blank'>D3.js</a>. The source code for this website is available <a href='https://github.com/formula-code/formula-code.github.io/'>here</a>, also under an MIT license.<!-- HTML_TAG_END --></p></div></div> </footer>
8686

8787
<script>

0 commit comments

Comments
 (0)