Document triple-barrier topics and classification scoring in Forge

[jefferythewind]

Summary

Adds a definitive triple-barrier reference under Forge Competitions → Scoring, alongside the existing log-return and volatility sections. Readers can link directly to /build/forge/competitions#triple-barrier-topics from the upcoming blog and Builder Kit material.

Covers testnet topics 87 (Gold), 88 (Silver), and 89 (WTI), their Atlas dataset names, and the shared 24-hour horizon.

Definition and evaluation

• Specifies UTC candle opening timestamps, the 2,400-hour historical selection, inclusive 25-opening rolling log ranges, and partial initial ranges matching the supplied reputer SQL. Distinguishes this log-range statistic from conventional ATR.
• Defines the minute-close base, exponential barriers, exact minute testing interval, inclusive high/low touches, down-first same-minute ties, and unresolved targets when required coverage is missing.
• Documents labeled probability submissions, declared class order, deterministic evaluation argmax, causal previous-100-resolved-target baseline, all six strict eligibility criteria, paired circular bootstrap settings, focal clipping, and undefined-statistic handling.
• Separates optional cost-declared directional payoff from eligibility and realized trading PnL.
• Links to the reviewed Builder Kit walkthrough, README, target builder, and evaluator at commit cb63774 (introduced by Add triple-barrier targets, evaluation, and labeled worker walkthrough allora-forge-builder-kit#43), so links resolve before that PR merges.

Scopes the existing return/volatility promotion criteria to those topic families rather than implying they apply to classification. Clarifies that offline reports do not establish live participation or guarantee promotion. No new page, route, or sidebar entry is added.

Sources and generated files

Target details were checked against the reputer SQL supplied for ENGN-9552; classification criteria and edge-case conventions come from the supplied evaluation contract. The linked Builder Kit implementation was independently tested against raw minute candles. Frontmatter records that provenance rather than claiming verification against an unidentified reputer release.

Regenerates public/llms-full.txt and public/raw/build/forge/competitions.md. The generated page index remains unchanged because the page title and description are unchanged.

Validation

• yarn build (Node 20.12.2)
• yarn fixlinks
• yarn checkllms
• yarn checkraw
• git diff --check

No executable snippets or dependencies changed. Build warnings concern existing optional sharp installation, Browserslist age, and Nextra _app.jsx guidance; link fixing reports existing duplicate filenames.

---

Summary by cubic

Adds a definitive reference for triple-barrier topics under Forge Competitions → Scoring, covering testnet topics 87 (Gold), 88 (Silver), and 89 (WTI). The new section documents the 24-hour target construction, labeled probability submissions, the six classification eligibility criteria, and the optional directional-payoff diagnostic. Existing return/volatility promotion criteria are now scoped to those topic families so they do not imply they apply to classification. Links now point to the merged Builder Kit implementation on main.

Details

• Regenerates public/llms-full.txt and public/raw/build/forge/competitions.md; the generated page index is unchanged.

^{Written for commit 3a9750b. Summary will update on new commits.}

[netlify]

✅ Deploy Preview for alloradocs ready!

Name	Link
🔨 Latest commit	3a9750b
🔍 Latest deploy log	https://app.netlify.com/projects/alloradocs/deploys/6abffd0b1a76d50008c43fcf
😎 Deploy Preview	https://deploy-preview-187--alloradocs.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify project configuration.

[jefferythewind]

Just pushed and checked, all links are now to public GH locations.