Clarify Forge scoring confidence intervals

[jefferythewind]

Summary

• Expand Forge scoring docs to explain point estimates vs confidence intervals and lower-CI threshold checks
• Add mainnet promotion metric criteria and link to the Allora Research promotion/relegation post
• Add separate notes for log-return and volatility topic evaluation
• Regenerate raw markdown and llms-full output

Verification

• corepack yarn build

---

Summary by cubic

Clarifies Forge scoring docs to distinguish point estimates from confidence intervals and how lower-CI thresholds drive promotion checks. This helps builders read the dashboard correctly and understand mainnet promotion criteria; documentation-only, no behavior change.

• Explains dashboard CI notation and that promotion checks use the lower CI, not just the point estimate.
• States per-worker, per-topic evaluation and the horizon-based adjustment applied to directional accuracy, WRMSE improvement, and WCZAR improvement.
• Adds explicit promotion gates: effective sample size ≥ 20; directional accuracy one-sided 95% lower CI > 50%; Pearson correlation two-sided 95% lower CI > 0; WRMSE and WCZAR improvement lower CIs > 0%; log aspect ratio CI overlaps [-0.5, 0.5]; participation > 90%.
• Adds topic notes: log-return topics are evaluated directly in log-return space; volatility topics evaluate log volatility change with a causal mean-reversion baseline.
• Links to the Allora Research promotion/relegation policy and updates verification metadata; regenerates public/raw and public/llms-full.txt.

^{Written for commit 1898bb8. Summary will update on new commits.}

[netlify]

✅ Deploy Preview for alloradocs ready!

Name	Link
🔨 Latest commit	1898bb8
🔍 Latest deploy log	https://app.netlify.com/projects/alloradocs/deploys/6a7dd57f56fa350008132678
😎 Deploy Preview	https://deploy-preview-181--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]

lgtm

[cubic-dev-ai]

1 issue found across 3 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="public/llms-full.txt">

<violation number="1" location="public/llms-full.txt:3281">
P3: The threshold text conflates "raw" and "effective" sample sizes, which are opposite concepts in statistics and directly opposed in the paragraph above it. "Raw" (unadjusted count of observations) and "effective" (autocorrelation-corrected count) cannot describe the same quantity, so readers can't tell whether the required 20 are raw or corrected. Write one term only, or clarify the relationship.</violation>
</file>

Architecture diagram

sequenceDiagram
    participant Docs as Forge Scoring Docs
    participant Chain as On-chain Scoring
    participant Forge as Forge Dashboard
    participant Worker as Worker

    Note over Docs,Worker: Forge Scoring & Promotion Flow

    Chain->>Chain: Score submissions vs ground truth (per epoch)
    Chain->>Forge: NEW: Summarize worker history into promotion metrics
    Forge->>Forge: Compute point estimates + confidence intervals
    Note over Forge: Per worker, per topic evaluation

    Forge->>Forge: Apply effective sample size correction
    alt Long-horizon topics
        Forge->>Forge: CHANGED: Apply horizon adjustment to accuracy/WRMSE/WCZAR
    end

    Forge-->>Worker: Display: "your value (lower CI, upper CI)"
    Note over Worker,Forge: Promotion uses lower CI (first value in parens)

    alt Log-return topics
        Forge->>Forge: Compare predictions in log-return space directly
        Note over Worker: No price conversion needed
    else Volatility topics
        Forge->>Forge: Evaluate log-volatility change
        Forge->>Forge: CHANGED: Subtract causal mean-reversion baseline
    end

    Forge->>Forge: Check promotion gates
    alt All gates passed
        Note over Worker: Eligible for mainnet promotion
    else Some gate failed
        Note over Worker: Stay on testnet - CI threshold not met
    end

    Note over Worker,Forge: See Allora Research forum for full policy

Loading

<sub>Reply with feedback, questions, or to request a fix.

Fix all with cubic | Re-trigger cubic</sub>

[cubic-dev-ai]

P3: The threshold text conflates "raw" and "effective" sample sizes, which are opposite concepts in statistics and directly opposed in the paragraph above it. "Raw" (unadjusted count of observations) and "effective" (autocorrelation-corrected count) cannot describe the same quantity, so readers can't tell whether the required 20 are raw or corrected. Write one term only, or clarify the relationship.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At public/llms-full.txt, line 3281:

<comment>The threshold text conflates "raw" and "effective" sample sizes, which are opposite concepts in statistics and directly opposed in the paragraph above it. "Raw" (unadjusted count of observations) and "effective" (autocorrelation-corrected count) cannot describe the same quantity, so readers can't tell whether the required 20 are raw or corrected. Write one term only, or clarify the relationship.</comment>

<file context>
@@ -3264,9 +3264,49 @@ The cycle repeats every epoch, so a competition is not a one-shot submission: yo
+
+For mainnet promotion, a worker must pass every promotion metric on the topic:
+
+- **Effective sample size** — at least 20 raw effective observations
+- **Directional accuracy** — one-sided 95% lower CI greater than 50%
+- **Pearson correlation** — two-sided 95% lower CI greater than 0
</file context>

Suggested change

	- **Effective sample size** — at least 20 raw effective observations
	- **Effective sample size** — at least 20 effective observations