Repository navigation
Separate 24h-volume floor for shadow forecasts (R209) - #6
Merged
Merged
Conversation
With an economics-only mandate and the $5k 24h-volume floor nearly nothing can be forecast, so no bucket can reach the 30 scored forecasts it needs. Lowering min_liquidity_usd would also loosen live orders. - mandate.yaml markets.min_forecast_volume_usd (optional, at most min_liquidity_usd; unset keeps today's single floor): submit_forecast needs only that 24h volume. Orders keep the trading floor; the risk engine is unchanged. Opt-in: commented out in the default and example files; config change effect neutral. - search_markets flags markets between the floors "forecast only" and below the forecast floor "no forecast either". - Forecast Executor skips markets whose 24h volume is below the trading floor (ctx.risk.getMandateSummary + ctx.polymarket.getMarket) and cancels its bids there, instead of collecting liquidity rejections that would trip the rejection cooldown. Unknown volume and backtests (no calls) behave as before. - Agent system prompt v6 and guide §3, Ledger page help text, guides, PROGRESS R209 and log. Co-Authored-By: Claude Opus 5.5 <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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Why
With the default mandate (economics / weather / non-US politics allowed) and the $5k 24h-volume floor, almost no market can be forecast: a real agent session on 2026-11-01 had all four forecasts refused, and the Mac test found the same. No (category, horizon) bucket can then reach the 30 scored forecasts it needs to earn trading rights. Lowering
min_liquidity_usdwould fix forecasting but also loosen the floor for live orders.What
mandate.yaml markets.min_forecast_volume_usd: optional, must be ≤min_liquidity_usd. When set,submit_forecastneeds only that much 24h volume. Orders are unchanged: the risk engine still requires the trading floor (the larger of the mandate's andlimits.yaml per_market.min_liquidity_usd). Unset = exactly today's behaviour. Opt-in: the default and examplemandate.yamlcarry it commented out. Config-change effect isneutral(it doesn't raise trading risk).search_markets): markets between the floors are flaggedforecast only (orders need min_liquidity_usd); below the forecast floorno forecast either.inMandateOnlystill drops both.ctx.polymarket.getMarket) is belowctx.risk.getMandateSummary().minLiquidityUsd, and cancels its bids there. Without this it would keep quoting forecast-only markets, collectliquidityrejections and trip the rejection cooldown (R193), blocking its good quotes too. Unknown volume is still left to the risk engine; backtests (no such calls) are unchanged.docs/guide/operations.mdandstrategies.md, PROGRESS decision R209 and log entry.Behaviour change to note
The executor skip also applies to existing forecasts whose market volume has fallen below the trading floor since they were made: the executor now cancels its resting bids there instead of leaving them. Template source changed, so (as with every template change) running executor bots go back to paper on upgrade.
Tests
min_liquidity_usd), ledger gate with and without the key, search flags with and without it, executor skip / cancel plus the unchanged cases (volume at the floor, unknown volume, no mandate summary).npm run typecheckclean; full suiteSKIP_NETWORK_TESTS=1 npx vitest run1241 passed; ESLint 0 errors (excluding a leftover local worktree under.claude/).To use it
Risk → Mandate →
markets.min_forecast_volume_usd: 1000(for example), and add the categories you want shadow-forecast tocategories.later.🤖 Generated with Claude Code