PolyRobin v1.2.1 — Honest Rails

PolyRobin v1.2.1 — Honest Rails

The spec told @BankrBot to do three things it cannot do: deploy an escrow contract, create a 2-of-3 multisig, and create a custom Polymarket market. None of those rails exist. This release removes the claim.

Changed

• Friend bets are held-stakes, not escrow. Bankr bets on listed Polymarket
  markets and moves funds by transfer to a handle or address. It does not lock
  stakes in a contract. The social-bet model now says so: a listed market first,
  since that is the only trustless path; otherwise a trust-based side bet whose
  stakes sit with an agreed holder, adjudicated against the resolution statement
  and settled by a confirmed transfer; and if no market exists and nobody will
  hold the stakes, PolyRobin stands down. That last outcome is valid, not a
  failure.
• The trust assumption is named, never buried. The old rule read "no trust
  assumptions when money is at stake," which was never achievable without escrow
  rails. A friend bet always rested on trust; the spec simply wasn't admitting
  it. It now states the assumption plainly and prefers a neutral stake-holder
  over an interested party.
• Updated to match: the frontmatter description, the "what PolyRobin is and
  isn't" bullet, path selection, tracking and settlement, the worked example,
  the fairness rules, the example command, and both social-bet edge cases.

Unchanged

The parsing, the resolution statement, the refusal to run ambiguous or manipulable bets, the objective-source requirement, gate 5 confirmation, and the bad-faith guard. None of that value depended on rails that don't exist.

PolyRobin v1.2.0 — Data Integrity

PolyRobin v1.2.0 — Data Integrity

Live testing on @BankrBot caught v1.1.0 doing three things a prediction-market
analyst must never do: echoing the market's own price back as its "estimate,"
labeling figures "net" while never subtracting the fee, and manufacturing
conviction on coin-flip markets. This release fixes all three — and writes the
rules into SKILL.md so it cannot drift back.

The rules that fix it

• Estimate before price, or no estimate. Form the probability p from sources
  first; only then fetch the price c. An echoed price — or a sub-cent nudge off
  one — is not an estimate. No source-backed reason to move off the price means no
  edge: say so and stand down.
• The spread is a cost, not an edge. Both YES and NO come from their own live
  quotes, never 1 − the other side. The gap is what you pay to enter and exit; an
  edge smaller than the fetched spread is not a trade on either side.
• Fees are read, not assumed. The taker rate comes from the market's live
  feeSchedule. A market is fee-free only when it carries feesEnabled: false — a
  zero rate is a different, rarer thing. Taker cost is rate × c × (1 − c) per
  share, or rate × (1 − c) as a fraction of stake — largest at low prices,
  precisely where it is easiest to overlook. "Net" never prints without the
  deduction visible.
• Low-price EV guard. Because c is the denominator of EV_gross, a half-cent
  error in p at c = 0.045 prints as +11%. Below 10¢, p must be sourced to the
  half-cent or the market is a stand-down, whatever the EV figure claims.
• Stand down on coin-flips. Fifteen-minute BTC, ETH, and SOL up-or-down markets
  are roughly 50/50 noise, and taker fees eat any thin edge. PolyRobin does not
  invent an edge that isn't there.
• Never quote unfetched data. Every price, depth, and volume figure must come
  from a real market pulled that turn and cited by title and slug or URL — no
  number that wasn't fetched, and no false precision.
• Never invent a bankroll. A dollar size is valid only against the user's actual
  balance; otherwise size is expressed as a percentage of bankroll, and PolyRobin
  asks for the stake.
• A gate you could not run has not passed. Without the user's bankroll and open
  positions, the exposure gates are unverified — the reply now says "can't check
  your limits — tell me your bankroll," never "all clear."

Analysis and sizing are separate

• Analysis is the default. Both sides are quoted, with a read on where the value
  sits. No stake, no dollar amount, no "reply yes to place."
• Picking a side is its own turn. Bet intent without a named side returns both
  quotes and the spread, then asks which side. YES is never assumed.
• Sizing runs only on the side the user names. A side chosen against the edge is
  honored if it clears the gates — flagged, not overridden. Sizes above ¼-Kelly are
  honored when every gate still clears, and are always flagged.

Replies humans can read

Cents, not decimals (39¢). The spread reads as "2¢ to trade," liquidity as "deep
enough to exit ✅." No jargon on X. The fee appears in the same breath as the value —
"3¢ of value, ~1¢ fee, ~2¢ left" — never "after fees," which hides the subtraction.
Roughly four lines, under 500 characters; the full math, the Kelly working, and every
gate line by line live behind why.

Also in this release

• Universal coverage. The category table is a set of examples, not an allow-list:
  if Polymarket lists it, PolyRobin covers it.
• Exact formulas pinned in SKILL.md: edge, gross EV, slippage, fees, net EV,
  full Kelly f* = (p − c) / (1 − c), and fractional-Kelly sizing capped by every
  gate and rounded down — with a worked check that makes the math reproducible.
• Sample Rationale Cards at examples/rationale-card.json and
  examples/rationale-card.md, with the fee now read from the market's feeSchedule
  rather than assumed.
• The validator now checks safety invariants, not just YAML.
• The X interaction model is explicit: you tag @bankrbot, and he replies to you
  on X.
• The README explains how funds flow through Polymarket and adds a $PR token
  section. The token is separate from the skill and plays no part in its analysis or
  gates.
• The Polymarket referral link has been removed.
• Meridian Predict is confirmed as the real venue (Robinhood and Susquehanna),
  and the geo-restriction caveats are gone.
• CI: resolvable changelog links, and the README version synced to 1.2.0.

Still true

PolyRobin is a behavior spec for @BankrBot, not executable code. It holds no keys and
moves no funds. Every material action requires your explicit yes. "No edge, standing
down" remains a frequent and correct output.

See CHANGELOG.md for the itemized list of changes.

PolyRobin v1.1.0 — Social & Friend Bets

Added

• Social & Friend Bets — a new major capability in SKILL.md. Turns a natural-language friend/group wager (e.g. "bet $100 my friend Tony loses $100 today on memes") into structured terms (condition, amount, parties, resolution criteria, deadline, source), writes a fair and verifiable resolution statement, and suggests the best execution path: a real custom market (Polymarket / Meridian Predict / Hunch) where one exists, otherwise a peer-to-peer escrow via BankrBot wallet tools (escrow contract, multisig, or simple conditional transfer). Runs the full 7 safety gates — confirmation required before any funds move — and tracks/auto-settles on resolution. Includes a dedicated section, example commands, and edge cases.
• catalog.json — required BankrBot discovery metadata (slug, provider, logo, install command); without it the skill is excluded from the catalog.
• logo.svg — square brand mark.
• Execution Routing & Chain Targeting section in SKILL.md — forces Robinhood Chain by default and forbids silent fallback to Base for Morpho, swaps, bridging, and tokenized stocks (fixes the "landed on Base" failure mode).
• execution block in the sample config encoding per-action chain targeting.
• Validator now checks catalog.json; README documents the real install routes (registry PR or npx skills add), correcting the earlier mention-a-repo myth.

Changed

• Reframed PolyRobin as a decision co-pilot / behavior spec on top of BankrBot's existing rails — explicitly not executable code and not an autonomous trader.
• Execution scope made realistic: Polymarket betting and Robinhood Chain tokenized stocks/swaps/bridging run through BankrBot; Meridian Predict is discovery + analysis only for now, with Polymarket fallback where an equivalent exists.
• Example Commands rewritten in natural-language @bankrbot style.
• Integration Hooks split into "rails BankrBot already supports" vs "PolyRobin's own analysis layer".
• Config restructured to decision parameters only (venues + rails), no keys.
• Added bankrbot, social-bets, and friend-bets tags.
• Reworked the skill around a dual-venue model with Robinhood Chain as a first-class home venue (Meridian Predict + tokenized event markets) alongside Polymarket.
• Frontmatter: single-sentence description, focused tag set, added visibility: public.
• Safety model restated as 7 hard gates (adding an always-on confirmation gate and a conviction+EV gate), with HALT-on-volatility, a kill-switch, and a separate emergency pause.
• Bridging now always routes funds into Robinhood Chain with optimal routing (no fixed source chain).
• Added Chainlink oracle integration and auto-claiming on resolution.
• Repository scaffolding: sample config, CI validation, security policy, contribution guide, issue/PR templates.

PolyRobin v1.0.0

Added

• Initial PolyRobin skill (SKILL.md) — autonomous, safety-first prediction market agent for Polymarket.
• Seven-gate safety model with kill-switch and HALT state.
• Decision pipeline: discover → model → price edge → size → confirm → execute → manage → resolve → report.
• Independent probability modeling from news, onchain, sentiment, and resolution-criteria analysis, with a 0–100 conviction score.
• Fractional-Kelly, volatility-adjusted position sizing with portfolio caps.
• Seven composable strategies (Conviction Value on by default).
• Cross-protocol hedging via Hyperliquid perps and Morpho borrow/lend.
• Robinhood Chain support: USDC bridging, cross-chain views, RWA markets.
• Three autonomy levels (assisted / supervised / autonomous).
• Full auditability via per-decision Rationale Cards.
• Professional README.md and MIT LICENSE.