> ## Documentation Index
> Fetch the complete documentation index at: https://docs.usetissue.xyz/llms.txt
> Use this file to discover all available pages before exploring further.

# Core Concepts

> Fair value, the five regimes, and the discipline that keeps Tissue from chasing the market.

## The fair-value model

Tissue prices a match in two phases:

1. **Freeze.** At kickoff, TxLINE's opening de-vigged consensus odds set the base
   scoring intensities for both teams. The free totals-only feed uses a neutral team
   split; when 1X2 odds are also present, they solve the relative scoring share.
2. **Reprice.** As the match progresses, Tissue reprices the remainder of the match
   using a Poisson goals model with Dixon-Coles low-score dependence, conditioned on
   verified current score, remaining playing time, and red cards.

This fair value is compared against TxLINE's live StablePrice consensus. When the two
diverge by more than `strategy.edge_threshold_bps` (after de-vigging), and every risk
gate clears, Tissue publishes a two-sided quote.

<Note>
  TxLINE's StablePrice is a single de-margined consensus stream, not per-bookmaker odds.
  This was verified directly against captured real data and TxLINE's current live docs —
  it shapes what kind of adverse-selection signal is honestly possible (see
  [Informed-Flow](#informed-flow), below).
</Note>

## The five regimes

Each regime is a flagged, individually toggleable heuristic addressing one specific
real-world trading question. All five are neutralized in the Strategy Arena's Baseline
agent and can be isolated individually via the regime ablation matrix.

| Regime                | Trading question it answers                                   | Effect                                                                                                                                                                                                                                            |
| --------------------- | ------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Stoppage-time**     | How should added time be priced?                              | Real playing time gets a bounded lambda floor instead of zeroing goal probability at minute 90 — a genuine bug fix that also functions as a regime                                                                                                |
| **Mutual-danger**     | How should sustained two-way pressure be priced?              | Sustained simultaneous high-pressure events (not a single spike) widen spread and cut size — the next-goal distribution is bimodal here, not the Poisson point estimate's confident middle                                                        |
| **Narrative regime**  | Is the market's recent behavior itself informative?           | A rolling window of the Radar's own signal-class history classifies the market as neutral, cautious, compounding, or oscillating, and sizing conditions on that regime, not just the most recent signal                                           |
| **Informed-flow**     | Is this price move unusually sharp for this market right now? | A Glosten-Milgrom-style adverse-selection check adapted to what TxLINE actually exposes: it classifies a move's *velocity* against that market's own trailing distribution, since there is no cross-bookmaker propagation data to compare against |
| **Stale-quote decay** | Should an unchallenged resting quote get tighter over time?   | The age of Tissue's *own* resting quote (real, already-tracked state) compresses spread as it sits unchallenged                                                                                                                                   |

## The discipline engine

Risk gates are the **only** module authorized to green-light execution. They check, in
order:

* **Feed-gap halt** — the feed went quiet longer than tolerance; every open intent is
  cancelled
* **Unexplained-movement halt** — odds moved with no feed cause Tissue can see
* **Informed-flow halt** — this market's move velocity is anomalous against its own
  trailing distribution
* **Model-divergence band** — Tissue's own price disagrees with the market by more
  than a sanity band allows (protects against Tissue's own model failure, not the
  market's)
* **Drawdown kill** — realized drawdown crosses a threshold; operator-restart-only,
  never auto-resumes
* **Portfolio-level caps** — the same exposure and drawdown checks, aggregated across
  every fixture the desk is concurrently running (the World Cup schedule stages
  multiple knockout ties in parallel; a per-fixture-only cap would multiply uncapped
  total risk by fixture count)
* **Proof-failure-rate circuit breaker** — an aggregate check distinct from a single
  rejected message: if the *rate* of recent source-proof failures crosses a threshold,
  the whole desk halts, because that signals a degraded proof service, not one bad
  message

<Warning>
  Unexplained-movement and informed-flow are the two halts that exist specifically
  *because* the market moved. Tissue's rule is simple and mechanical: it will not quote
  against information it doesn't have, even when doing so looks like leaving edge on the
  table.
</Warning>

## Network consistency

See [Architecture → Network consistency](/architecture#network-consistency).

## Signed decisions and receipts

Every decision record carries:

* A **hash** linking it to every prior decision (`H(prevHash ‖ canonical(record))`)
* An **Ed25519 signature** over that hash, from the same Solana keypair used for
  on-chain anchoring
* A **Merkle inclusion proof**, on request, against a specific anchored checkpoint root

See [Verifiability & Audit Trail](/verifiability) for exactly how to check any of this
independently.