The project's README from github.com/cioffivictor/ssp-protocol. As one markdown file: README.md.
SSP is liquidity infrastructure for Solana memecoin markets. It adjusts liquidity distribution, trading fees and capital allocation in response to volatility, pool depth and market conditions. A reserve backstop can deploy capital into pools whose depth falls below a threshold, which reduces price impact during sell pressure.
SSP does not peg prices and does not claim to prevent drawdowns. The objective is that pool depth responds to conditions instead of staying fixed.
Status. The risk kernel, the operations console and the on-chain Anchor program are implemented. The program targets devnet and is unaudited.
A set of liquidity agents attests to the state of each supported pool. Each agent carries a capital weight. Agents submit one of three stances on each pool assertion.
| Stance | Meaning |
|---|---|
healthy |
The agent asserts the pool state holds. |
stressed |
The agent asserts the pool state fails. |
abstain |
The agent has no signal. Recorded, carries no weight. |
The risk kernel aggregates the attestations into one liquidity regime per pool assertion. The regime determines the action applied to the pool.
| Regime | Condition | Action |
|---|---|---|
EXPAND |
Broad agreement, low stress. | Add depth, reduce fee, release reserve. |
WATCH |
Directional lean below the expansion bar. | Hold allocation, skew fee toward the lean. |
HEDGE |
Meaningful weight on both sides. | Rebalance inventory, raise fee. |
BACKSTOP |
Insufficient decisive signal. | Deploy reserve capital as a floor. |
BACKSTOP is a result, not an error path. An assertion with no decisive signal
is a distinct finding from an assertion that fails.
Figure 1. Component layout. The kernel is offline and dependency-free. The on-chain program is dashed because it is not implemented yet.
The kernel does not run agents, does not call models and does not touch the network. It reads a recorded epoch and derives a result. Identical telemetry and identical attestations produce identical regimes, identical report bytes and an identical attestation digest.
Figure 2. Record model. One epoch, many agents and assertions, many attestations, zero or more citations per attestation.
weight defaults to 1.0. confidence is clamped to [0, 1]. A position that
references an undeclared agent or claim is rejected at validation, so a typo
cannot silently drop a vote.
Figure 3. One epoch end to end. The kernel only reads the record; it never calls back into the agents.
Every non-abstaining attestation casts a signed vote:
vote = sign(stance) * capital_weight * oracle_confidence
where sign(healthy) = +1 and sign(stressed) = -1. Let S be the sum of
healthy votes and C the sum of stressed magnitudes. Then:
decisive_depth = S + C
regime_polarity = (S - C) / decisive_depth in [-1, +1]
stress_ratio = min(S, C) / decisive_depth in [0, 0.5]
regime_polarity is the direction of the lean. decisive_depth is the capital
behind the call. stress_ratio is the share of that capital on the losing side.
The classifier reads those three values against the risk policy:
BACKSTOP when decisive_depth <= minimum_depthEXPAND when |regime_polarity| >= expansion_threshold and stress_ratio < stress_ceilingHEDGE when stress_ratio >= stress_ceilingWATCH otherwise
Figure 4. Classifier decision flow. Four mutually exclusive outcomes, three policy parameters.
The epoch also produces one scalar, the Stability Index, in [0, 1]. It is the
depth-weighted mean of |regime_polarity| across assertions. A set of agents that
agrees on heavily attested assertions scores near 1.0. An evenly divided one
scores near 0.
Two agents attest to one assertion. Agent vlt, capital weight 1.5, attests
healthy at confidence 0.8. Agent rsk, weight 1.0, attests stressed at
confidence 0.6.
S = 1.5 * 0.8 = 1.20
C = 1.0 * 0.6 = 0.60
decisive_depth = 1.80
regime_polarity = (1.20 - 0.60) / 1.80 = +0.333
stress_ratio = 0.60 / 1.80 = 0.333
With the default policy (expansion_threshold = 0.66, stress_ceiling = 0.34),
stress_ratio 0.333 is below the ceiling so the result is not HEDGE, and
|polarity| 0.333 is below the threshold so it is not EXPAND. The regime is
WATCH, leaning healthy.
Figure 5. Decisive depth per assertion in samples/pool-stress.qf, with the resulting Stability Index. Assertion p5 carries no decisive depth and resolves to BACKSTOP.
Three parameters control classification. All are range-validated.
| Parameter | Flag | Default | Effect |
|---|---|---|---|
expansion_threshold |
--consensus |
0.66 |
Required polarity to release capital. |
stress_ceiling |
--dissent |
0.34 |
Losing-side share that forces a hedge. |
minimum_depth |
--min-mass |
~0 |
Decisive depth required to act at all. |
The defaults encode a two-thirds supermajority rule with a limited stress tolerance. Raising the threshold makes expansion rarer. Lowering the ceiling makes hedging more frequent.
Figure 6. Regime map. Depth rises up the y axis, volatility rises along the x axis. Points are illustrative pool positions.
An attestation is a canonical snapshot of an epoch together with its regimes, stamped with a content digest (FNV-1a). Attestations are byte-stable, so they diff cleanly and can be checked in CI before any on-chain submission.
cargo run --release -- bundle samples/pool-stress.qf -o attestation.json
cargo run --release -- verify samples/pool-stress.qf
Figure 7. Bundle and verify. Canonicalization is what makes the digest stable across a parser round trip.
The digest covers the canonical body with a fixed key order, positions sorted by
(assertion, agent, stance) and floats snapped to six decimals. The snap makes
the JSON writer idempotent, so a bundle survives a parser round trip without
changing its digest. Any mutation makes verify exit with code 4.
The on-chain half lives in programs/ssp-registry, an Anchor program targeting
devnet. It records the regime the kernel produced and applies it to the pool.
| Instruction | Effect |
|---|---|
register_pool |
Creates the pool config and its reserve vault. |
submit_attestation |
Records one agent's stance and confidence for the epoch. |
settle_regime |
Writes the regime, polarity and stress, and reprices the fee. |
deploy_backstop |
Pulls reserve capital into a pool below its depth floor. |
| Account | Seeds |
|---|---|
PoolConfig |
[b"pool", base_mint] |
RegimeState |
[b"regime", pool] |
ReserveVault |
[b"reserve", base_mint] |
Attestation |
[b"attest", pool, agent, epoch_le] |
The fee band is narrow by construction: expand shaves 5 bps and hedge adds
10, both clamped to 10..200 bps. The program never re-derives a verdict; the
kernel never reads chain state. docs/PROTOCOL.md is the normative reference and
viewer/src/solana.ts decodes the same accounts the program writes.
anchor build
anchor test
Requires Rust 1.74 or newer. The console additionally requires Node.js 18 or newer. No other toolchain is needed.
cargo build --release
cargo test
cargo run --release -- adjudicate samples/pool-stress.qf
The last command prints a risk report:
========================================================================
SSP RISK REPORT :: pool-stress
Question: Does SOL-BONK hold depth above $250k under a 15% drawdown?
========================================================================
Assertions: 5 Expand: 1 Hedge: 1 Watch: 2 Backstop: 1
Stability index: 58.7% Policy: expand>=0.66, stress<0.34
------------------------------------------------------------------------
[=] [p1] Pool depth stays above $250k through the drawdown.
consensus/affirmed topic=depth polarity=+1.00 depth=4.32 stress=0% cites=1
[!] [p2] A $10k sell moves price by less than 1.5%.
contested/negated topic=impact polarity=-0.24 depth=3.85 stress=38% cites=1
dissenting: lzr, mm1
...
cd viewer && npm install && npm run build && cd ..
cargo run --release -- adjudicate --format json samples/pool-stress.qf \
| node viewer/dist/cli.js -
cargo run --release -- adjudicate --format json samples/pool-stress.qf \
| node viewer/dist/cli.js --html - -o risk.html
The HTML renderer inlines all styling. No remote fonts, scripts or images, so the output opens offline and can be committed as a build artifact.
ssp <command> [options] <record-file | ->
COMMANDS
adjudicate Parse, normalize and print a risk report (text or json).
bundle Emit a digest-stamped stability attestation.
verify Re-derive an attestation digest and confirm it.
inspect Print the parsed, normalized epoch record as JSON.
help Print usage.
OPTIONS
--format <text|json> Report format for adjudicate. Default: text.
--consensus <0..1> Expansion threshold. Default: 0.66.
--dissent <0..1> Stress ceiling. Default: 0.34.
--min-mass <n> Minimum decisive depth to escape BACKSTOP.
-o, --output <path> Write to a file instead of stdout.
--json Treat stdin or extensionless input as JSON.
Exit codes: 0 success, 2 usage error, 3 parse or validation error, 4
digest mismatch. These make the tool usable as a CI gate.
src/ Rust risk kernel, standard library only
lib.rs public API and the end-to-end run
model.rs Agent, Claim, Position, Citation, Deliberation
json.rs compact idempotent JSON codec
parse.rs line-oriented .qf parser and JSON parser
normalize.rs assertion canonicalization and clustering
adjudicate.rs weighted regime engine and policy
amm.rs constant-product maths and depth targeting
bundle.rs deterministic digest-stamped attestations
report.rs text and JSON report renderers
programs/ssp-registry Anchor program: registry, regime state, reserve vault
viewer/ TypeScript operations console and web3 decoding
samples/ example epoch records
docs/PROTOCOL.md protocol reference
docs/EVIDENCE.md normative record format reference
Protocol vocabulary maps to code as follows.
| Protocol concept | Code |
|---|---|
| Liquidity agent and capital weight | Agent { weight } |
| Pool assertion | Claim |
| Attestation | Position { stance, confidence } |
| Liquidity regime | Outcome |
| Risk policy | Policy |
| Stability Index | Adjudication.cohesion |
| Stability attestation | Bundle and FNV-1a digest |
| Risk report | ssp.report.v1 |
| Operations console | ssp-view |
cargo test
cd viewer && npm test
make test
The Rust suite covers all four regimes, weight and confidence scaling, policy tuning, parser error paths, JSON round-trip stability including unicode escapes and surrogate pairs, attestation determinism and tamper detection. The console suite covers report validation, colored against plain output, HTML escaping, render determinism and on-chain account decoding.
std only. The console depends on
@solana/web3.js at runtime and on the TypeScript compiler at build time.f64 snapped to six decimals, not exact rationals.settle_regime and deploy_backstop.
Figure 8. Delivery plan. Kernel and console are done. On-chain work is in progress.
Two encodings are accepted and are information-equivalent. The line-oriented
.qf format is built for humans and version control: comments, one record per
line, readable diffs. The JSON format is built for machine generation. The full
grammar and schema are in docs/EVIDENCE.md.
MIT. See LICENSE and CHANGELOG.md.
SSP is liquidity infrastructure for Solana memecoin markets. It adjusts liquidity distribution, trading fees and capital allocation in response to volatility, pool depth and market conditions.