A monitoring framework is the post-launch instrument panel for a token model: a short list of metrics, each carrying a target, a warning level, an action level, and a response that was authorized before anyone needed it. The metrics come out of the risk register the audit produced, so the team watches what was shown to break this model rather than what a dashboard happens to plot.
Price is the last thing to move. By the time it confirms a problem, the mechanism causing it has been misbehaving for hours or days, and every one of those hours was visible in a number nobody had assigned to anybody.
Derived from the register, not bought off a shelf
If the adversarial pass found six vectors capable of breaking the model, the framework carries at least one leading indicator per vector, chosen because it moves before the vector becomes an incident. That is the whole selection rule, and it is what separates a monitoring framework from an analytics page.
Total value locked on a chart is not monitoring. Total value locked with a floor beneath which the simulation showed rewards stop covering operator costs is monitoring, because the number now carries a mechanism and a consequence. The simulation is what converts a generic metric into a threshold somebody can be woken up for.
Three levels and a response that already exists
Each metric gets three values. Target is the state the model was designed for. Warning is the level at which a human looks and nothing else happens. Action is the level that fires a response decided in advance: a circuit breaker, a borrow pause, an emergency vote, a reserve drawdown.
Pre-authorization is the part that gets skipped, and it is the part that matters. Convening a discussion at the moment the threshold trips is usually too slow, because on-chain conditions move faster than a group chat reaches agreement. Decide the response while nothing is on fire, write down who is permitted to pull it, and rehearse it once before launch so the first execution is not the first attempt.
A worked threshold: health factor distribution
Aave's own risk documentation gives the formula. Health factor is the sum of each collateral position valued in ETH multiplied by its liquidation threshold, divided by total borrows in ETH, and when it falls below 1 the position may be liquidated to maintain solvency.1 That is the mechanism. The monitoring question is which statistic of it to watch.
Not the average, which hides the tail that liquidates. Track a low percentile across active loans, say the 10th, and set the warning where your own liquidation throughput model says the queue starts building. Set the action where liquidators can no longer clear positions before they go underwater. Both numbers come out of the simulation you already ran, not out of a benchmark somebody else published for a different collateral mix.
What to watch that is not price
Concentration: share of supply, stake, or pool liquidity held by the largest ten addresses, and the direction that set is moving. Privilege: every admin role granted or revoked, every upgrade queued, every timelocked transaction pending. Continuous contract surveillance is a discipline in its own right, covering admin key changes, timelock bypasses and dependency changes on live contracts.2 Oracle: deviation between your feed and an independent reference, plus time since the last update, which catches a stalled feed that is still returning a plausible number.
Then two the register usually supplies. Incentive health, measured as emissions paid per unit of the behavior you are actually buying, because farming shows up in that ratio long before it shows up anywhere else. And exit velocity, measured as redemptions or withdrawals per hour as a share of float, compared against the depth that would have to absorb them.
Reporting cadence is a design decision too
The internal framework and the external one are separate documents drawn from the same source. Supervisors treat ongoing crypto activity as a continuing reporting question rather than a single filing, and the BIS Financial Stability Institute's 2023 policy paper surveying how authorities across jurisdictions approach crypto, tokens and DeFi is the map worth reading before you decide what you publish and how often.3
Pick the cadence before launch and hold it. Across the 100+ projects we have advised, teams that publish on a fixed schedule get asked fewer questions during a bad week than teams that publish when the numbers look good, because the second pattern makes silence informative.
Common questions
What should a token project monitor after launch?
One leading indicator per risk the audit identified, plus a standing set: holder and stake concentration, every privileged action executed or queued, oracle deviation and staleness, emissions paid per unit of the behavior being incentivized, and exit velocity against available depth. Price belongs on the list but near the bottom, since it confirms problems the other metrics have already shown.
What is the difference between a warning threshold and an action threshold?
A warning threshold means a human reviews the situation and nothing automatic happens. An action threshold fires a response that was authorized during design, such as a circuit breaker, a borrow pause or a reserve drawdown. Keeping them separate stops the team from either ignoring early signals or triggering emergency machinery every time a metric wobbles inside its normal range.
How often should monitoring thresholds be reviewed?
Whenever the model changes and on a fixed calendar in between. A parameter change, a new market, a new integration or a change in collateral mix all invalidate thresholds derived from the previous simulation. Thresholds age at the speed of the assumptions behind them, so a framework left untouched for a year is measuring a protocol that no longer exists in that form.
See Tokenomics Audit Services for how this applies in practice.
Sources
- Risk Parameters, aave/risk-v3
Aave (GitHub), 2026
Aave's own risk documentation: loan to value, liquidation threshold, and the health factor formula with liquidation at Hf below 1. Read 3 August 2026 from the repository, since the rendered docs site returns an empty shell to a plain fetch. - Smart Contract Post-Deployment Monitoring
ChainScore Labs, 2026
Practitioner scope for continuous surveillance of live contracts, naming admin key changes, time-lock bypasses and library dependencies as monitored events. Read 3 August 2026. - FSI Insights No. 49: Crypto, tokens and DeFi: navigating the regulatory landscape
Bank for International Settlements, Financial Stability Institute, 2023
Garcia Ocampo, Branzoli and Cusmano, May 2023. Cross-jurisdiction survey of supervisory approaches, cited as the reference for setting a reporting cadence rather than for any single normative requirement.
Last reviewed 2026-08
Know the terms but not sure how they apply to your project? That is what an engagement is for. We design, document, and stress-test the whole token economy inside the Tokenomics Data Room.
100+ projects advised. Complete tokenomics in 4 to 6 weeks.