Telegram warning points: build a policy members can understand

A volunteer moderator remembers asking someone to stop promoting a service. The next moderator sees another promotional post but knows nothing about the earlier conversation. One gives a reminder; the other reaches for a ban. A warning system can make those decisions more consistent by recording points and connecting a known total to a known consequence.

The useful starting point is a rule you can explain to a member. “Three strikes and you are out” is incomplete when a single filter match can carry several points. You also need to say when the counter resets, which action the threshold triggers, and how someone can challenge a mistake. This guide focuses on that accounting; escalation design is the next step if one threshold is insufficient.

A warning has several separate parts#

Do not configure the system as if a filter match, a warning and a ban were the same event.

Relevant filter controls offer warning weights from 1 to 5. Three points can therefore mean three one-point events, a one-point event followed by a two-point event, or one three-point event. If the member-facing rule promises two chances before a restriction, a weight of three contradicts that promise.

Message deletion is another independent choice. A filter may remove a message without adding points, or issue a warning while leaving the message available. Seeing a post disappear does not establish that the author received a warning.

Write the rule before choosing the numbers#

Name an observable behavior: repeated unapproved promotion, an invitation prohibited by the group policy, or a specific matching rule. “Being disruptive” leaves too much room for moderators to interpret the same exchange differently.

Keep ambiguous cases available for human judgment. A member reporting a scam may quote the same advertisement that a scammer posts. A support request may contain a word on a prohibited list. A counter records the chosen signal; it does not establish the writer's intent. A doubtful signal is a poor reason to assign enough points for an immediate restriction.

For a hypothetical hobby community, a first policy could give one point per confirmed violation, apply a one-hour mute at three points, and clear the total 24 hours after its window starts. This is an example, not a universal recommendation. In a time-sensitive support community, preventing a legitimate question may carry a very different cost.

Configure the consequence in the Mini App#

Open Settings in your private chat with @defendy_bot, select My groups, and check the group name before changing anything.

1. Open Warnings and choose the basic, single-threshold mode.
2. Set the warning limit, action and action duration. The example above uses three points and a one-hour mute.
3. Choose the reset period. The current editor offers 1, 6, 12, 24, 48, 72, 168 and 720 hours.
4. Recheck the duration after selecting the action. Changing the punishment type resets that field to zero; for warning mutes and bans, zero represents an indefinite action.
5. Wait for the save confirmation, leave the section and reopen it to verify the saved values. These controls save changes as you make them; the screen is not a draft workspace.
6. In the relevant filter, select Warning and its weight. A stop-word list, for example, has separate controls for its action and message removal.

Avoid clicking through presets in a live group just to inspect them: choosing a preset replaces the tiers. A profile described as mild still includes a ban. Start with the policy you intend to use rather than assuming the least severe label contains no consequential actions.

The warning clock and the mute clock are different#

Defendy starts the counter's expiry window when the first points are added to an empty counter. Later warnings do not restart that window. The accumulated total expires together; each incident does not have its own gradually fading score.

Here is a hypothetical sequence with a clear counter, a 24-hour reset period, a threshold of three and a one-hour mute. All times use the same time zone.

This is why “warnings expire a day after a violation” is an inadequate explanation. It does not identify which violation starts the clock. Do not change the reset period midway through a test and assume already-running windows have been recalculated. Use a fresh counter when checking a new configuration.

Reaching the threshold also does not create a free pass for the rest of the window. Further points can trigger the final configured action again. A member returning after a short mute may still have active points.

Inspect and correct points deliberately#

For one person's case, reply to their message so the target is unambiguous:

• /warn repeated promotion adds one warning and may immediately trigger the configured consequence;
• /warns reports the current total;
• /unwarn subtracts one point;
• /resetwarns clears the counter and the record of applied tiers.

Your Defendy role must allow the command. See the moderation command reference for targeting and other supported forms.

The one-point correction matters when filters have larger weights. If a false positive added three points, one /unwarn removes only one. It does not locate the most recent incident and reverse its entire weight. Check the result with /warns; use a complete reset only if you intend to remove the person's other active points as well.

Neither correction automatically lifts a mute or ban that has already been issued. Review the restriction separately and use the appropriate recovery command. Conversely, unmuting someone does not erase their warning points. Treat access and accounting as two checks, even when they arose from the same mistake.

A test that distinguishes setup from wishful thinking#

Use a separate test group and a consenting ordinary member. Confirm the bot's moderation permissions and the moderator's command access. Testing only against the group owner is not a representative check of member behavior.

Test one addition below the threshold, an addition that reaches it exactly, a high-weight addition that reaches it in one step, and a correction afterward. Before repeating a scenario, remove any test restriction and clear the counter separately. Defendy's minimum temporary mute is 60 seconds, so a five-second mute is unsuitable as a quick test.

To check the reset window, select one hour in the test group, record the first addition, add another point later, and inspect the counter after the original window ends. Until that check has actually happened, the expiry path remains untested. These are recommended test steps, not claims that we performed them in a live group.

Make an appeal possible outside the restricted chat#

Publish a contact route a muted member can still use. Explain which behavior earns points and when the counter resets. Configure selected-event logging before you need to reconstruct a case: /warns is a total, not an incident history, and the log is not automatically an archive of every message.

When an appeal is justified, fix the matching rule as well as the points and restriction. Otherwise the same legitimate text can trigger the same chain again. If a message was removed, reversing a warning does not restore it. Ask the author to repost an acceptable version when that is useful, and tell the moderation team what changed so the next shift does not repeat the mistake.

Sources and verification#

Checked on 5 October 2026 against Defendy's current warning implementation, warning settings guide and command handlers. Telegram documents the distinction between individual and default chat permissions. A particular group's saved policy and actual outcomes still require verification; this article is not a report of a live deployment.