Reveal and Trace

XpressPurge never removes a post without recording why. Where that reason appears depends on your hide behaviour setting (Options → Settings → Hidden posts):

• Leave a one-line receipt — the reason appears in the feed, in place of the post.
• Remove from feed — the post is gone, and the reason is recorded under Recently removed on the Activity page (Options → Activity).

Either way the decision is reversible, and reversing it teaches the engine.

The strip appears in place of the post's content and includes:

• the primary reason for the decision, such as a rule or pack name;
• the strongest matching terms (up to three), when available;
• a Show button to reveal the post;
• a Not <topic> button when the hide came from a classifier, so you can correct it.

Crypto & Airdrops (airdrop, $SOL)        [Show] [Not crypto & airdrops]

The first time XpressPurge catches a post for you, the receipt carries a one-time "your first catch" note so you know the receipt is working as intended.

Click Show to dismiss the receipt and restore the post. It stays visible for the rest of the session.

After you tap Show, XpressPurge may offer one-tap correction follow-ups so the same hide is less likely to happen again:

• Hide less of <topic> — softens the pack or topic that caused the hide;
• Always show posts like this — protects similar posts going forward.

Click Not <topic> when a soft classifier hide is wrong. This:

• reveals the post;
• records a local soft label so the same post or similar posts are less likely to be hidden for the same reason;
• is used only for classifier decisions, not for your own explicit rules.

Labels are stored locally and never sent to the XpressPurge backend.

When posts are removed outright there is nothing left in the feed to carry an explanation, so the record lives under Recently removed on the Activity page instead. It lists what was removed, the reason, and — for Smart Feed decisions — how close the match was.

Wrong call on any entry restores the post, whitelists it so it is not removed again, and counts against whatever made the decision. A Smart Feed example that you overturn at least half the time, over four or more removals, is retired automatically.

When the hide was pack-driven, Help prevent this may appear after Wrong call. Local suggestions (for example switching a short substring term to whole-word matching) apply only after you review them. Optional AI Coach is separate and never blocks the restore.

• Trust. A wrongly removed post is one click from recovery, whichever hide behaviour you use.
• Auditability. Removal you cannot inspect is indistinguishable from a bug. Every decision keeps the evidence that produced it.
• Supervision. Corrections feed back into the local model, so the same mistake gets less likely rather than repeating.

Evidence is shown strongest-first, so the receipt always leads with the deciding term. For packs, the strongest matching term is the highest-weight hit. For rules, the matched values are shown.

When a post is hidden by one of your own rules, the receipt still shows the reason and a Show button, but it does not offer Not <topic>. A rule you wrote is a command, not a classifier guess, so correcting it means editing or disabling the rule.