Skip to content
Closed
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 5 additions & 1 deletion home-mixer/candidate_hydrators/vf_candidate_hydrator.rs
Original file line number Diff line number Diff line change
Expand Up @@ -174,11 +174,15 @@ pub(crate) fn should_drop_ancillary(
false
}

/// Hard-drop only. Interstitial on a quote / RT / ancestor is intentional
/// keep-and-warn (Home NSFW rules emit Interstitial; Recs has Drop twins).
/// Do not treat Interstitial as Drop here. TES flag misses that skip those
/// rules are handled in visibility-filtering hydration (fail closed).
fn should_drop_reason(reason: &FilteredReason) -> bool {
match reason {
FilteredReason::SafetyResult(safety_result) => {
matches!(safety_result.action, Action::Drop(_))
}
_ => true,
_ => true,
}
}
7 changes: 7 additions & 0 deletions visibility-filtering/hydration/batch.rs
Original file line number Diff line number Diff line change
Expand Up @@ -25,6 +25,10 @@ impl<V> Hydrated<V> {
Hydrated::NotFound | Hydrated::Failed(_) => None,
}
}

pub(crate) fn is_failed(&self) -> bool {
matches!(self, Hydrated::Failed(_))
}
}

impl<V, E: Display> From<Result<Option<V>, E>> for Hydrated<V> {
Expand Down Expand Up @@ -194,12 +198,15 @@ mod tests {

assert_eq!(batch.get(&1), Some(&7));
assert_eq!(batch.hydrated(&2), Some(&Hydrated::NotFound));
assert!(!Hydrated::Found(7).is_failed());
assert!(!Hydrated::<u32>::NotFound.is_failed());
assert_eq!(
batch.hydrated(&3),
Some(&Hydrated::Failed(HydrationError::Rpc(
"backend unavailable".into()
)))
);
assert!(batch.hydrated(&3).is_some_and(Hydrated::is_failed));
assert_eq!(batch.get_or_default(&2), 0);
assert_eq!(batch.get_or_default(&3), 0);
}
Expand Down
3 changes: 2 additions & 1 deletion visibility-filtering/hydration/mod.rs
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,7 @@ use safety_label_hydrator::{SafetyLabelHydration, SafetyLabelHydrator};
use socialgraph_hydrator::SocialgraphHydrator;
use std::collections::HashMap;
use std::sync::Arc;
use tes_hydrator::TesHydrator;
use tes_hydrator::{retain_candidates_with_usable_tes_flags, TesHydrator};
use viewer_hydrator::ViewerHydrator;
use xai_core_entities::gizmoduck_client::GizmoduckClient;
use xai_core_entities::tweet_entity_service_client::TESClient;
Expand Down Expand Up @@ -207,6 +207,7 @@ impl HydrationPipeline {
label_response,
} = safety_labels;

let candidates = retain_candidates_with_usable_tes_flags(candidates, &tes_tweet_keyed);
let tweet_features = self.tes_hydrator.assemble_tweet_features(
&candidates,
&core_datas,
Expand Down
129 changes: 128 additions & 1 deletion visibility-filtering/hydration/tes_hydrator.rs
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
use crate::hydration::batch::TweetHydrationBatch;
use crate::hydration::batch::{Hydrated, TweetHydrationBatch};
use crate::hydration::metrics::{record_batch_size, timed_keyed_rpc, timed_results};
use crate::models::{
CoreFeature, MediaFeature, NsfwFeature, TweetCandidateInput, TweetFeatures, TweetId,
Expand Down Expand Up @@ -28,6 +28,37 @@ pub(crate) struct TweetHydration {
pub(crate) media: TweetHydrationBatch<MediaFeature>,
}

impl TweetHydration {
/// TES flag RPCs that decide Drop / Interstitial. A Failed read must not
/// be treated as "flag unset" — that fail-opens NSFW, takedown, nullcast,
/// and DMCA/geo media rules. Genuine NotFound (no flag) is not Failed.
pub(crate) fn safety_lookup_failed(&self, id: TweetId) -> bool {
self.nsfw_user
.hydrated(&id)
.is_some_and(Hydrated::is_failed)
|| self
.nsfw_admin
.hydrated(&id)
.is_some_and(Hydrated::is_failed)
|| self
.takedown_reasons
.hydrated(&id)
.is_some_and(Hydrated::is_failed)
|| self.nullcast.hydrated(&id).is_some_and(Hydrated::is_failed)
|| self.media.hydrated(&id).is_some_and(Hydrated::is_failed)
}
}

pub(crate) fn retain_candidates_with_usable_tes_flags(
candidates: Vec<TweetCandidateInput>,
tweet_keyed: &TweetHydration,
) -> Vec<TweetCandidateInput> {
candidates
.into_iter()
.filter(|c| !tweet_keyed.safety_lookup_failed(c.tweet_id))
.collect()
}

impl TesHydrator {
pub async fn fetch_pure_core(
&self,
Expand Down Expand Up @@ -474,4 +505,100 @@ mod tests {
assert!(f.core.text.is_empty());
assert!(!f.media.has_media);
}

fn failed_bool(id: u64) -> TweetHydrationBatch<bool> {
TweetHydrationBatch::from_results(
[TweetId(id)],
HashMap::from([(TweetId(id), Err::<Option<bool>, _>("tes unavailable"))]),
)
}

fn found_false(id: u64) -> TweetHydrationBatch<bool> {
found(id, false)
}

fn not_found_bool(id: u64) -> TweetHydrationBatch<bool> {
TweetHydrationBatch::from_results(
[TweetId(id)],
HashMap::from([(TweetId(id), Ok::<_, anyhow::Error>(None))]),
)
}

#[test]
fn safety_lookup_failed_is_false_when_flags_are_found_or_absent() {
let keyed = TweetHydration {
nsfw_user: found_false(10),
nsfw_admin: not_found_bool(10),
..Default::default()
};

assert!(!keyed.safety_lookup_failed(TweetId(10)));
assert!(!TweetHydration::default().safety_lookup_failed(TweetId(10)));
}

#[test]
fn failed_nsfw_user_lookup_is_safety_failure() {
let keyed = TweetHydration {
nsfw_user: failed_bool(10),
..Default::default()
};

assert!(keyed.safety_lookup_failed(TweetId(10)));
assert!(!keyed.safety_lookup_failed(TweetId(11)));
}

#[test]
fn failed_takedown_nullcast_or_media_lookup_is_safety_failure() {
let takedown = TweetHydration {
takedown_reasons: TweetHydrationBatch::from_results(
[TweetId(10)],
HashMap::from([(
TweetId(10),
Err::<Option<Vec<TakedownReason>>, _>("tes unavailable"),
)]),
),
..Default::default()
};
let nullcast = TweetHydration {
nullcast: failed_bool(10),
..Default::default()
};
let media = TweetHydration {
media: TweetHydrationBatch::from_results(
[TweetId(10)],
HashMap::from([(
TweetId(10),
Err::<Option<MediaFeature>, _>("tes unavailable"),
)]),
),
..Default::default()
};

assert!(takedown.safety_lookup_failed(TweetId(10)));
assert!(nullcast.safety_lookup_failed(TweetId(10)));
assert!(media.safety_lookup_failed(TweetId(10)));
}

#[test]
fn retain_drops_only_ids_whose_safety_flag_rpc_failed() {
let keyed = TweetHydration {
nsfw_user: TweetHydrationBatch::from_results(
[TweetId(10), TweetId(11)],
HashMap::from([
(TweetId(10), Err::<Option<bool>, _>("tes unavailable")),
(TweetId(11), Ok::<_, anyhow::Error>(Some(false))),
]),
),
..Default::default()
};
let kept = retain_candidates_with_usable_tes_flags(
vec![candidate(10, 100), candidate(11, 100)],
&keyed,
);

assert_eq!(
kept.iter().map(|c| c.tweet_id.0).collect::<Vec<_>>(),
vec![11]
);
}
}