Struct ThinkingSummary 

pub struct ThinkingSummary {
    pub scanned_qnames: usize,
    pub matched_for_query: usize,
    pub surfaced: usize,
    pub by_kind: BTreeMap<String, usize>,
    pub by_kind_dropped: BTreeMap<String, usize>,
    pub entries_in_workspace: Option<usize>,
}

Metadata accompanying the prior_thinking projection. Parallel in shape to FeedbackMetrics — same idea, different domain.

Token economy (1.0.78): every count field that’s zero is skipped during serialization, and the all-zero by_kind / by_kind_dropped maps are skipped entirely. On a query where no thinking exists, the serialized form collapses to {"scanned_qnames": N} (or empty if even scanned_qnames is 0), saving ~200 chars/call. Non-zero counts still emit — accuracy preserved.

Fields

scanned_qnames: usize

Number of qnames the caller passed in for scanning. Skipped when zero (no qnames provided → nothing to summarize).

matched_for_query: usize

Qnames that had at least one thinking entry (any kind). Skipped when zero; agent infers “no match” from absence.

surfaced: usize

Entries kept after kind-filter + confidence-floor cuts. Skipped when zero; agent infers “no entries surfaced” from absence (and from prior_thinking being null/absent).

by_kind: BTreeMap<String, usize>

Surfaced counts broken out by kind. Skipped entirely when every kind is zero. When emitted, ALL kinds appear (so the agent doesn’t have to defensive-check) — this is the established feedback_summary pattern.

by_kind_dropped: BTreeMap<String, usize>

Entries the kind/confidence filters dropped, broken out by kind. The load-bearing field for the “thinking exists but isn’t showing” case — when by_kind_dropped["hypothesis"] > 0 AND by_kind["hypothesis"] == 0, the agent knows hypotheses exist for these symbols but all fell below the confidence floor. Skipped when all-zero (most calls).

entries_in_workspace: Option<usize>

Workspace-wide thinking count (across ALL indexed symbols). Only populated when matched_for_query == 0 — gives the agent a signal whether to broaden the query (workspace has entries elsewhere) or capture new ones (workspace is genuinely empty). None otherwise to keep the typical hot path cheap.

Implementations

impl ThinkingSummary

pub fn is_quiet(&self) -> bool

True when the summary carries no signal — useful for the caller’s “emit at all?” decision per the sentinel rule (emit only when prior_thinking is null AND scanned_qnames > 0).

Trait Implementations

impl Clone for ThinkingSummary

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for ThinkingSummary

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Serialize for ThinkingSummary

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.