Lucent Hit Rate Methodology

Lucent Hit Rate Methodology

This page documents how Lucent measures its own catalyst-coverage performance. The headline metric is the Timely Capture Rate on unscheduled catalysts. A descriptive read-through activity statistic and an internal unmatched-move-day count are surfaced alongside the rate as transparency aids, not as additional benchmarks.

Definitions — Timely Capture Rate (unscheduled catalysts)

For each clinical readout, SEC filing, approval, or capital raise (offering) in the trailing 90 days that actually occurred on a known exact date and was linked to a tracked biotech ticker: did Lucent first register the catalyst on or before 9:30 ET on the next NYSE trading day after the catalyst date? Multiple catalysts hitting the same ticker on the same date are collapsed to a single observation using the earliest surviving registration time so one market reaction is not double-counted. Wilson 95% confidence intervals are computed on the resulting numerator and denominator.

Example: a clinical readout posted at 5:00 PM ET on a Tuesday. If Lucent's first record of it was registered on or before 9:30 ET Wednesday, it counts as captured on time. If the next trading day is delayed by a holiday, the deadline shifts to 9:30 ET on the following NYSE session.

Universe

Lucent's tracked biotech ticker registry on the refresh date. 90-day trailing window anchored on the refresh timestamp. Catalysts without a linked ticker, without an exact known date, or that have not yet actually occurred are excluded from the denominator.

Counterfactual disclosure

The metric uses Lucent's ingestion timestamp — the moment a catalyst first entered Lucent's records — as the proxy for awareness. Live alert delivery to users may lag this by minutes; this metric does not measure user-perceived alert latency. Read it as Lucent's counterfactual capture timing against the next market session, not as a user-delivered alert benchmark.

Read-through activity volume

Not a rate. This page reports the volume of read-through alerts fired since Lucent's read-through alert system launched (2026-05-21), along with the median and mean absolute target-ticker move over the five trading days following each alert, plus a count of alerts whose target moved more than 2%. The central-tendency stats are computed over the subset of alerts whose 5-trading-day endpoint has arrived at compute time; alerts that fired too recently for the window to close are surfaced in the volume figure but excluded from the median, mean, and >2% count. The resolvable subset size is displayed alongside the volume so the gap is auditable. The underlying data is user-cohort-scoped — different platform usage mixes would produce different numbers. We publish this as activity volume, not a quality benchmark, and render no confidence interval to avoid implying one.

Filter (V1.4.0, 2026-06-22): alerts attributable to Lucent's curated backtest-universe synthetic user (system:backtest-universe-v1) are excluded from this rollup. Phase 25.12 closed a Phase 25.8 substrate-coupling leak in which the synthetic user's curated watchlist was being walked by the read-through evaluator as if it were a real pilot user, generating 1,109 alert_history rows between 2026-06-16 (Phase 25.8 ship) and 2026-06-22 (Phase 25.12 deploy) that entered earlier read-through volume and central-tendency stats. Phase 25.12 layers the fix at three sites for defense in depth: (a) the read-through evaluator filters synthetic-user pairs at the write path so no further synthetic-user rows are generated; (b) this page filters at the read path; (c) a one-shot cleanup script purged the 1,109 pre-fix rows as part of the v1.4.0 rotation. The descriptive numbers published under v1.3.0 or earlier (volume, median |Δ|, mean |Δ|, >2% count) should be read as inclusive of those synthetic rows for the 6-day window 2026-06-16 through 2026-06-22.

Unmatched move-day count

In the trailing 90 days, Lucent's internal audit scans daily price action across the full tracked-ticker registry and flags trading days with |1d Δ| > 5% that lack a matched catalyst within ±1 calendar day. We publish the count of such days. Biotech tickers commonly move more than 5% on sector rotation, macro tape, or momentum without a discrete catalyst, so this count is not a direct miss-rate — many of these days reflect price action that no catalyst-anchored system would capture as a discrete event. The number is the honest disclosure of the gap the catalyst-anchored coverage rate cannot speak to. The detailed audit log remains internal and feeds Lucent's source-expansion backlog.

Known gaps

• SEC 8-K and 6-K filings and news RSS are polled hourly during US market hours (13:00–22:00 UTC weekdays) by a dedicated intraday ingestion job; other event types (FDA approvals, ClinicalTrials.gov, FINRA short interest, earnings transcripts) remain on daily or weekly cadences. Capture timing for SEC and news catalysts that occur during market hours typically resolves within the hour rather than at the next daily close; capture timing for other event types is resolved at daily-close granularity.
• Conference abstract release coverage is incomplete prior to the ASCO 2026 backfill.
• Foreign Private Issuer annual filings (20-F) are tracked but excluded from this rate because they are scheduled, not unscheduled — issuers know months in advance when they are due, so on-time capture is not informative about coverage performance. 6-K (current-events) filings ARE included in the rate; the per-event-type breakdown distinguishes 8-Ks (event-driven material information) from 6-Ks (largely routine FPI quarterly updates) since the two carry materially different return-shape distributions.
• The read-through descriptive statistic reflects the current user mix and is not directly comparable across cohorts or time periods with different user composition.
• Catalysts first registered more than 7 days after the event date (typically historical back-ingestion) are excluded from the rate; the skip count is surfaced on the page so the universe remains auditable.
• Same-real-world-event multi-filing patterns — most commonly two SEC filings of the same form-type describing one real-world transaction (e.g. a 424B5 prospectus supplement co-filed with its own amendment for one capital raise), or multiple FDA approvals announcing different labels of one decision on the same date, or a clinical trial whose results page is amended same-day — are deduplicated at the substrate-materialization layer per the canonical earliest-createdAt tie-break. The per-event-type breakdown surfaces post-dedup cohort N; pre-V1.7.0 (rotation 2026-06-30) cohort N silently included these ghost rows. The offering / prospectus_supplement cohort N shifts from 27 to 24 at the 1d window as a result of the rotation (signed_mean +1.83% → +1.41%, abs_p50 2.96% → 2.85%); approval and clinical_readout cohorts also shift by smaller per-row margins. See the per-event-type breakdown for cohort-level numbers post-rotation.

Per-event-type breakdown

The headline Timely Capture Rate aggregates all event types. The per-type breakdown surfaces four catalyst families (clinical readouts, SEC filings, approvals, offerings), with SEC filings further partitioned by form subtype (8-K residual / 8-K Item 1.01 Material Definitive Agreement / 8-K Item 2.01 Completion of Acquisition / 8-K Item 2.02 earnings / 8-K Item 5.02 executive changes / 8-K Item 7.01 Reg FD / 6-K) since the six 8-K Item codes plus 6-K carry materially different return-shape distributions. Phase 24b (2026-06-25) added Items 1.01 + 2.01 at the front of the routing precedence chain to support partner-relation and acquirer_of-relation densification — see the routing rule pin in src/events/sec.ts's EIGHT_K_ITEM_PRIORITY.

Refresh

Weekly. Sundays at 10:00 ET. The methodology version hash is rendered in the page footer. Any change to the methodology body or to the underlying compute logic produces a new hash; Lucent's drift gate resets its baseline when the hash changes and annotates the reset in the underlying snapshot row.

Disclaimer

Historical / backtested. Counterfactual capture-time proxy, not user-delivered alert timing. Not client returns. Not investment advice.