--- name: emerging-target-radar description: > Fuse literature/preprint velocity, scientific mindshare, and orthogonal data-engine convergence into a ranked watchlist of emerging target x modality candidates — each tagged with realized/expected literature-to-first-in-human lead time, the responsible NewCo-creation events, and which historical pattern it follows (genetics-first, modality-unlock, or resistance-ladder). Activate when scanning the discovery frontier for what is inflecting before consensus, building or refreshing an early-stage target watchlist, or sourcing candidate objects for downstream arc-placement and conviction scoring. metadata: author: nirav version: "1.0" parent: frontier-intelligence compatibility: Designed for Claude Code allowed-tools: Read, WebSearch, WebFetch --- # Emerging Target Radar — The Pre-Consensus Watchlist Integrator The single most valuable thing a frontier investor or a clinical scientist entering a new field can know is *what is inflecting before everyone agrees it is*. Scientific mindshare accreting to an emerging target or modality is measurable, with lead times running from ~5-7 years (NIH grant flows) down to near-real-time (conference share-of-voice). The radar's job is to fuse those signals into one ranked, defensible watchlist instead of chasing whatever was loudest at the last JPM. This skill is the **integrator**. It does not generate raw velocity, attention, or convergence signals itself — three upstream skills do that (signal-scanner, mindshare-tracker, data-generation-monitor). The radar consumes their outputs and produces a single ranked list of `{target × modality}` candidates, each carrying its **lead-time estimate** (the literature-inflection-to-first-in-human gap, typically 3-7 years for a novel target), its **NewCo-creation events** (the sharpest venture signal there is), and its **historical-pattern tag** (genetics-first like PCSK9/GPR75, modality-unlock like KRAS/VAV1, or resistance-ladder like BTK). These candidate objects are the input contract for the downstream analog engine (moa-analog-engine, modality-lifecycle, frontier-conviction-scorer). The core discipline is anti-hype. Mindshare is reflexive — capital manufactures the attention it claims to detect — so a candidate's rank is a *positioning/timing* instrument, never a measure of biological merit. Every velocity signal must be gated by an evidence-maturity check and normalized against the financing regime (Q1 2025 was the lowest US biotech startup formation in a decade, ~70% off the 2021 peak, so a single NewCo means more in 2025 than it did in 2021). > **Watchlist:** the dated target×modality candidate roster is versioned separately in `references/watchlist.md` (a point-in-time snapshot to refresh on cadence); the scoring method stays here. ## How to Run ### Input | Parameter | Source | Required? | |---|---|---| | Scope (therapeutic area, modality, or "broad frontier scan") | User | Yes | | Signal-scanner output (velocity + acceleration + breadth per entity) | signal-scanner skill / WebSearch | Recommended | | Mindshare-tracker output (0-100 momentum + Consensus evidence gate) | mindshare-tracker skill / WebSearch | Recommended | | Data-generation-monitor output (orthogonal-convergence counts) | data-generation-monitor skill | Recommended | | Time window for velocity (e.g. trailing 3 yr) | User | Optional (default: 3 yr) | | Existing watchlist to refresh | User / Read | Optional | | Financing-regime normalization year | User | Optional (default: current) | If the three upstream signal streams are unavailable, reconstruct them inline: query PubMed E-utilities for `total_count` velocity, bioRxiv/medRxiv for preprint lead, conference abstract databases for share-of-voice, and EDGAR/press/Crunchbase for NewCo events. The skill degrades gracefully but is strongest when fed clean upstream objects. ### Steps #### Step 1 — Assemble the Candidate Universe Pull every `{target × modality}` pair surfacing in the three signal streams within scope. Do not pre-filter on merit — the universe is deliberately broad; ranking happens later. Seed sources to query directly when upstream streams are thin: - **PubMed E-utilities** — `total_count` per `target[tiab] AND modality[tiab]` per year; compute YoY velocity and the second derivative (acceleration). Strip review/perspective publication types before counting. - **bioRxiv / medRxiv** — preprint volume per entity (leads peer-reviewed by 6-18 months); the earliest formal-literature signal. - **Open Targets** — genetics-to-target associations; flags genetics-first candidates. - **ClinicalTrials.gov v2** — first IND/Phase-1 entry is the *translation-event ground truth* that closes the lead-time measurement. - **ChEMBL** — is there tractable chemical matter yet (IC50/EC50/Ki), or is this still a tool-compound gap? #### Step 2 — Fuse the Three Signal Streams per Candidate For each candidate, combine the upstream signals into a fused signal profile. The streams are deliberately orthogonal — they catch different false positives: ``` FUSED SIGNAL PROFILE — {target × modality} Velocity (signal-scanner): accel of PubMed total_count + bioRxiv preprint vol; author-affiliation breadth (Herfindahl — high concentration = single-lab artifact, discount) Attention (mindshare-tracker): 0-100 momentum from conference share-of-voice (AACR/ASCO/ASH/ESMO/JPM); GATED by Consensus evidence-maturity (volume high + quality low = HYPE flag) Convergence (data-gen-monitor): count of net-new ORTHOGONAL data-engine convergences (cis-pQTL+MR AND selective DepMap dependency AND clean Perturb-seq AND druggable ChEMBL matter AND cell-type-restricted expression) ``` A candidate with high velocity but a HYPE flag and zero orthogonal convergences is froth. A candidate with moderate velocity, multi-lab breadth, and 3+ convergences is a durable inflection. Rank rewards the second, not the first. #### Step 3 — Estimate Lead Time (Literature-Inflection → First-in-Human) For each candidate, locate the literature-velocity inflection year and the first-in-human (FIH) entry, or project the FIH if not yet reached. ``` LEAD-TIME ESTIMATE Inflection year = year the second derivative of publication/preprint volume turned sharply positive FIH year = first Phase-1 entry on ClinicalTrials.gov v2 (REALIZED) — or projected from arc position Realized lead = FIH year − Inflection year (only if FIH has occurred) Expected lead = 3-7 yr from inflection for a novel target (use 3 yr if tool compound + chemical matter already exist; 7 yr if still target-ID/biology stage with no tractable handle) ``` The realized/expected lead time is what tells an investor *how much runway remains before consensus prices it* and tells a scientist *how mature the field is*. A candidate already past FIH with a short realized lead (PCSK9 ~12 yr target-to-approval) is late-stage radar; a candidate at tool-compound stage with a 5-7 yr expected lead is the earliest actionable read. #### Step 4 — Capture NewCo-Creation Events (the Sharpest Venture Signal) A venture builder spinning out a NewCo around a *specific* target/modality front-runs consensus by 1-2 years — far more informative than aggregate sector financing. Record each event and **normalize against the financing regime**: ``` NEWCO SIGNAL Events: [company, round/$ , date, builder (Flagship/ARCH/etc.), specific target×modality] Big-pharma deal proxy: large licensing/M&A on the exact mechanism counts as a confirmation NewCo-equivalent (e.g. Novartis × Monte Rosa VAV1 degrader up to $2.1B; Sanofi × Vigil TREM2 $470M+; AbbVie × Capstan ~$2.1B) Regime normalization: weight each event UP in a contractionary year. Q1 2025 = decade-low US startup formation (~70% off 2021 peak), so a 2025 NewCo carries more conviction than a 2021 one. ``` #### Step 5 — Tag the Historical Pattern Assign each candidate exactly one of three named patterns (the analog engine keys off this tag). The pattern predicts *how* the class will explode and *what* the ignition event will be: ``` PATTERN TAG genetics-first — human-genetics (esp. LoF-protective) validation came BEFORE the drug. Ignition = CV/hard-outcome trial. Analogs: PCSK9, GPR75, INHBE/ALK7, TL1A. modality-unlock — a flat/disordered/pocketless target made tractable by a CHEMISTRY/PLATFORM fix, not new biology. Analogs: KRAS (covalent switch-II), VAV1 (molecular glue), in vivo CAR (tLNP), siRNA targets (GalNAc). Ignition = first-in-class pivotal readout. resistance-ladder — each new sub-class triggered by a defined resistance mutation. Analogs: BTK (ibrutinib → acalabrutinib → pirtobrutinib C481S → BTK degraders). Ignition = the resistance mechanism itself is the leading indicator of the next rung. ``` #### Step 6 — Rank and Emit the Watchlist Score each candidate and sort. The composite rank is a positioning score, not a PoS: ``` RADAR RANK (0-100) = 0.35 × fused_signal_strength (velocity accel + breadth + convergence count, gated by evidence-maturity) + 0.25 × newco_signal (count × regime-normalization weight) + 0.25 × lead_time_actionability (more remaining runway before consensus = higher; penalize already-crowded) + 0.15 × pattern_clarity (clean genetics-first or post-ignition modality-unlock ranks above ambiguous) HARD GATES (apply before scoring): — HYPE flag (high volume, low Consensus evidence quality) → cap rank at 40, route to "watch, don't underwrite" — Single-lab artifact (Herfindahl breadth fail) → cap rank at 40 — "AI-discovered target" claim with no independent wet-lab/clinical validation → discount; AI moving molecular DESIGN is credible, AI "finding a novel target" usually is not (targets were already implicated) ``` ### Output ``` EMERGING TARGET RADAR — {Scope} Date: [assessment date] | Velocity window: [window] | Regime year: [year] RANKED WATCHLIST | Rank | Target × Modality | Pattern Tag | Arc Position | Lead Time (R/E) | NewCo / Deal Signal | Fused Signal | Evidence Gate | |------|-------------------|-------------|--------------|-----------------|---------------------|--------------|---------------| | 1 | RAS(ON) tri-cplx | modality-unlock | FIH→1st-appr (Ph3) | realized ~8 yr | RevMed (public) | accel↑↑ / conv 4 | PASS | | 2 | GPR75/INHBE × siRNA | genetics-first | tool→FIH | expected 4-6 yr | Arrowhead/Regeneron | accel↑ / conv 5 | PASS | | ... | | | | | | | | CANDIDATE OBJECTS (handoff to analog engine) — one per row: { target, modality, indication, arc_position, pattern_tag, lead_time, newco_events, supporting_signals: {velocity, mindshare, convergence}, evidence_gate, radar_rank } TOP-OF-RADAR NARRATIVE: 1. [Highest-conviction pre-consensus read and why now] 2. [Steepest adoption curve this cycle] 3. [The one to "watch, don't underwrite" — high mindshare, evidence gap] DEPRIORITIZED (gated out): - [Candidate] — [HYPE flag / single-lab / AI-novelty discount] ``` ## Error Handling | Scenario | Response | |---|---| | Upstream signal streams unavailable | Reconstruct inline via PubMed E-utilities, bioRxiv/medRxiv, conference abstracts, EDGAR/press — flag as lower-fidelity | | High velocity but no convergence and HYPE flag | Hard-gate to "watch, don't underwrite"; do not let loud literature inflate rank | | Single-lab dominates the literature (Herfindahl fail) | Discount as artifact (>50% of preclinical findings irreproducible); require multi-lab breadth before ranking | | No FIH yet (no ClinicalTrials.gov entry) | Use *expected* lead time from arc position (3 yr if tractable chemical matter exists, 7 yr if still biology-stage) | | NewCo events sparse | Check big-pharma licensing/M&A on the exact mechanism as a confirmation-equivalent; do not treat absence as a kill | | "AI-discovered target" claim | Down-weight unless AI acted in molecular *design* with independent validation; AI does not yet move Phase 2/3 odds | | Crowded/post-explosion class | Lower lead-time actionability; flag that alpha has likely left (convenience/cadence reformulation = maturity) | | Contractionary financing year | Up-weight each NewCo event; one 2025 spinout > one 2021 spinout for conviction purposes | ## Cross-Domain Connections This skill is explicitly **dual-use**: a clinical-scientist entering a field reads the watchlist as a map of *where the science is inflecting and how mature each frontier is*; an investor reads the identical output as a *screen of early opportunities with remaining runway before consensus pricing*. Lead-time and pattern-tag serve both readers. - **frontier-intelligence/signal-scanner** *(depends_on)*: supplies the velocity/acceleration/breadth primitive the radar fuses — the literature inflection that starts the lead-time clock - **frontier-intelligence/mindshare-tracker** *(depends_on)*: supplies the 0-100 attention momentum and the Consensus evidence-maturity gate that powers the anti-hype hard gate - **frontier-intelligence/data-generation-monitor** *(depends_on)*: supplies the orthogonal-convergence count — the strongest durable-merit signal that separates inflection from froth - **frontier-intelligence/moa-analog-engine**: primary consumer — takes each candidate object, finds its nearest validated analog, and names the still-pending ignition trial to watch - **frontier-intelligence/modality-lifecycle**: consumes the modality half of each candidate to supply the P(modality deliverable) term and confirm the delivery unlock has landed - **frontier-intelligence/frontier-conviction-scorer**: ultimate consumer — converts the ranked candidate into a decomposed conviction score, handing off to phase-weighted rNPV once a clinical asset exists - **competitive-intelligence/pipeline-mapper** *(depends_on)*: once a candidate crosses into the clinic, pipeline-mapper enumerates the competitive set and confirms the "pre-explosion crowding" signal (e.g. every major immunology pharma owning a TL1A asset) - **product/frontier-antenna** *(depends_on)*: shares the cross-domain frontier-scanning methodology; the radar is the biotech-specialized instance feeding product-level frontier signals - **neocortex/foresight** *(depends_on)*: the radar's lead-time + pattern-tag output is a structured foresight signal — second-derivative detection of an emerging trend before consensus