Northstar discovery benchmark
intelligence
Prompt
You are allocating scarce research time across the 100 stocks below. Select the FIVE situations most worth investigating for possible long-only trades over the next 2–8 weeks. You may select fewer only if you explain why additional research would have little value. You are selecting investigations, not trades. Use your own economic reasoning to develop hypotheses from these observations. Do not just restate the strongest signals, largest moves, or supplied headlines. Explain the transmission mechanism and why resolving an uncertainty could reveal an actionable opportunity. A lack of proof of mispricing is not itself a reason to discard an investigation. Conversely, an imaginative story without a concrete observation and a discriminating research question does not earn a place. All companies and records are fictional. Use only this packet for case-specific facts. General business/economic knowledge is allowed; label its application as an inference and its prerequisites as assumptions. Do not browse, invent facts, or claim to have conducted research. No prescribed winning symbols exist. Research budget: five initial investigations, up to 45 minutes each, using public filings, operating disclosures, and market data. You need not spend all slots. Your aim is the expected usefulness of an investigation, not certainty that its thesis is true. No valuation, stop, or confirmation requirement blocks shortlisting. OUTPUT (maximum 1,000 words) 1. Briefly identify the most relevant cross-universe patterns. 2. Rank your shortlist. For each symbol give: - the exact observations that earned attention (cite row ID and context ID); - a causal opportunity hypothesis, including direction and horizon; - the strongest ordinary/adverse explanation of those same observations; - the FIRST decisive question to research, the source to inspect, and how different findings would advance or terminate this investigation; - why that question deserves a slot ahead of other candidates. 3. Name three tempting alternatives you deprioritized, with specific reasons. Discuss overlap between selected cases if they share the same underlying driver. 4. State the most consequential limitation of your shortlist. DATA CONTRACT - Synthetic snapshot: 03 Sep 2026 before market open. All stock prices/volumes refer to completed 02 Sep sessions; there is NO partial-day volume here. - R1/R5/R20: percent price returns over 1/5/20 trading sessions, adjusted for splits; dividends excluded. Sector context uses the same 20-session window and a broader external peer universe, not just the ten shown per sector. - V: latest completed-session volume / preceding 20-session median volume. - E0/E1: percent change over SEVEN CALENDAR DAYS in consensus EPS for the same current/next annual fiscal period of that company. They are estimate REVISIONS, not EPS growth rates or earnings surprises. Fiscal calendars can differ across companies; no period rollover or share-basis change occurs within these windows. - N: number of analysts contributing to the current FY0 consensus. FY1 contributor counts and revision breadth are unavailable. '?' means unavailable, never zero. - Signals are descriptive scanner labels, not scores, confirmations, or recommendations. - Headlines summarize reported claims; they do not supply independently researched economics. '-' in Reported item means no headline supplied, not no real event. - The horizons of R20 and E0/E1 differ. Price/revision disagreement is a lead, not proof of same-window divergence or causation. Valuations are not supplied. - No outcome after the snapshot is provided or permissible evidence. COMMON CONTEXT [M1] Sector codes (ID prefixes) and median R20: BNK=Banks -3%; IT=IT services -5%; PH=Pharmaceuticals +2%; EN=Energy -6%; CH=Chemicals -2%; IN=Industrials +3%; CO=Consumer -1%; TR=Transport -4%; ME=Metals -7%; UT=Utilities +1%. These describe peers, not estimated market betas. [M2] Over the same 20 sessions: USD crude -12%; USD/INR +3% (INR weakened); container spot freight rates -9%. Individual hedges, contracts, and exposures unknown. [M3] Domestic 10-year government yield fell 25 basis points over those sessions. Borrowing-cost pass-through, credit spreads, and company debt structures unknown. [M4] Broad equity index R20 -2%. No universal market explanation is established. UNIVERSE (each row ID is its symbol; all 100 must be considered) |ID|Business|R1%|R5%|R20%|V|E0%|E1%|N|Signal|Reported item| |---|---|---:|---:|---:|---:|---:|---:|---:|---|---| |EN07|lubricants producer|-0.1|-2.9|-7.5|0.8|-1.8|-2.3|9|recovery|-| |UT10|district energy supplier|-0.8|-0.5|1.1|1.14|0.4|-2.2|5|none|-| |BNK02|retail lender|0.0|-1.9|-1.4|1.4|-0.3|3.6|6|pullback|-| |BNK09|deposit-led lender|-0.2|-2.5|-3.7|0.92|-2.1|-2.1|15|none|-| |IN01|power equipment|1.1|4.0|8|1.7|3|6|14|range breakout|01 Sep: issuer reports firm orders up 25% year-on-year; advances, execution schedule, and margins absent.| |IN07|defence components|1.4|-0.8|3.9|1.04|0.7|-0.6|6|pullback|-| |ME10|aluminium fabricator|0.3|-1.8|-5.9|1.31|-2.1|-3.0|19|none|-| |CH02|specialty chemicals|1.5|2.5|-3.7|0.7|0.2|2.0|13|none|-| |TR03|road freight|1.0|1.6|-7.2|1.47|2.3|-4.2|7|pullback|-| |UT04|electricity transmission|-0.7|-0.5|0.4|1.13|-0.5|-1.2|19|none|-| |IN03|rail components|-0.5|1.4|9.4|0.65|-2.4|-0.3|7|none|-| |IT10|small IT consultancy|2.2|6.0|9.0|2.1|200.0|?|1|EPS revision|Single-analyst FY0 EPS estimate changed from INR 0.10 to INR 0.30 in seven calendar days.| |TR06|rail freight|-0.5|-2.2|-6.7|1.18|0.4|1.3|15|none|-| |CH10|contract synthesis|0.4|-1.0|-4.7|0.77|-2.3|-3.9|17|none|-| |TR08|travel operator|-0.3|-2.7|-3.4|1.1|0.5|0.7|10|recovery|-| |IN05|construction equipment|-1.1|1.1|5.3|1.0|-0.3|0.6|16|recovery|-| |IN09|electrical cables|1.0|-1.5|7.2|1.28|-1.3|-0.7|10|none|-| |UT02|thermal generator|1.4|-1.0|-0.8|1.07|0.7|-5.8|14|pullback|-| |ME03|copper smelter|0.7|-3.4|-7.4|1.45|-3.2|-2.2|16|recovery|-| |IT06|healthcare IT services|-0.6|-4.4|-8.5|0.82|-0.8|-1.4|13|recovery|-| |IT04|banking software services|-1.9|-1.7|-7.0|0.74|0.7|-0.9|12|pullback|-| |PH03|API producer|-1.9|4.6|0.6|1.22|2.3|-0.6|9|none|-| |UT09|solar equipment operator|0.7|1.5|-3.5|0.84|-1.2|-3.1|8|none|-| |CH06|dyes manufacturer|-0.5|-3.4|-1.4|1.44|2.1|-0.1|5|recovery|-| |IT09|cybersecurity services|-0.1|-1.6|-2.5|0.77|-0.5|-0.2|13|none|-| |EN02|oil refiner|0.2|2.5|-3.2|1.25|-1.1|-0.2|6|recovery|-| |EN06|integrated oil company|1.0|-1.4|-6.5|1.42|1.8|3.1|14|none|-| |EN04|gas distributor|-0.7|-1.7|-6.0|0.66|0.9|3.4|6|recovery|-| |PH02|domestic branded drugs|-1.1|-1.8|7.5|1.21|-3.8|-2.3|8|recovery|-| |TR10|logistics warehousing|-0.3|0.1|1.2|0.89|-0.3|0.3|18|none|-| |EN10|fuel logistics|-1.1|-3.0|-10.1|1.45|3.0|2.4|12|pullback|-| |EN05|offshore services|0.6|-0.7|-4.6|1.01|2.2|0.2|7|none|-| |CO09|personal care|-1.1|-2.3|0.3|1.29|0.1|2.6|8|none|-| |ME01|steel producer|-1.0|-2.4|-4.4|1.31|3.6|-4.8|14|recovery|-| |CO04|premium apparel|-1.7|-0.1|-2.5|1.21|-0.6|-0.4|9|pullback|-| |CO08|jewellery retailer|-0.0|0.7|-1.7|1.02|1.3|-2.7|12|none|-| |PH01|generic drug exporter|-2.1|-7.8|-6.0|2.8|0|0|6|none|02 Sep: regulator lists observations at one plant; affected revenue, severity, and remediation timeline absent.| |TR02|container shipping|-1|-4|-10|1.03|-5|-8|18|recovery|Issuer carries container freight; contract/spot revenue split absent.| |CH08|basic petrochemicals|0.1|1.8|-3.2|0.89|0.8|1.5|18|none|-| |UT03|renewable developer|2|7|15|2.0|0|1|6|range breakout|02 Sep: issuer announces capacity ambitions; financing terms and contracted offtake absent.| |ME06|steel tubes|-1.0|-0.9|-1.0|0.9|-1.8|0.6|11|none|-| |CO06|quick-service restaurants|0.4|1.2|0.6|1.24|-1.4|0.5|16|none|-| |PH08|biologics manufacturer|-0.3|-2.7|1.6|1.23|1.0|1.7|14|none|-| |ME02|aluminium producer|1.3|-6.1|-12.1|0.76|1.1|2.3|14|none|-| |PH06|hospital injectables|1.9|3.5|-0.3|0.95|2.3|-2.4|19|none|-| |TR04|airport operator|-2.6|0.5|1.1|0.74|1.0|0.3|15|pullback|-| |CH01|solvent producer|-1.0|-4|-9|1.27|0.5|1|8|none|Issuer describes feedstock as crude-linked; output-price pass-through and inventory cycle absent.| |ME04|iron ore miner|1|3|2|1.4|-1|0|16|none|02 Sep: provider reports a competitor mine outage; duration, substitutability, and ME04 spare capacity absent.| |TR01|passenger airline|-0.5|-2|-8|0.74|0|0.5|9|none|02 Sep: issuer reports unchanged passenger load factor; yields, fuel hedges, and currency exposure absent.| |IN10|small engineering contractor|-0.6|-1.5|-0.2|0.89|1.2|-1.5|7|none|-| |CO03|discount retailer|-0.4|-1.4|2.8|0.71|4.0|0.3|15|none|-| |EN09|oilfield equipment|0.6|-1.4|-7.9|1.34|-0.2|-0.1|16|recovery|-| |BNK06|diversified lender|1.5|-1.9|-7.9|0.92|1.2|-0.7|14|none|-| |BNK08|rural lender|-2.5|0.1|-5.8|0.85|-0.5|-1.7|6|pullback|-| |UT08|merchant power producer|1.4|2.2|-2.2|1.17|-0.5|-3.0|11|pullback|-| |EN03|fuel retailer|-0.7|-2|-7|1.19|1|2|10|none|02 Sep: issuer reports stable retail fuel volumes; retail repricing and inventory effects absent.| |CO02|home appliances|-1.0|-4|-8|1.0|-4|3|12|recovery|02 Sep: issuer announces channel inventory reduction; consumer sell-through data absent.| |BNK07|consumer lender|0.1|-3.0|-4.9|1.23|0.1|-0.4|7|none|-| |EN01|oil producer|-0.9|-4.2|-10.0|1.41|-5|-7|6|recovery|No issuer explanation supplied; business is upstream oil production.| |CH05|agrochemicals|0.8|-1.6|-3.3|0.75|0.8|3.5|17|none|-| |IT01|large outsourcing vendor|-0.8|-3.0|-9.7|1.39|-1.9|0.4|15|none|-| |BNK03|regional lender|-1.2|-5.4|-10.2|1.7|1.1|2.0|19|none|02 Sep: issuer says deposit growth accelerated; net-interest-margin and funding-cost details absent.| |CH03|commodity resins|0.6|3.0|-0.6|0.88|0.5|1.3|17|pullback|-| |PH07|consumer healthcare|-1.1|-1.6|7.3|1.18|-0.9|-4.6|16|none|-| |IT02|cloud integrator|0.4|1.5|4.0|1.25|2.0|4.0|15|none|02 Sep: issuer announces a 3-year contract renewal; value and incremental revenue undisclosed.| |CH07|industrial adhesives|0.9|-4.0|-1.1|0.84|-0.2|2.1|18|recovery|-| |IT07|enterprise software services|3.4|8.8|14.0|2.4|-2.0|-4.0|7|range breakout|02 Sep: provider headline says AI partnership; revenue commitments absent.| |BNK01|corporate lender|-0.3|-3.8|3.0|1.5|-2.6|2.3|15|recovery|-| |PH04|contract manufacturer|1.3|0.9|-0.7|1.3|0.6|-2.3|9|recovery|-| |ME05|zinc producer|0.4|-7.5|-5.9|0.94|2.7|3.7|14|none|-| |CO01|packaged food|1.3|-0.2|0.6|0.96|-0.1|-1.0|19|none|-| |TR09|air cargo|-1.4|0.6|-2.4|1.06|3.4|3.3|18|pullback|-| |UT05|hydropower generator|0.2|1.9|0.8|1.28|2.9|2.2|15|none|-| |CH09|fluorochemicals|-0.6|3.8|-2.3|0.74|1.7|0.2|11|none|-| |CO07|beverages|-0.7|1.7|-0.7|0.79|0.4|1.2|9|none|-| |PH09|diagnostic reagents|-0.8|-1.5|0.5|1.48|-0.6|-3.6|19|recovery|-| |CO10|footwear retailer|-0.5|2.0|-1.1|1.12|1.3|1.0|9|recovery|-| |ME07|metal recycler|-0.2|-3.9|-3.8|1.34|1.7|0.8|7|none|-| |UT01|regulated electricity distributor|-0.5|-2|-6|1.13|-1|-1|19|none|02 Sep: regulator opens a tariff review; proposed recovery, timing, and allowed-return effects absent.| |BNK10|infrastructure lender|-1.3|-1.1|-7.8|1.48|-1.4|1.1|16|recovery|-| |IN04|industrial automation|-0.1|1.1|4.8|1.34|2.1|2.0|16|recovery|-| |UT07|water utility|-0.7|0.5|3.3|1.1|-1.4|-1.6|14|pullback|-| |IT03|engineering services|-1.2|-0.5|-3.4|1.17|5.1|4.2|8|pullback|-| |PH10|small generic exporter|0.2|4.1|1.4|0.83|-5.5|0.2|12|none|-| |ME08|specialty alloys|-1.1|-3.7|-1.6|1.24|0.9|-1.3|12|recovery|-| |TR07|express parcel delivery|0.6|0.9|-1.3|1.0|-1.7|-3.2|14|none|-| |CO05|household products|1.2|-1.5|2.7|1.12|-1.3|-0.4|11|none|-| |ME09|copper miner|0.1|-4.1|-8.9|0.7|3.1|-1.1|16|none|-| |BNK04|mortgage lender|0.5|-1.0|0.5|1.39|2.6|3.1|10|pullback|-| |IN08|industrial pumps|-1.1|0.8|1.7|1.09|-0.9|4.0|10|none|-| |EN08|gas producer|-0.0|-1.6|-7.1|1.44|0.4|0.4|14|none|-| |IT08|offshore outsourcing|0.3|-1.4|-3.1|0.9|0.0|1.2|7|none|-| |IN02|capital equipment|4.0|10|18|4.8|0|0|19|range breakout|Confirmed index addition announced 28 Aug, effective 04 Sep close; no earnings update supplied.| |IN06|precision bearings|0.2|2.0|-1.2|1.13|3.1|-2.9|14|pullback|-| |IT05|IT staffing|0.6|-5.6|-7.3|0.93|-1.7|-2.4|17|none|-| |UT06|gas transmission|-0.1|0.9|7.4|0.75|-0.7|-2.7|13|none|-| |PH05|specialty drug exporter|2.4|6.2|12.5|1.8|1|2|10|range breakout|02 Sep article repeats product approval first announced 05 Aug; launch timing and competition absent.| |BNK05|SME lender|-0.0|-0.0|-6.5|0.84|2.4|-1.1|16|none|-| |TR05|port operator|2.1|-1.2|-5.2|1.4|-2.3|-1.0|16|none|-| |CH04|industrial gases|-1.7|1.9|-5.8|1.1|-0.2|-3.0|11|none|-|
Answer guidance
What this measures The task is research allocation across 100 fictional stocks, using report-shaped evidence and general economic reasoning. It measures hypothesis formation, prioritization, numerical attribution, and the quality of proposed investigation. It does not measure browsing, source verification in practice, trade execution, profitability, or full performance on Northstar's real reports. Paste only `discovery-100-prompt.md` into MicroEvals. This guide must not be part of the evaluated model's input. The fixture is synthetic, not an exported market snapshot. Values were constructed for an exercise, not calibrated to a market return distribution. Some rows have company-specific headlines; others have no headline supplied. This can favor well-described cases, so assess whether the model still considers the broader price/estimate/sector evidence. ## Scoring (100 points) Score each dimension from 0 to 4 and multiply by weight / 4. Intermediate scores 1 and 3 represent performance between the stated anchors. Apply the rubric to the whole answer, considering how consistently the behavior appears across cases. These are starting weights, not empirically validated predictors of usefulness. | Dimension | Weight | 0 | 2 | 4 | |---|---:|---|---|---| | Evidence fidelity | 20 | Material invented or misattributed facts | Mostly correct; some unsupported comparisons | Correct symbols, values, windows, units, and missing-data interpretation | | Economic hypothesis | 25 | Repeats signals/moves or invents a story | Plausible mechanism with weak prerequisites or magnitude discussion | Evidence-linked mechanism with explicit conditions, countereffects, and fit to horizon | | Discriminating research | 25 | Generic “research financials/news” | Relevant question but vague source or decision consequence | Feasible first question, appropriate source, and different findings that advance/stop the case | | Research allocation | 20 | Mechanical ranking or no rationale | Some comparisons but weak marginal-value reasoning | Explains why these questions beat alternatives and handles shared-driver overlap | | Judgment under uncertainty | 10 | Treats hypotheses as proven, or rejects all because data is incomplete | Acknowledges uncertainty but acts inconsistently | Separates promising investigations from validated trades; applies fair competing explanations | Record material fabricated facts, cross-stock substitutions, percentage/basis-point errors, and treatment of partial versus completed sessions separately as critical errors. A case whose main justification depends on such an error is unsupported; do not let polished prose conceal it in the aggregate score. Do not impose an arbitrary universal pass threshold from this one fixture. Different selected symbols can earn the same high score. Agreement with the notes below is not itself a scoring dimension. Neither novelty nor contrarianism earns credit without an observation, a mechanism, and a useful test. ## Case-specific reviewer notes — illustrative, not a target shortlist - **BNK03:** Underperformance despite reported deposit acceleration is worth explaining. Deposit volume does not establish cheap funding, margin improvement, or good asset quality. First look at deposit mix/cost and margin disclosures; credit quality could dominate. Do not call relative underperformance a valuation. - **IT02:** Strength relative to a weak sector and upward estimates could support resilience or differentiated demand. A renewal may merely preserve existing revenue. Contract incremental value, bookings/conversion, and client concentration distinguish the stories. Currency can also affect exporters; exposure is unknown. - **IT07:** A price/estimate disagreement around an AI headline might deserve verification, but does not establish either irrational enthusiasm or hidden earnings upside. Binding commercial commitments differ from a partnership label. - **IT10:** A 200% revision is a rise of INR 0.20 from INR 0.10 to INR 0.30, based on one FY0 analyst. It is not broad confirmation or an EPS growth rate. A model can investigate it, but must justify the value of that slot. - **PH01:** Plant observations may create a temporary dislocation or reveal a serious earnings impairment. Read the regulator notice and map the plant to products/revenue and remediation. No fine, shutdown, or severity is established. - **PH05:** The article is recent; the approval is old. Do not call the approval new information. Launch economics or competitive change could still justify investigation if framed as unknown rather than asserted. - **EN01:** Lower crude is a plausible headwind to an upstream producer. An oversold/recovery label alone supplies little causal edge. Currency, realization, costs, and policy can alter the effect; simple commodity beta is an alternative. - **EN03 / CH01 / TR01:** Falling crude can affect fuel retail, feedstock costs, and airline fuel expense differently. Retail price resets, inventory losses, product-price pass-through, hedges, fares, and INR weakness can offset apparent benefits. Favor precise questions over blanket “oil down means profit up.” Multiple selections can be justified, but these may overlap in research exposure. - **IN01:** Firm orders, revisions, and price strength may point to unfinished repricing; confirmation does not require contrarianism. Order economics, timing, funding/advances, conversion, and what valuation already discounts remain unknown. Much of an order's earnings contribution may fall outside 2–8 weeks; near-term information-driven repricing is a hypothesis to explain, not assume. - **IN02:** Index inclusion offers an observable alternative explanation for volume and price. Its effective date is future but the announcement is already known. A flow-based investigation can be valid if mechanics, remaining demand, and post-event reversal risk are the actual research questions. Mechanical does not mean automatically unattractive; it also does not imply operating improvement. - **CO02:** FY0 revisions down and FY1 revisions up are different horizons, not contradictory data. Inventory reduction can mean normalization or weak demand. Consumer sell-through, channel inventory, and the basis of next-year estimates are relevant discriminators. Do not claim estimate breadth from analyst count. - **TR02:** Lower freight rates can offset cheaper fuel for a shipping company. Revenue contract exposure and bunker-fuel pass-through matter. Do not transfer an airline's input-cost hypothesis unchanged to shipping. - **ME04:** Strength relative to a weak sector plus a reported competitor outage gives a potentially coherent supply hypothesis. Verify the outage and duration, substitutability, commodity exposure, and ability to capture higher prices/volume. Relative performance is not proof of causality or a guaranteed remaining edge. - **UT01:** A tariff review could alter recovery or returns, but the direction is absent. Lower benchmark yields do not establish company refinancing benefits. The regulator's proposal and decision timetable are natural first sources. - **UT03:** Capacity ambitions differ from funded, contracted projects. Financing, offtake, returns, and the relevance of lower rates need investigation. Strong price action alone neither proves the story nor makes it exhausted. - **Other rows:** These are eligible selections. A good model may infer a useful sector/price/estimate question without a supplied headline. Require the same evidential standards; do not penalize a model for departing from this list. ## Useful manual comparison procedure 1. Run the identical prompt across models; record date, model version, reasoning setting, response length, latency, and cost when the service exposes them. 2. Hide model names while grading. Score the full answer and note critical errors. A second human reviewer helps reveal subjective ranking differences. An LLM judge is optional and should not be the sole authority for this subjective task. 3. Compare with simple baselines: largest absolute R20 moves, largest positive E0 revisions, and scanner-highlighted names. Merely choosing different symbols is not success; look for better mechanisms and higher-value research questions. 4. Repeat with the same rows reordered. A shortlist can legitimately change near ties. Large unexplained swings, changed numbers, or lost key conditions matter more than exact top-five overlap. Do not claim actual all-100 consideration can be proven from a five-name answer alone. 5. Optionally repeat with the Signal column removed while retaining underlying facts. This probes label dependence; it is a changed-input condition, not strict invariance, because some labels carry otherwise absent structural information. 6. Keep this initial fixture unchanged while comparing models. Add independent fixtures later; repeated tuning on this one exercise would weaken the conclusion. ## Implementation boundary No runtime harness, tools, upstream LLM, or reporting-code change is needed for this prompt. Its structure could later be generated from Northstar's data using ordinary code, with missing inputs preserved. The current synthetic fixture assumes clean fixed forecast periods and completed sessions; real exports must verify those properties or explicitly report uncertainty instead. Company news and common market context would need the relevant source data, not invented fills. No model run or score is included in these files. This is a ready-to-run prompt and a proposed grading procedure, not evidence that any model performs well.
