Research · Today’s picks

How our daily picks are produced

We combine independent sports forecasts, available sportsbook prices and sourced research to build a list of potential bets to review before deciding. The process can return no bets. All four sports remain experimental for betting decisions; a completed review does not establish a profitable edge.

Policy: morning-edition-7 · All displayed times use U.S. Eastern Time (EST in standard time, EDT during daylight saving). Recurring schedules use ET. This page is generated from the workflow configuration and checked for drift in pull requests.

The simple flow: Refresh sports data and prices → run the models → research the slate independently → combine the candidate pools → check the price and sourced context → publish up to 10 reviewed ideas → after the 4:30 p.m. refresh, price-check those picks and add new afternoon picks → you decide and track.

When the work runs

These are scheduled starts, not guaranteed publication times. Jobs can queue, feeds can fail and books can post markets later. A research job waits for the relevant sports refresh when triggered by that refresh; it does not assume that a scheduled job has finished. The quote timestamp in the picks table is the sportsbook’s timestamp, never the page refresh time.

NHL and MLB refresh at 7:05 a.m. and 4:30 p.m. Eastern. The morning workflow also refreshes NFL, waits for the sports jobs, then runs research. Recovery starts at 7:35, 8:05 and 8:35 a.m. check for a completed edition first; if one exists they skip data pulls and paid research and check delivery. Otherwise they refresh NHL and MLB again but reuse an NFL refresh less than 45 minutes old, because NFL quotes stay valid for 90 minutes at publication. Since September 29, an independent Supabase database timer has requested these four morning starts on time, with the existing completed-edition and spending checks; GitHub’s own schedule, which missed the September 28 morning window, remains a backup. The same timer starts the 4:30 p.m. refresh, because GitHub’s schedule started it after 7 p.m. from September 29 to October 4. A GitHub backup at 4:45 p.m. skips any sport that already refreshed that afternoon and does nothing after midnight. Each board publishes when its refresh finishes, usually within minutes of the start; a start time is not a publication time. When both afternoon refreshes finish, the afternoon picks edition runs (below); it typically publishes between 4:40 and 5 p.m., and never after 7 p.m. Existing NFL game-day updates and the separate price rundown continue. After all three sports refresh jobs finish, the workflow separately starts editorial writing and Top Picks research, and records Pinnacle’s prices for the events in our feeds for later evaluation. From October 24, 2026, each morning pick shows Pinnacle’s no-vig chance at the same line as a sharp reference, or that Pinnacle had not posted that line when asked; Pinnacle never chooses, ranks or reprices a pick. Editorial targets two articles by 8:30 a.m. Eastern; this is a target, not a guarantee. Automatic articles require successful model checks after 7:05 a.m. and within 90 minutes, plus their existing quote, history and reporting checks. Missing inputs wait rather than produce stale analysis. Before publication a separate audit checks every draft against its sources and prices; a rejected draft gets one revision against the audit’s findings and is published only if a fresh audit passes. All our models are experimental: any article that uses one must say so and that it has not been validated against betting prices. NHL articles use the experimental NHL forecast; NBA is not written about automatically. Editorial recovery runs only before noon and does not repeat the sports pulls. Hourly maintenance continues for approved posts and freshness; it is not hourly article production.

The morning card preserves its original analysis and prices, including after games start. It is clearly dated and is not continuously reassessed. The home page features the first pick on that card whose game has not started, or, once every game has started, the card’s first pick marked as started; it reads the same edition, so it never adds a research-pool offer or a different order, and on days with no picks the slot is hidden. See NFL, MLB, NHL and NBA for newer offers. Missing morning markets can appear on these pages later. Today’s Picks can show one sportsbook at a time (the choice lasts for the Eastern day); the filter only hides rows and changes no pick or ranking. NBA player props joined the card with policy morning-edition-7 (from November 2, 2026). Their prices and forecasts come from the separate NBA refresh at 6:45 a.m., which finishes before the 7:05 a.m. morning pipeline starts; NBA game lines are not eligible.

Completion and delivery: An empty completed card means no reviewed offers qualified. Failed feeds or research are labeled “Research incomplete,” with per-sport assessment counts and reasons saved in the card. Recovery attempts may complete the work; they never reset the daily budget. Until the day’s edition publishes, Today’s Picks and the homepage say it is not published yet; the previous day’s card is not carried over and remains in its dated archive. The workflow checks that the exact edition is publicly reachable before reporting success. Once that check passes, the site owner receives a private email copy of the card; it changes nothing on the card and places no bets. Late runs outside the morning window and unverified delivery fail visibly in the workflow history.

The afternoon edition (since October 19, 2026)

After the 4:30 p.m. NHL and MLB refresh, a separate afternoon edition (policy afternoon-edition-3 from November 2, 2026, which admits NBA props when fresh NBA prices are published; afternoon-edition-2 before it; afternoon-edition-1 had no game or player cap) does two things. It never rewrites the morning card.

• Price check of the morning picks. Each morning pick is checked at the best book now posting its exact outcome and line, under the same rule that put it on the card: at least two books post the line, the model gives the side at least the market’s chance, and at least 1% expected value after blending. Each pick then shows whether it is still a pick at that price, no longer a pick (with the price it would now need), or that its line moved. The check is recorded once with its time and stays with the pick for the evening; it is not a new review. It costs nothing: the prices are the ones the afternoon refresh already bought.
• New afternoon picks. The same selection runs on the 4:30 p.m. prices for NHL and MLB (NFL prices are not refreshed that afternoon, and NBA’s evening refresh starts at 5 p.m., so NBA props enter only when NBA prices under two hours old are already published). Markets already on the morning card are left out at any line or side, and so is any side the morning review passed on; sides the morning review left waiting on a fact (a goalie, a lineup) are reviewed first, then new candidates by expected value. Close to the start the market has absorbed most news, so the afternoon runs no discovery search and no targeted searches; reviews use the official reporting the review step already collects. Reviews are paid only from what the morning left of the same daily allowance; an afternoon with nothing left publishes no new picks and says so, which is not a failure. Up to 10 picks, at most 4 per sport, 2 per sport and market, 2 per game and 1 per player in a game, with the same three-hour review rule. They appear in their own table under the morning card, and the owner gets an afternoon email when there is a new pick or a morning pick no longer qualifies.

Afternoon picks are recorded as their own dated edition and graded beside the morning picks in the weekly recaps, under their own policy version, so the two can be compared.

The order inside a successful run

1. Refresh sports data and prices. The sports job retrieves its configured statistics, schedule and authorized odds feeds, retains stable identities and quote times, and archives inputs. This is not a claim that every injury, lineup or weather observation is a numerical model input.
2. Run the sport’s model pipeline. Current inputs produce independent quantity forecasts and outcome probabilities where supported. NHL uses versioned trained models; MLB refreshes its cached daily training bundle as needed; NFL refreshes player/history estimates and its scoring projections. NHL live forecasts use retrieved, validated final results from completed dates before today Eastern, including the previous night before 8 a.m.; the historical 7 a.m. EST / 8 a.m. EDT availability assumption is kept only in reconstructed evidence. Each NHL board refresh checks model and goalie history again. MLB checks final games through yesterday Eastern and rebuilds its daily bundle when the actual observations change, even on the same date. NFL uses completed earlier weeks. NBA forecasts (experimental model nba-v1.0) are published by the separate 6:45 a.m. and 5 p.m. NBA refreshes from ESPN box scores through the previous night, and withhold players ESPN lists as out. Missing models stay missing. The data and model update finishes before the review job consumes its published board.
3. Research the slate independently. A separate AI-assisted search sees the available games and market types, without our model predictions or rankings. It starts with that day’s published expert picks, best bets and prop projections from the allowed outlets, each tagged with its outlet and the line as published, then looks for its own opportunity, matchup and participation questions. A published pick is an opinion that raises a question; it is never treated as verification of an edge. Suggestions must resolve to real current offers with the exact game, player, market and side; the offer supplies the line, book, price and time. A suggestion that breaks a format rule is set aside on its own; the run’s other suggestions are kept. Search suggestions are unverified leads until the underlying dated reporting is retrieved. Each run makes at most one bounded discovery request, with games submitted and source coverage recorded separately. A large slate may not fit; unsubmitted games have not received this independent search. The later candidate-review stage uses multiple batches. A game being included in a search request does not mean every market or report was researched.
4. Build the union of candidates. Keep offers passing the sport’s quantitative screen and, since October 5, 2026, the market blend described below, plus independently researched offers. Neither route must agree with the other. Model quality gates, quote freshness and exact identities apply. The best available price is selected for an identical outcome and line. Since October 19, 2026, each research lead enters as one bet: its market’s main line, the line the most books post, which at least two books must post, at the best price there (books with verified settlement rules first). Alternate lines are not separate leads, and a lead that matches only alternate lines is withheld. When research leads take both sides of one market, for example two outlets’ opposite spread picks, they cancel: neither enters the research pool or the review queue. Coverage counts both withholdings. Earlier research pools listed every matching line and book, and both sides. The pool then keeps one offer per market (game, player and market): never both sides, whether two research leads or a research lead against a model pick, and never several lines, alternate thresholds or settlement terms of one bet. The highest-ranked offer stays, and coverage counts the rest. Leans skip markets already listed. There is no four-pick quota and no one-bet-per-game rule.
5. Collect evidence and investigate the case. Retrieve dated league/team reporting, injury information and relevant search leads. Before review, the strongest candidates in queue order (six per run, two per batch, so each sport’s leaders are covered) also get one targeted web search each about that exact bet: injury and practice status for the player, teammates and opponents, role or lineup changes, weather, and published analyst views for and against, contrary evidence first. The search only supplies links; each trusted, dated page is fetched and the review must quote that fetched text. Check expected opportunity, starters, lineups, opponent effects, weather and settlement where evidence is available. Compare the projected quantity with the offered line, the model probability with the actual price, and the offered threshold with other books. Follow-up questions and unresolved source failures are retained. Professional betting recommendations are opinions, not verified support by themselves.
6. Assess and rank. Each completed assessment says consider, wait for a specific material fact, or pass, with its case and countercase. Current consider assessments precede pending/wait/pass rows across all sports, so completed MLB or NHL opportunities remain visible above unreviewed NFL rows. The research pool retains within-sport rankings; the published card uses the common ranking described below. Related bets are labeled as shared exposure. No joint-bet probabilities are obtained by multiplying marginal chances.
7. Build a manageable daily card. Publish up to 10 reviewed ideas, aiming for five to ten when the evidence supports them, with no minimum. At publication, only exact current consider assessments no older than three hours or explicit analyst selections qualify. Explicit analyst selections take priority. Consolidate alternate thresholds and equivalent bets, never carry both sides of one market, and take at most one bet per player in a game (a player’s points and assists, or hits and total bases, ride on the same minutes and role); keep their original forecasts intact. The research pool never shows the opposite side of a card pick. Wait, pass, unreviewed and changed-price assessments stay in the separate collapsed research pool. Additional reviewed opportunities also remain there. At most two bets in one game reach the card, and they carry shared-exposure labels. This card limit is a workflow choice, not evidence of profitability.
8. Publish the review record. Preserve the original model, offer, source times, review, exclusions and budget usage. The table shows the actual book’s line and odds, quote time, model and market comparison, review status and tracker action. An independent research candidate without an eligible model says “Not established”; it receives no invented probability or EV.
9. Keep the edition and track decisions. Save an immutable, dated card with original offers, forecasts and reviews. Later odds updates cannot erase it or silently approve a changed price. The research pool remains separate and continues to check freshness. A repeated morning run cannot replace a completed morning edition or spend again unless an operator explicitly requests replacement, and a morning edition cannot be published in the afternoon window. A test edition does not complete or block the normal morning run; its archive and spending are retained. Incomplete research can recover during the morning window under the same daily budget. Manual afternoon tests are labeled test editions. Bets can be logged in Bet Tracker using the actual executed price; the site does not place wagers. Tracked bets settle automatically from final box scores: a check runs every 30 minutes and settles a bet once its game is final and started at least 4 hours ago. Bets it cannot match with certainty stay pending.

Player statistics alongside each pick

On the sport boards, a player’s name opens a snapshot (hover with a mouse, or select it) that pairs the saved model projection (its average and, for NHL, MLB and NBA, its median) with recent games, workload and relevant forecast inputs. Today’s Picks shows the same panel within each pick’s research. Observed averages are separate from prior-adjusted estimates. NHL panels show production and ice time; MLB shows innings, strikeout rates or batting opportunity; NFL passing panels show recorded components and recency weights. NBA panels show the model average and median with recent games and minutes. Snapshots also explain the model: a chart of recent games against the line with the model’s recency weights, the math behind the projection, the chance of each outcome, opponent strength with league rank, how much comes from the player’s own games, what the model does not know and, for MLB and NHL, a published track record from games the model never saw. Each snapshot retains its sample and dates; missing or older context is labeled, and later results never rewrite a published assessment.

Contributor articles require manual publication

Reader suggestions and other articles submitted through the private Editorial Desk remain private until an editor reviews a complete saved draft, chooses Publish saved article, and confirms that version. Requesting research, saving text, switching to Opinion, or selecting homepage placement never authorizes publication. Analysis and Opinion use the same Generate draft, review, rewrite and manual Publish controls. Opinion research checks factual claims against current reporting without requiring betting lines or a model forecast. The hourly publisher only delivers explicitly approved versions and honors their earliest publication dates. Editing an approved version withdraws its approval. This is separate from the automatic daily analysis articles described above.

The formulas and selection rules

Let d be decimal odds, p the unconditional chance of winning and u the chance of a refunded push. Then loss probability is 1 − p − u.

• Expected return per dollar: EV = p × (d − 1) − (1 − p − u) = p × d + u − 1.
• Fair decimal price: d_fair = (1 − u) / p. A price’s break-even probability, conditional on no push, is 1 / d.
• Two-way de-vig: for paired opposite prices, q_i = (1 / d_i) / [(1 / d_1) + (1 / d_2)]. Only matching identities, thresholds, settlement and timestamp windows can be compared. A median line is not the book’s expected outcome.
• Market blend (since October 5, 2026): with m the model’s chance and c the median no-vig chance across books at the exact line, both conditional on no push, the final chance is f = sigmoid(w × logit(m) + (1 − w) × logit(c)). A model candidate qualifies only if at least two books post that exact line, counting the offered book, the model’s chance is at least the market’s (m ≥ c), and (1 − u) × (f × d − 1) ≥ 1% at its own price. An NFL calibrated estimate of exactly 50% is the calibration curve’s flat centre, so it is withheld as no information.
• Model weight w by sport (since policy morning-edition-5): 0.10 for NHL, NFL player props, NBA player props (from morning-edition-7) and MLB moneylines, run lines and totals; 0.25 for MLB player props. Editions from October 5 used 0.25 for every sport and keep that weight in their records. The weights come from a fit on graded lines on October 14, 2026: across 5,112 NHL lines (86 games) and 567 NFL prop lines (weeks 3–5) the best weight was close to zero, and 0.10 is near the top of its uncertainty range. MLB player props were the only market where the model beat the market, over six games, so they keep 0.25 until more games are graded. A model below the market is withheld even at a long price: a better price alone is a Market Watch signal, not a pick. Pinnacle prices are recorded each morning for evaluation only; they are not yet part of c.
• Confidence tiers (policy morning-edition-3): a pick whose blended expected value is 3% or more is labeled high confidence; from 1% to 3% it is moderate confidence. The label says how much room the price leaves if the model is too optimistic; it is not a chance of winning, and neither tier is a validated edge. Under morning-edition-2 the only bar was 3% and MLB and NHL needed two books besides the offered one; together these left the October 6 card empty. The 1% and 3% bars are starting values, and the weights will be refit as graded lines accumulate.

The only weighting is the model-and-market blend above, with its fixed weights by sport; qualitative confidence is never added to it. The high and moderate confidence labels come only from the blended expected value, and research does not change them. Research evaluates assumptions and price sensitivity; it never manufactures a confidence percentage or changes the original model probability, which is kept beside the blended one. Tracked bets record the model, market and blended chances, the blend weight, expected value at the price taken and the decision time. Market Watch remains the separate price-comparison feature.

Leans: each sport’s best offer that is still positive after the blend but below the 1% floor is listed under Today’s Picks as a lean, for interest only. Leans are not picks, not ranked onto the card and have no tracker button.

Prediction-market observations

The separate Kalshi contract board shows dated prices, available quantity, estimated fees and exact settlement rules. An operator starts each initial refresh: series metadata and fees, contracts, event fee overrides and order books are collected before the snapshot is published. There is no scheduled refresh or promised publication time yet. Quotes older than 15 minutes are historical; market close is not game start. A private paper simulator requires quotes and fees within 30 seconds, verified future game starts and unchanged contract rules. It records simulated fills, not real trades.

Exchange contracts do not enter Today's Picks or Market Watch in this release. A sportsbook comparison requires an exact verified outcome and matching settlement rules; a 50-cent tie settlement is not a sportsbook push. Missing model probabilities remain unknown, and prior research results do not validate this new integration. Paper-only spending limits are separate from the research budget and do not authorize real-money orders.

Arbitrage scans (since October 22, 2026)

Each NFL, NBA, NHL and MLB refresh above saves its prices, then scans those saved quotes for arbitrage before it publishes, so results appear when that refresh publishes, not at its scheduled start. The scan makes no requests and uses one engine for every league: a pair is locked only when both quotes were taken within five minutes of each other before the game, at two books whose settlement rules for that market are on file and match; other pairs under 100% are listed as checks. Each league also keeps a paper ledger of every pair seen. Editors can press Scan now on the finder to rescan current game lines for a league (about 3 Odds API credits) or one game with props (up to 7), twice that with the optional second group of books. Those presses share the Live Odds credit reserve and its 60-second reuse. Arbitrage pairs never enter Today’s Picks or Market Watch, and the one-time history replay on the arbitrage plan is a paper exercise, not evidence for any pick policy.

Freshness, coverage and cost

NFL quotes expire after 90 minutes. MLB quotes and model checks expire after 90 minutes. NBA quotes, forecasts and the NBA refresh itself expire after two hours, which covers the morning recovery starts after the 6:45 a.m. NBA refresh. NHL Top Picks quotes and forecasts expire after 30 minutes, with model inputs checked within 36 hours. These checks apply when publishing and to the live research pool. Started games, future-dated quotes and failed model feeds cannot create new picks. Published editions retain their original timestamps and label started games as historical assessments. The research pool checks expiry every 30 seconds and reloads feeds every five minutes.

All Top Picks discovery and review share one fixed allowance per Eastern calendar day. It includes search and review requests, legacy same-day charges and conservative reservations for failures with unknown usage. The full allowance is available to morning research; there is no reserved afternoon allowance. The afternoon edition reviews only with what the morning left, and runs no discovery or targeted searches. Unspent amounts do not roll over. Discovery and reviews use Anthropic’s Claude Opus 5.5 at high reasoning effort under policy morning-edition-7. Earlier editions used OpenAI’s GPT-6.1 Sol (from October 1, 2026) and, before that, GPT-6 Astra at low reasoning; every review records the model that wrote it, and earlier editions keep their original policy version in results. Opus costs more per token than Sol and high effort spends more output on reasoning, but per-request ceilings and this allowance still bound every run. Existing odds-feed costs and the separate daily-article writer are outside this research allowance. No new paid subscriptions are enabled. A request is skipped if its conservative maximum cannot fit; this can leave money unused and candidates awaiting review.

Research batches rotate across sports. Independent discovery ideas receive an early review turn; explicit replacement runs prioritize previously reviewed offers with changed facts or prices. Within each group, the strongest screened case (the highest expected value after the market blend) is reviewed first, and kickoff time only breaks ties; offers in the lower-reliability research tier follow. This schedules research without assigning a higher win probability to any idea. The main card holds at most 10 reviewed ideas. The separate collapsed research pool initially shows 20 additional offers, with a button to show all; all eligible candidates remain available for research. The batch size controls API payload size. The batch limit decides how many candidates are reviewed, and only reviewed candidates can reach the card; strongest-first ordering keeps the limit from skipping a strong later game for a weaker earlier one. Budget and source limits can leave incomplete research. That coverage is recorded explicitly. No system can promise to examine every possible market exhaustively within a fixed allowance, and no daily picks minimum is imposed.

One-day release test: The owner authorized a larger shared research ceiling for September 27, 2026 only, including spending already recorded that day. It expired automatically at midnight Eastern; the usual daily allowance resumed September 28. This does not change the maximum of ten reviewed ideas on the main card.

Ordering the daily card: After the completed-review gate, explicit analyst selections lead, followed by forecast reliability, then high-confidence before moderate-confidence picks, then an expected-log-growth score on a common scale across sports. Since October 5, 2026 the score uses the blended chance for every sport; the calculation uses a fixed 0.0025 fraction and is an ordering convention, not staking advice. Missing independent models receive no numerical score. For variety without forced picks, each sport’s best eligible bet is taken first, then the rest in order, with at most four bets per sport, two per sport and market, two per game and one per player in a game. A sport appears only when one of its bets qualifies. This ordering has not been validated as a profitable selection strategy.

Each research run reviews at most twelve batches of up to three candidates (36 candidates). Additional candidates remain in the research pool; reaching this limit is recorded separately from exhausting the daily budget. Each returned review is checked on its own against the reviewer’s rules: field lengths, cited excerpts that appear word for word in their sources, and no percentage figures or certainty wording in the reviewer’s own prose. A review that breaks a rule is dropped and the reason recorded; its candidate stays unreviewed and cannot reach the card, while the batch’s other reviews stand. The edition counts as incomplete only when a whole batch fails. Until October 13, 2026, one broken review discarded its entire batch.

Weekly recaps by sport

MLB Monday · NHL Tuesday · NFL Wednesday · NBA Thursday. MLB, NHL and NBA each review the seven completed Eastern calendar dates before their publication day. NFL reviews the completed football week and previews the next one.

Closing-line value report card (from October 20, 2026). Every published pick, morning and afternoon, is priced again in the final minutes before its game: the median no-vig chance across US books at the pick’s exact line. Each recap reports, by edition, how much more the published price paid than that closing chance justified (value at the close), how often that was positive (beat the close) and how far the market moved toward or away from the pick after it was published, for the week and for every pick since tracking began. Positive value at the close over many picks is the earliest reliable sign of an edge, well before win–loss results settle down; it is a measure of price, not a guarantee. A pick whose line moved has no same-line close and is counted separately. Taking the close costs about one odds credit per pick’s market per game and shares the snapshot with any tracked bet on that game. NBA picks begin with policy morning-edition-7 (from November 2, 2026); earlier NBA recaps show none.

The existing morning pipeline is the primary delivery path. After data and card work finish, a recap check looks for each sport's expected dated article and matching review data. Successful complete reports are skipped. Daily 10 a.m., noon and 6 p.m. Eastern checks recover missing reports; waiting states back off for two hours. Unresolved published reports can be updated once per day for the next two days; explicit operator reruns can incorporate later corrections. NFL also has Wednesday 10 a.m., noon and 6 p.m. starts and can recover missing weekly work during Wednesday/Thursday morning refreshes. These are scheduled starts, not guaranteed publication times.

Published morning picks are graded at their exact offered price, using the final morning edition for each date. Policy versions stay separate. A separate full-board comparison uses each game's last saved pregame snapshot and the most-offered paired line. Missing player participation and missing statistics remain unresolved. Model and market probabilities use matched, non-push outcomes and game-level uncertainty intervals. MLB and NBA official result responses are saved for reproducibility.

Coverage is stated in every recap. MLB full-board archives began October 7, 2026; older saved morning cards can still be graded. The October 12 report will have partial board coverage for October 7–11. NHL already has durable pregame and results archives. NBA now saves durable snapshots and has a Thursday grader; it reports waiting for evidence or results until games can be graded. No current forecast reconstructs an earlier prediction. Report failures remain visible in the workflow and delivery status; one sport's failure does not discard another's completed report.

What remains to prove

NBA player props joined the card under morning-edition-7 by the owner’s decision, before the proposed readiness bar of 150 graded games. Through October 31, 2026 the NBA shadow grade covered 88 games: the model alone trailed the market on every market, and the 10% blend beat the market on props by about 0.001 in log loss, with a 95% interval just below zero. Problems with the NBA feed or its reviews are recorded as notes and do not mark an edition incomplete. The market blend’s 0.25 weight and its 1% and 3% bars are not yet fitted; Phase 4 of the model plan scores raw model, market and blend out of sample by market. Moderate-confidence picks have not been shown to be profitable and should be graded separately from high-confidence ones. Editions from policy morning-edition-6 cap the card at two bets per game and one per player in a game; the October 20, 2026 edition under morning-edition-5 carried four correlated overs on two players in one game. Editions with the October 14 blend weights carry morning-edition-5; editions researched by Claude Opus 5.5 with strongest-first review order and twelve batches carry policy morning-edition-4; editions with the confidence tiers before that carry morning-edition-3; editions from October 5, 2026 carry morning-edition-2 and earlier ones morning-edition-1. Earlier editions are not regraded as evidence for a later policy. New independent discovery and review decisions are prospective shadow records. We have not shown that this larger candidate pool, new ordering or qualitative intervention improves returns. Grade original forecasts and reviewed selections separately, retain correlated exposure and execution prices, and assess forecast calibration, returns, drawdown and closing-price diagnostics where trustworthy data exist. MLB and NHL refreshes now record each published pick’s latest pregame snapshot after publication (scheduled at 7:05 a.m. and 4:30 p.m. Eastern), summarized on the briefing; it is a later snapshot, not the closing line, and it does not change selection or ranking.

Implementation sources: research policy, scheduled workflows, and decision log. Configuration fingerprint: dffb55e5aff57ce3eb1c.

Return to today’s picks →