A lens is a way of seeing with a scoreboard. It is not a mood, a slogan, or a fresh label for whichever side looks attractive in NFL Week One. A lens makes a claim before the result is known. The claim is graded at final. The lens then carries that grade into the next place where the same claim is applied.
That last sentence is the contract. The record belongs to the claim in its exact context. It does not belong to the author, the page, or the week. A useful lens can be wrong on Sunday and remain useful. An ungraded lens can sound precise and still tell the reader nothing about whether it deserves weight.
The lens contract
Every lens needs a claim that can be written without the result. The claim names the market, the side of the comparison, the information available at decision time, and the condition that turns an observation into a call. It also names the conditions that produce no call. A lens that fires on every game is often hiding the fact that its evidence only applies to a narrow set of games.
The decision is timestamped before the grading boundary. The final result then produces a settled grade under the market's published rules. A push remains a push. A void remains a void. A game excluded by the lens contract does not become a win because the outcome happened to agree with the thesis. The scoreboard is useful only when the eligibility rule stays fixed after kickoff.
The public record should make the misses easy to find. A card that leads with the best streak but buries the losing contexts is an advertisement, not an audit. The lens page should show the claim, the current gate, the eligible sample, the settled record, and the contexts where the claim has not held. The result is not a promise about the next game. It is evidence about how the same claim behaved before.
A lens is not a system and it is not a pick
A system is the machinery. It can collect prices, transform inputs, fit a model, rank choices, and publish decisions. One system may contain several lenses because it can ask several distinct questions of the same board. The system can also change without changing a lens, provided the claim and its grading contract stay intact.
A pick is one terminal decision. It is tied to one event and one market at one moment. The pick can be graded, but the pick alone has no reusable memory. Without the lens, a reader sees the answer but not the rule that produced it or the evidence that should determine how much weight to give it.
A lens sits between those two. It is narrower than a system and more durable than a pick. It says, in effect: when this defined situation appears, this is the direction the evidence points, this is when the call was made, and this is what happened on every eligible prior call. The lens turns a stream of picks into a falsifiable claim.
Context is part of the score
A record without context is an invitation to mix unlike decisions. A spread call made against an opening number is not the same claim as a spread call made near close. A side lens is not a total lens. A lens evaluated before a major status change is not interchangeable with the same label evaluated after the market absorbed that change. The name can stay the same while the claim has quietly changed.
The scoreboard therefore travels with a context key. That key can include sport, market, decision window, competition phase, eligibility filters, and claim version. The point is not to create a taxonomy for its own sake. The point is to stop a strong record in one context from laundering a weak record in another.
Week One makes this discipline visible. New roles, new combinations, and limited current-season evidence put pressure on every prior assumption. A lens may still apply, but its public card should show whether the record comes from the same context or from an adjacent one. If the current application is outside the proven context, the label should say so before the call is read.
The four evidence gates
GRADING means the claim is live and its eligible calls are being settled, but the public evidence is not yet strong enough for the highest gate. This is not a negative label. It is the honest state for a new claim, a changed claim, or a claim entering a materially different context. The lens can be followed, but its record is still being formed.
PROVEN means the lens has cleared the product's evidence standard in the context being displayed. The label should depend on settled calls, sample size, uncertainty, and the stability of the claim across the relevant evaluation window. It is not a lifetime title. A proven lens can move back when the evidence changes.
FEED_ONLY means the source contributes information but does not make a clean, gradeable claim. An injury feed, a price feed, or a stream of attributed observations may be valuable without choosing a side. Keeping that material feed-only prevents information coverage from being mistaken for forecasting skill.
DECAYING means the lens has a prior record but current matched evidence is weakening. The cause may be market adaptation, data drift, a changed decision process, or a context that no longer resembles the one in which the lens earned its gate. The label says the old record remains part of the history while warning that it should not carry its old weight automatically.
Why a coin-flip lens is demoted
A public scoreboard should answer a hard question: does the settled evidence support treating this claim as more informative than a coin flip? A point estimate alone cannot answer it. Small samples move sharply after one result. A wide uncertainty band can include both a useful effect and no effect. The gate must account for that uncertainty rather than turning the latest streak into a status change.
When the evidence cannot separate the lens from the baseline, the public product should demote it. That may mean moving from PROVEN to GRADING, from GRADING to FEED_ONLY, or from a visible recommendation role to an observational role. Demotion is not punishment. It is the mechanism that keeps the interface from presenting confidence the record does not support.
The reverse also matters. A lens should not be promoted because its recent run looks clean. The underlying claim must remain fixed, the calls must be eligible under the same rule, and the settled sample must support the change. A gate earned by rewriting the definition after each miss is not evidence.
How to read a lens in Week One
Start with the gate, then the context, then the record. A PROVEN label in a mismatched context should carry less weight than a GRADING label whose current calls match the displayed situation exactly. Check whether the lens made a call or correctly stayed silent. A no-call is part of the contract when the required evidence is absent or the available price falls outside the lens's decision rule.
Then read the misses. The useful question is not whether the lens has ever been wrong. It is whether the losses cluster around a condition the current card also contains. A lens whose misses concentrate after late market moves should not be read the same way before and after such a move. A lens whose record weakens outside its original decision window should not borrow the stronger record without disclosure.
Finally, separate the lens from the bet. A lens can make a sound directional claim while the available market price makes the wager unattractive. The scoreboard grades the claim under its contract. The decision to act still depends on the price, the market, and the bettor's own constraints.
What the public record must preserve
An auditable lens record preserves the claim version, the decision timestamp, the market snapshot used by the claim, the call or no-call, the final grade, and any exclusion reason. It also preserves changes. When the rule changes, the version changes. The new version may inherit lineage, but it does not silently inherit the old version's scoreboard.
This is why a lens is more than a card treatment. The interface can compress the record into a gate and a short line, but the underlying contract must remain inspectable. The current live figure belongs on the lens surface that owns the claim. This explainer does not substitute a static number for that live record.
Sources and method
This article describes the grading contract used by the product concept: claim before outcome, settlement at final, context-specific record, and evidence-based gate changes. It intentionally contains no live team result, no current lens rate, and no fabricated Week One scoreboard. Current records should be read on the live surface where the claim, context, sample, and final grades can be inspected together.
The standard is simple to state and difficult to maintain: define the claim, freeze the eligibility rule, grade every eligible call, expose the misses, and let the evidence control the gate.
Model calibration: predicted vs observed
Predicted win probability bucket vs the empirical win rate inside that bucket on the test set. Points on the y=x reference line are perfectly calibrated; points below mean the model is overconfident in that bucket.
EV per $100 across win rate × odds grid
Expected value of a $100 stake at each combination of true win rate and market odds. Anywhere the cell is positive you have a long-run profitable bet; the magnitude shows how aggressive Kelly will size it.
