Next Level Dynasty Football

One number per manager, league mean 100, one standard deviation 15. 115 means "one standard deviation better than the rest of this league". It is a ranking within your league, not an absolute rating — the average manager in every league scores about 100.

Manager Score = 100 + 15 × (0.50·zdraft + 0.35·ztrade + 0.15·zwaiver)

Weights are renormalised over whichever components your league actually has, so a league that never trades is not scored on a component nobody played.

The unit: value capture

Everything below is built from one measurement. For any asset that changes hands on a date, we take its KeepTradeCut value on that date and the highest value it reached afterwards. The difference is what was captured:

captured = peak value after the transaction − value on the day of the transaction

Acquiring an asset captures that gap; giving one up forfeits it. So acquiring a player just before he breaks out scores well, and trading away a player who then breaks out scores badly — which is the thing everyone actually argues about in a dynasty league.

Draft skill — did the pick beat its slot?

Surplus = what the pick captured, minus what a pick at that slot typically captured in that same draft. The par curve is fitted from the draft's own picks (median per six-slot bin, forced never to rise as slots get later), so the average pick in a room scores zero by construction. That is what makes startup drafts, rookie drafts, 10-team and 14-team leagues comparable without any hand-tuned constant. A draft with fewer than 12 evaluable picks is skipped rather than scored badly.

Trade skill — did value come in or leak out?

Net = captured by what you received, minus captured by what you gave, counting players and draft picks on both sides. Within any trade these sum to exactly zero, so it is a genuine transfer measure. A trade containing an asset we cannot price is reported but not scored — pricing one side and not the other would invent a steal that never happened.

Waiver skill — was there value on the wire?

Captured by each waiver or free-agent add. A replacement-level body that never rose captures zero; plucking a player who then becomes a starter captures a lot. FAAB spent is shown in the audit but not scored: KTC points and FAAB dollars have no exchange rate, and inventing one would be the least defensible number on the page.

Why volume cannot buy a good score

Each component is a per-transaction mean, not a total, then shrunk toward zero by n / (n + k) (k = 6 picks, 3 trades, 5 adds). Shrinkage only ever moves a manager toward average — it never flips a sign and never overshoots — so one lucky pick cannot top the table, and making 40 mediocre trades cannot either. A manager with no activity in a component is scored as league-average for it, flagged, and not penalised for abstaining.

KeepTradeCut publishes a live superflex consensus board, not an archive, so historical values are not available from KTC directly. The dated boards behind this page were recovered from public web-archive captures of that page and are stored in this repository, one small file per date. Nothing here is interpolated, modelled or invented — every number was published by KTC on the date it is filed under.

The archive is real but sparse: a few dozen dated boards rather than a daily series, with gaps of weeks to months. Consequences, stated rather than hidden:

• A transaction is priced at the nearest board on or before its date — never after, because a price that did not exist yet is not a point-in-time price. The date actually used is shown on every row.
• A "peak" is the highest value we observed. The true peak may fall in a gap, so captures are best read as a floor on what was really available.
• Transactions with no recorded board after them are not scored at all and are counted separately. You cannot grade foresight on a trade made last week.

Going forward the daily job files a new dated board on every run, so the series densifies from here even though the past cannot be filled in.

Any Sleeper league, and gently. Enter any league id (or find one by username) — nothing about this page is specific to one league, and the whole dynasty chain is discovered by following previous_league_id backwards from whatever you type. Sleeper's API is public, unauthenticated and free, so the reads are kept modest: every response is cached for the tab, completed seasons are cached indefinitely because a finished box score cannot change, the live season is only asked for weeks that have actually started, requests are batched by week rather than per player, and no more than a handful are ever in flight at once. Scoring the same league twice costs zero further requests.

Everything above prices a trade against the market. It answers was this a good bet at the time. It cannot answer the question you actually remember — did it work out. Trade for a back who then carries you to a title, and trade for a back who then tears an achilles, and on the day you made them the two can price identically.

So every trade now carries a second, independent figure: the points the players you acquired actually scored for you, from the week of the trade onward.

Where the points come from. Sleeper's own weekly matchup records, which report each rostered player's score computed with your league's scoring settings. Not a generic PPR approximation: this league is half-PPR with a tight-end bonus, and a generic feed would be wrong for it in a way that quietly favours receivers. The realized lens therefore needs no scoring configuration and is exact rather than close.

Who held whom is read, not inferred. The same weekly record says which roster each player was on that week. So a player traded on again, dropped, or acquired the day before kickoff needs no special case: we count the weeks he actually spent on the acquiring roster and no others. Production before the trade stays with the manager who earned it.

Normalisation. Raw points would make a week-2 trade beat a week-12 trade for no better reason than having ten more weeks to accumulate. Two defences are reported:

• PAR — points above replacement, the headline. Replacement is the median score of a started player at the same position, in the same week, in this league. Being position-relative it does not reward acquiring a quarterback simply because quarterbacks score more; being week-relative it self-normalises for span; being league-relative it does not inflate in a high-scoring format.
• Points per week held, a plain rate, for readers who want the unadjusted version.

This counts all rostered production, bench included. Sleeper reports both the full roster and the weekly lineup, and the choice is deliberate: what is being graded is the acquisition, not the manager's weekly lineup card. Trading for a player who goes on to produce is one skill; leaving him on the bench while he does it is a different one. Charging the trade for a benching would conflate them, and would punish acquiring a stash who breaks out while blocked behind somebody else on your roster.

The honest caveat. Bench points did not directly win anybody a game. A manager who acquires a producer and never starts him gets full credit here for a benefit he never actually banked. That is a real consequence of scoring the acquisition in isolation, and it is not offset anywhere: lineup management is not measured on this page at all. So that the gap stays visible, every trade also reports how much of that production was actually in the lineup — scored on the total, shown on both.

Replacement level follows the same population. Because every rostered week is scored, the bar a player is measured against is the median of all rostered players at that position that week, not the median of those who were started. Mixing the two would charge a benched or bye-week player a starter's replacement level for a week he was never asked to play, and would manufacture large negative PAR out of nothing. Measured across four real seasons this puts the RB bar at about 3.5 points a week rather than 10.9, because roughly a third of rostered players score zero in any given week. PAR figures are correspondingly larger than a starter-based version would give — a change of scale, not of accuracy, since every trade is measured against the same bar.

The two lenses are never averaged. They answer different questions and will sometimes disagree; where they do, the page says so explicitly. A manager who lost value by the market and won by the scoreboard bought win-now production cheaply — which is a real way to win a league and a real way to bleed value doing it. Averaging deletes exactly that signal.

One property worth stating: unlike the market lens, realized value is not zero-sum. Points are produced, not exchanged, so both sides of a trade can genuinely come out ahead.

The obvious follow-up is "what was he projected to score at the time, and did he beat it?". That was investigated directly against the live API rather than assumed, and the answer is more awkward than either yes or no.

Sleeper does return per-week projection rows for past seasons — 2023 through 2026 all answer. But they are not a trustworthy point-in-time archive, for two independent reasons:

• The rows are rewritten after the fact. Each carries an updated_at. For 2025 week 11 it reads 2025-11-18, a day after those games. For 2025 week 2 — games played 11-15 September — it reads 2025-10-06, three weeks later. For 2023 the field is absent entirely. There is no way to recover what was on screen before kickoff, which is the only number worth comparing against.
• The units are wrong for your league. Projections are published as generic pts_ppr, pts_half_ppr and pts_std. None of them is your league's scoring. Comparing a generic projection against a league-scored actual produces a difference that is partly just the two formats disagreeing.

So no expected-vs-actual figure is shown, and none has been invented. Realized production answers the owner's question without projections anyway: "he scored this much for you, and that ranked him here" is a stronger claim than "he beat a forecast", because it is measured in the only units that decided your matchups.

Going forward this becomes answerable honestly. scripts/archive_sleeper_projections.py files a dated snapshot per run, in the same pattern the KTC archive uses, stamped with the date it was actually captured. Once a trade is made after a snapshot exists, the expected-versus-actual comparison can be made from a projection that provably predates the outcome. It cannot be backfilled.

• Hindsight is baked in, by design. This measures what the assets went on to do. A defensible process that ran into an injury scores badly here, and a reckless punt that hit scores well. It is a measure of results, not of reasoning.
• Realized production says nothing about draft picks. Sleeper's trade record names a traded pick, but the draft-results feed does not say which selection descended from which traded slot, so attributing a drafted player's points back to the pick that bought him would be a guess. Picks in a trade are reported as unattributed for this lens. The market lens prices them directly, which is the clearest argument for showing both rather than one blended number.
• A trade made this week has produced nothing yet. Realized value needs weeks to accrue, and trades still accruing are labelled ongoing rather than presented as settled. This is the mirror of the market lens's blind spot, which is strongest exactly where this one is weakest.
• KeepTradeCut is a crowd, not an oracle. Where the consensus was wrong, this page is wrong the same way.
• Only players on the board are priced. KTC publishes a top 500; assets that never appear cannot be scored and are counted as unpriceable rather than guessed at.
• Future picks are priced at the "Mid" tier. A traded 2027 1st could land anywhere; KTC prices Early/Mid/Late separately and we take Mid, because the final slot depends on standings that have not happened.
• Roster management is not measured. Lineup decisions, injury stashes, taxi squads and contending-vs-rebuilding timing are all invisible here.
• Sleeper only. The live reads target Sleeper's public API. MFL leagues are pre-baked elsewhere in this project and are not covered by this page.
• Credit is given for points that never won a game. Realized value counts all rostered production, so a manager who acquires a producer and leaves him on the bench is credited in full. That is deliberate — the acquisition is being graded, not the lineup — but it does mean the number can overstate what a trade actually delivered on the field. Every trade shows how much of its production was started, so the gap is visible.
• Lineup management is not measured anywhere. Neither lens rewards or punishes starting the right players. A manager who consistently benches his best available starter looks identical here to one who never does.