/ Methodology

How a posted pick becomes a record you can check.

This is the plain-language version of how the system captures, prices, grades, and reports picks. It is deliberately specific — the whole product is that our numbers mean exactly one thing, and you can see what that thing is. Where a convention disagrees with a capper's self-reported record, that disagreement is the point, not a bug.

/ The plain-language version

Every convention here has one job: make a number mean exactly one thing.

Capture, pricing, grading, reporting — each step is written down so you can see what a figure counts and what it doesn’t. Where a convention disagrees with a self-reported record, that disagreement is the point.

01

From post to record

The bot watches your picks channel passively. There is no /pick command — cappers write in plain English, exactly as they always have. When a pick appears, the system does five things in order, and none of them can be undone later:

  • Timestamp at post (T0). The pick's clock starts at the exact second the message lands. We never backfill a record to make an account look older or more established than it is.
  • Parse and normalize. The language model reads the message into a structured pick — league, market, side, line, stated odds, units. When confidence is low, it asks for a one-tap confirmation rather than guessing. The model interprets; it never invents an odds number, a result, or a player.
  • Snapshot the market. We capture prices across monitored US-licensed books at T0, then re-capture as the line moves. Those snapshots are what "attainable" is measured against.
  • Preserve immutably. The pick is written to an append-only record. Corrections are new dated rows, never overwrites — there is no delete privilege anywhere in the system.
  • Grade against game data. Once the event settles, the pick is graded straight from the official game result, by fixed rules. The capper never grades their own pick.
Why passive ingestion

A record only means something if the thing being recorded can't choose what gets recorded. Because the bot sees every pick the moment it's posted — not the ones someone later decides to log — there is no survivorship editing at the source.

02

Attainable odds & the three-price model

A single "the odds were X" hides most of what matters. So every pick carries three prices, kept separate on purpose:

  • P_stated — the price the capper claimed.
  • P_att0 — the best price actually attainable at the moment of posting, across monitored US-licensed books.
  • P_real — the price a follower actually got, when a tail is recorded.

Every price is converted from American to decimal the moment it's captured and stored with the probability it implies, so every comparison we make is on a fair-odds basis — the book's built-in margin taken out — rather than on raw American odds. If a pick is posted without odds, we fill in the fair-odds consensus across books at post and mark it odds_imputed — never a best-case number. If the stated price wasn't actually gettable at post, the pick is flagged stale_line: that flag is an honesty signal we show on purpose, not something we quietly clean up.

Why we take the margin out

Every book builds in a margin — the vig. Comparing raw prices across books compares their margins as much as their opinions. Taking the margin out first lets us compare the real probability each price implies — the only fair way to say whether an edge was actually there.

03

How grading works

Grades are not binary. A pick can land in any of six states, and every one of them is counted on the record — nothing is silently dropped:

WIN LOSS PUSH VOID UNGRADED PENDING_REVIEW

Grading follows fixed rules and runs from official game data, cross-checked against a second source when one is configured. When two sources disagree, or when a case is inherently ambiguous, the pick is routed — loudly — to a review queue rather than auto-graded. That "unsafe set" includes things like a player who didn't play a prop, a postponed or suspended event, and a stat line still inside its correction window.

  • Corrections are append-only. If a stat correction flips a grade, we add a dated GradeRevision and a public correction-log entry — we never overwrite the original grade in place.
  • Overtime counts toward totals, spreads and moneylines. A postponed event that isn't completed within 24 hours of its scheduled start is voided.
  • Official and realized are never merged. A capper WIN that is a follower VOID under a different book's rules is correct output — two true facts about two different questions — not something to be "reconciled" away.

Parlays are graded leg by leg. Each leg of a multi-leg pick is settled on its own from official game data. A void or push leg drops out and the combined odds are recomputed over the surviving legs; any losing leg settles the whole parlay as a LOSS; a leg that routes to review holds the parlay in PENDING_REVIEW until it clears; if every leg voids, the parlay is a PUSH. A parlay is settled only once every leg’s event is final — and every leg stays visible on the record.

Every graded pick stores the exact grading_rules_version it was settled under, so the rule set that produced any number is always recoverable.

04

Reading a record

We publish two ROI records side by side, always — and we lead with the one that can't be gamed:

  • Flat-stake ROI (headline). Every pick is treated as one unit. "5u lock" talk can't retroactively reshape a flat record, which is exactly why it's the number we lead with.
  • Stated-units ROI (secondary). Computed at the capper's declared stake, frozen at first sight — shown with the flat number, never instead of it.

ROI is profit divided by total risked, pooled — never an average of per-pick percentages. Win% is W / (W + L) with pushes excluded from the denominator, and we always show the full W-L-P plus the count of voids and ungraded picks. Units and rates only — you will never see a dollar-winnings figure on any surface we publish, because that is exactly the opacity marketing we exist to replace.

Why we don't rank on ROI

Detecting a real edge from raw ROI at typical prices can take thousands of picks; closing-line value converges in dozens to a few hundred. So ROI is a reporting metric and closing-line value is the ranking metric. No surface we build sorts cappers on raw ROI or win-rate — those make good screenshots and bad evidence.

05

Slippage & the official-vs-realized gap

Slippage is the distance between the number a capper posts and the number a follower can actually get. We measure it directly: capture the pick the second it posts, snapshot every monitored US book, then re-snapshot at 30 seconds, 2, 5 and 15 minutes. Posted price minus what's still gettable — compared on a fair-odds basis, with the book's margin taken out — is the slippage, and the baseline is always the best price attainable at post, never a best-case screenshot price.

Aggregated across a capper's picks, this produces the official-vs-realized gap: a record can look excellent at posted prices while the attainable record — what a follower could realistically get — is meaningfully worse. We ship how much of a capper's activity a figure covers alongside every aggregate, count one-sided markets without adjusting their margin, and label every figure with how many picks it's built on, the dates it covers, and the version of how we track. We publish our own error rate too.

On the numbers you see today

Until a pilot community opts a period into public sharing, any slippage figure on this site is an illustrative preview on demo data from our own test server, clearly labeled as such. We would rather show a thin, honest launch than launder a made-up number as a finding.

06

Tailability

Tailability is a single score for how followable a capper actually is — not an opaque "AI score." It's a plain weighted blend of pieces that each come straight from the locked record, and every piece is shown on its own, always. The main input is follower closing-line value; the rest capture whether the posted price was actually available, how the line held up as money came in, and how long the window to bet it stayed open.

  • Sample-size gated. No score at all below 30 closing-line-eligible graded picks. Correlated same-event picks count once, so volume can't be farmed into a score.
  • Confidence shown, not hidden. Between 30 and 99 picks the card is marked provisional with the count printed. The core estimate shows its 95% confidence interval; if that interval spans zero edge, the card says so in plain words — "not yet distinguishable from market."
  • Built so it can't be gamed. Coverage gaps pull the score down; volume never lifts it; per-sport subscores are published so one strong category can't hide a weak one.

Weights are frozen on first publication. Any change means a new methodology version, a changelog, operator sign-off, and a 30-day period where both versions are published side by side.

07

Our independence

We say independently tracked — and we mean tracked, not rubber-stamped. A badge is something someone can sell you; "independently tracked, append-only, timestamped at post, graded by a system that can't be paid to look away" is a sentence you can check. We keep the language of formal attestation in reserve for a future book-synced execution tier that does not exist yet, and we won't borrow its credibility early.

  • We take no sportsbook money. No affiliate links, no referral or discount codes, no "bet here" buttons, no paid partnerships — on any surface, ever. Affiliate revenue is modeled at zero, permanently, because taking it would detonate the independence that is the entire product.
  • We don't sell picks and we issue no advice. We never tell anyone to place, tail, or fade a bet. We describe records; we do not issue picks.
  • The evidence outlives the vendor. Raw vendor odds feeds live in a separate, purgeable cache; our permanent record holds only the measurements we computed. A data-vendor termination — or a pressured takedown — can empty the cache and change not a single published number.
  • Named records are consent-gated. A community's real record is only ever published with written consent and a period the owner opted into public sharing. Until then, everything you see is demo data, labeled.
The one-line version

We are a measurement layer, not a tout. We win by showing our work — including our own error rate and the losing weeks — not by showing you a screenshot and asking for trust.

08

RC Verified & the publication floor

RC Verified means a community’s betting record is independently tracked by Ranked Cappers: every pick is timestamped when it is posted, graded against game data, and written to an append-only ledger that no one — not the community, not Ranked Cappers — can edit or delete. The community’s owner has opted in to publishing this record. RC Verified is a statement that the record is independently tracked and owner-published — it is not an endorsement, rating, ranking, or prediction of future results, and it does not assert the record is good or profitable.

  • The floor is earned, not bought. A community becomes eligible to publish only once its own append-only ledger holds at least 50 graded picks spanning at least 14 days. The floor alone never publishes anything.
  • Publication is an explicit owner action. Going public takes a recorded owner opt-in with a chosen public slug, and it is revocable at any time. An administrator may force-eligible a below-floor community for a labeled demo — the pilot profile is exactly that.
  • The public page is official-record-only. It shows W-L-P, net units (flat and stated), ROI, sample size, date range, the record table with losses shown, and an integrity summary. Closing-line value, follower/realized records, slippage and Tailability are never reachable on the public route — the code that builds a public profile never loads them.
  • The badge states one fact. The embeddable badge says only “record independently tracked” with a pick count and the profile link. It is a status, never a rating or a claim of accuracy.
On communities that aren’t listed

Only communities whose owners have published a record appear in the directory. A community that isn’t listed simply has no owner-published record on file — that is the only thing its absence means. We never label any community, and we publish only true facts measured from our own tracked data plus our verification status.

← Back to home Browse the directory Get your Discord tracked

/ A measurement layer, not a tout

We win by showing our work — including our own error rate.

No sportsbook money, no affiliate links, no “bet here” buttons on any surface, ever. The evidence outlives the vendor, and named records are published only with a community’s written consent.

Independent · append-only · consent-gated