Claimed model edges
Backing the model's claimed disagreement returned −8.67% ROI over 6,870 observations. Performance worsened to −10.61% as claimed edge grew — a signature of model error, not model edge.
n = 6,870RetrospectiveModel v3.4
Independent market researchEvidence before conviction
A market desk
that shows its work.
EdgeDesk prices markets independently, investigates meaningful disagreement with the market, and opens every call to inspection — from model inputs and book-by-book prices to cited team news and the final grade at close.
A research subscription, not a tip feed. No promise of profit.
ED–2841 / LIVE OBSERVATION
UNDER REVIEW
Premier League · 24 Jul · 19:45 UTC
Updated 14s ago
Arsenal v Newcastle
EdgeBook price
2.18
45.9% implied
Disagreement
+4.7%
research trigger
Sharp anchor
2.29
43.6% de-vigged
12:0014:0016:00Now
EdgeBook Market anchor
~36Books monitored
821kPrice snapshots archived
33,364Matches in fitted model
18Football leagues
Archive snapshot · July 2026
Architectural boundary
No funds held.No bets placed.No bookmaker credentials.No execution path.
01 / The desk
See the full case,
not a pick.
Every market observation opens into a complete research file: the independent price, the de-vigged benchmark, movement across books, model inputs, supporting evidence, and the grade at close.
-
01
Watch the market
Dense histories across football, tennis, MLB, NBA, NFL, and cricket.
-
02
Build an independent price
A fitted Dixon–Coles scoreline model, separate from consensus.
-
03
Test the disagreement
Compare with a de-vigged market anchor. A flag begins research; it proves nothing.
-
04
Validate the context
Retrieve cited reporting in full. Retain, revise, or reject.
-
05
Grade the close
Publish the closing-line result whether it supports the process or exposes a flaw.
01 / MARKET ARCHIVETRACE: ED–2841
Pinnacle2.29Sharp anchor
Book 022.3414s
Book 112.45outlier
Book 192.3121s
Book 312.2709s
EVIDENCE STATE · 821,000 SNAPSHOTS
Every move keeps a timestamp.
Book-by-book prices are retained as a dense event history, not reduced to a single screenshot. The research file can show what moved, when, and against which reference.
02 / One coherent price
One scoreline matrix.
Every market.
1X2, Asian handicaps — including quarter lines — totals, and both-teams-to-score all derive from the same underlying distribution. Related markets cannot contradict each other.
Away goals →
Home goals →
01234+
04.28.67.44.12.0
17.314.912.87.03.4
25.811.910.25.62.7
32.95.95.12.81.3
4+1.53.12.71.50.7
Illustrative probabilities · competition-specific decay and correlation
Derived from one distribution
Home win45.9%2.18
Draw27.4%3.65
Away win26.7%3.75
Over 2.551.2%1.95
Both teams score53.8%1.86
Home −0.25—1.84
Internally consistent by construction
01Market archive~36 books
→
02Dixon–Coles model18 leagues
→
03Scoreline matrixone source
→
04Market comparisonShin de-vig
→
05Evidence passsources fetched
→
06CLV ledgergraded at close
03 / The evidence ledger
The failures are
part of the product.
The Ledger records what failed, what may be real but is not yet usable, and what still needs more evidence. Negative findings do not disappear when the page refreshes.
Recorded · not buried
The standalone model did not beat the closing benchmark.
That is the strongest conclusion in the current evidence. The research subscription is access to a disciplined process and its record — not a claim of a proven edge.
Goal model0.6293
Pinnacle close0.5912
Lower Brier score is better · standalone benchmark test
League-level edge
Scotland's apparent +23.7% result evaporated under split-period retesting. Only 9 of 18 leagues retained the same direction.
9 / 18 sign agreementSplit-period test
Chance-quality proxy
A shots-based chance-quality feature did not improve the model. It remains excluded from production pricing.
Feature rejectedOut-of-sample
De-vig method choice
Five tested methods landed within 0.045% of one another. Method selection is not where the useful disagreement lives.
5 methodsΔ ≤ 0.045%
Real information. Not yet bankable.
The model's disagreement with Pinnacle predicts a small part of Pinnacle's error: r = +0.047, 95% CI [+0.017, +0.076]. The relationship is real; the resulting Brier improvement remains too uncertain to support a usable-advantage claim.
Incremental informationGain CI crosses zeroNo performance claim
Encouraging archive replay. Still unproven.
The screen replayed on its own archive at +2.50 CLV points, 95% CI [+1.19, +3.81], n=97. It remains below the predefined 300-observation gate and is retrospective, not live proof.
n = 97Archive replayBelow 300 gate
Research findings, not a claim of future performance. Past results, model scores, and CLV do not establish profitability.
04 / The book
Every call meets
the close.
A research note is not complete when it is published. EdgeBook retains the available price, the closing benchmark, resulting CLV, evidence state, and model version — whether the grade is favorable or not.
LIVE EVIDENCE GATECurrent snapshot · Jul 2026
62/ 300
300
Below this threshold, no performance conclusion is made. Reaching 300 permits evaluation; it does not prove an edge.
No selective windowsNo deleted lossesNo retroactive reasoning
GRADE / ED–2794Negative CLV
Liverpool v Brighton · Over 2.75
Published1.9811:42 UTC
Close2.04Pinnacle de-vig
CLV−2.94%retained in book
The market moved against the published observation. The reasoning file remains open with the original inputs, source excerpts, and timestamps.
01
The desk
Live market screen and every research note open to its full reasoning.
02
The analyst
Interrogate any call, assumption, movement, or evidence state.
03
Your book
A complete personal CLV record with the evidence gate always visible.
04
The ledger
The experiment archive: what worked, what failed, and what remains unknown.
05 / Interrogate the desk
Ask why it fired.
Ask what would
change its mind.
EdgeBook is not a one-way feed. Question any observation, inspect what moved, compare model and market, or ask what the current CLV record can honestly support.
ANALYST / ED–2841 Evidence linked
Illustrative interaction · analyst answers may contain errors and should be independently checked.
Advisory by design
Research only.
No execution path.
01Market dataprices · results · sources
→
02EdgeDeskanalyse · explain · grade
→
03Client decisionindependent and external
×Bookmaker executionno credentials · no integration
Bespoke research subscription
Access a disciplined process —
not a stream of tips.
EdgeBook access includes the live desk, complete reasoning behind every market observation, the analyst, your CLV book, and the full evidence ledger.
No funds heldNo bets placedNo execution integrations
Availability depends on client location and applicable law. Research subscription only; not betting or financial advice.
07 / Frequently asked
Questions,
answered plainly.
01Does EdgeBook place bets?
No. It holds no funds, accesses no bookmaker account, and has no execution capability. Its output ends at research presented to the client.
02Is this a tipster service?
No. EdgeBook is an advisory research desk. It publishes inspectable reasoning and grades every eligible observation against a defined closing-line benchmark.
03Does the model beat bookmakers?
That has not been demonstrated. The standalone football model does not beat the Pinnacle close, and the current live sample remains below the predefined evidence gate.
04What is a “disagreement”?
A measurable difference between EdgeBook’s independent price and a de-vigged market benchmark. It is a trigger for investigation, not proof of value.
05Why grade on closing line value?
CLV records whether the published price improved or deteriorated relative to the defined close. It is a process grade, not a guarantee of profit.
06Can the news review reject a call?
Yes. The second pass retrieves cited articles in full and re-tests the original assessment. It can retain, revise, or reject the observation.