Public methodology
Historical data. Clear assumptions. Reproducible evidence.
Our evidence standard
Number99 aggregates public result records. It is not an official market operator. A source chart date is not necessarily a draw timestamp. We preserve source-specific date and timing context and never infer a universal schedule.
Source selection and independence
Tracked sources must expose a readable, dated chart with identifiable markets. Domain separation alone does not prove independent reporting. Independence groups are assigned only after documented review of ownership, feeds and copying. Unreviewed sources cannot automatically verify a result.
Collection and normalization
Collectors accept only unambiguous dated tables from an allowlist. Blocked pages, uncertain dates and parser failures are recorded as errors, not results. Gali/Galee, Disawar/Desawar/Deshawar/Deshwar/DS, Faridabad/Farida and Ghaziabad/Gaziabad/Gazi/GZB map to four canonical markets. Raw observations are append-only.
What the status means
HIGH requires at least three reviewed independent source groups agreeing. MEDIUM requires two. LOW means one reviewed independent group. Agreement between sources whose independence is unknown remains unverified. Any numerical conflict stops automatic publication; no majority winner is selected. Pending cells do not count as zero or a numerical conflict.
Data Confidence v1 — experimental
Base 20; first independent agreeing source +20, second +20, third +15, fourth +10; documented historical consistency up to +5, exact retained capture timestamps +5 and archived evidence +5. Meaningful conflict −30, changed source value −15, uncertain parsing −15. Scores are bounded from 0 to 100. Consistency and archive bonuses remain inactive until the corresponding evidence exists. No usable evidence means no score.
Score labels and verification are separate
96–100 Very High; 80–95 High; 60–79 Medium; below 60 Low. Conflicts always display Manual Review. Source-count status and the experimental score measure different aspects of the record. Neither is a calibrated probability, draw fairness assessment, or probability of a future result. Manual review does not automatically increase the score.
Corrections and review
A reviewer must select a value supported by preserved evidence and provide a reason. Published value changes create correction and audit entries atomically. New contradictory observations keep the previous value in the audit trail but withdraw it from normal result cards pending review. Concurrent reviews must reload if the record has changed. The public log excludes user identities.
Preserved legacy records
Existing charts are retained as LEGACY_UNVERIFIED when they lack linked original source evidence. Their values are not invented or replaced with sample numbers. Original publication times and complete earlier correction histories are not available. Empty dates are omitted, never filled as 00.
Historical seed evidence
The 1–6 September 2026 source sample consists of manually transcribed table excerpts, not original screenshots or raw pages. Capture times have day precision only. It is labelled TRANSCRIBED_EXCERPT. Import time is not substituted for original first-seen time. Hashes protect stored bytes; they do not prove source authenticity or publication time.
Creator evidence score — planned framework
Timestamp transparency 25; record completeness 25; prediction specificity 20; edit transparency 10; result corroboration 10; claim discipline 10. This measures transparency and auditability, never profitability or prediction accuracy. No creator scores are published until evidence has been reviewed and a reproducible rubric applied.
Access-level disclosure
Creator reviews distinguish FULL_VIDEO_REVIEWED, TRANSCRIPT_REVIEWED, FULL_PAGE_REVIEWED, CAPTION_ONLY, DESCRIPTION_ONLY, SEARCH_SNIPPET_ONLY, SCREENSHOT_ONLY and INACCESSIBLE. Automated source-table captures are labelled TABLE_CAPTURED; manually transcribed historical excerpts are TRANSCRIBED_EXCERPT. Neither implies a full human review.
What Number99 cannot verify
Public agreement does not establish the true draw mechanism, operator independence, fairness, actual wager volumes or creator profitability. Missing posts do not prove deletion of losses. Insufficient evidence is not equivalent to falsehood or fraud. Historical association does not guarantee future outcomes.