A practical guide to manual review of disagreeing sources in TEMIRIN using configured records, explicit market identity, current controls, and human review.
August 3, 2026 · 5 min read
Two sources can appear to conflict while answering different questions. One may report a preliminary count, another a certified total; one may describe an announcement, another the date when a rule takes effect. Begin by writing the smallest factual claim that matters to the configured market, including actor, event, threshold, jurisdiction, and time window. Then reopen the exact market wording and resolution source. A precise claim prevents a broad headline disagreement from being mistaken for evidence about the contract itself.
The workspace stores configured RSS, HTTPS web, and provider-configured Telegram or X records with deterministic fields and provenance. Human reviewers can place records side by side, record their interpretation in attributable notes, and leave the conclusion unresolved when the source material does not justify one. Uncertainty is a valid review result, and the canonical records remain available for another reviewer to inspect.
For each record, inspect publisher identity, canonical URL or provider reference, observed time, supplied publication or event time, and the configured source that collected it. An article quoting another outlet is not independent confirmation merely because it has a different domain. Follow the attribution chain until the original document, statement, or dataset is visible when possible. Canonicalization and deduplication help prevent repeated web or RSS items from inflating the record count, but they do not certify factual accuracy.
Separate source quality from transport health. A failed reader can hide a publisher update, while a successful reader can store a weak claim perfectly. Likewise, a successful email, Telegram, Slack, or webhook attempt proves only that an alert destination accepted a message. Use ingestion status and the alert-delivery ledger for operational diagnosis, not as votes for one interpretation. If an item arrives late, preserve both publisher and observation times rather than silently moving it earlier in the research timeline.
Use one row per evidence record and columns that a reviewer can verify directly: precise claim, source type, named primary authority, timestamp, quoted scope in paraphrase, relevant contract clause, and reviewer note. Add a correction status when a publisher has changed its account. Avoid a single blended confidence number that hides why records differ. TEMIRIN’s deterministic brief organizes stored fields while the reviewer remains responsible for opening and evaluating the canonical material.
Classify the relationship in plain reviewer language such as same claim with different values, preliminary versus final, prediction versus reported fact, different jurisdictions, or unresolved attribution. If several outlets repeat the same wire report, keep the shared origin visible. If the records cannot be reconciled, state what additional document would change the review and who will check it, rather than manufacturing a tie-breaker from popularity.
Evidence belongs beside a market only when it carries a supported explicit market identifier, and opportunity records originate from the configured watchlist. A conflicting story should not be attached to every contract that shares a person or topic. Verify outcome side, date, threshold, closing condition, and resolution authority for the exact market ID. Unlinked evidence can remain in the source record while a user investigates; forcing a relationship would create false precision and weaken provenance.
Preserve the opportunity’s first-recorded trigger price and show the current public price with separate timestamps. The movement may help prioritize a human review, but it does not establish which source is correct or prove that either report caused repricing. Inspect public bid, ask, spread, and visible depth when available so a thin book is not mistaken for broad consensus. Market prices aggregate participant behavior; they are context for the evidence review, not an oracle that settles the underlying fact.
When a reviewer is ready, save an attributable note describing the accepted facts, disputed points, uncertainty, and sources consulted. An in-app approve or reject event is a separate authenticated record; the configured price-move guard may block it if market context changed. A paper order, if the user creates one, is another simulation subject to current notional, exposure, and concentration checks. None of these records resolves the factual dispute or performs a real Polymarket transaction.
If an official correction arrives, add the new evidence and a later journal entry instead of editing history to make the earlier decision look cleaner. Review alert attempts separately if the correction failed to reach a configured destination, and use manual retry where appropriate. This append-only habit makes it possible to examine how the team handled uncertainty at the time. TEMIRIN organizes the record; users remain accountable for verification, interpretation, eligibility, capital, and any real venue decision.
Teams should also agree on what an unresolved review means operationally. One reviewer may wait for an official document, another may reject the thesis because the contract wording cannot be reconciled, and a third may record no action while monitoring continues. Preserve those choices as attributable notes instead of forcing consensus through a numeric average. During a later audit, compare the cited records and timestamps first; judge the process that existed then rather than rewarding the conclusion that happened to match resolution.
TEMIRIN organizes deterministic source records and provenance so a human reviewer can compare canonical material and record a bounded conclusion or unresolved state.
Not necessarily. Reviewers should trace syndication and attribution to the common origin, while canonicalization and deduplication reduce repeated RSS or web records.
No. Trigger price, current price, spread, and visible depth provide public market context, not proof about the disputed real-world fact.
Store it as a later attributable evidence record and add a new journal note. Do not rewrite the earlier evidence or decision timeline.
