A practical guide to repeatable deterministic evidence research in TEMIRIN using configured records, explicit market identity, current controls, and human review.
August 3, 2026 · 3 min read
Repeatable Polymarket research begins with the same identity check every time. Record the explicit market ID, outcome ID, exact question, close time, status, and stated resolution source before reading the newest evidence. Titles can change and similar contracts can coexist, so a copied headline is not a reliable join key. A saved watchlist record gives the workflow a bounded starting point without claiming that the system searches or ranks the full market universe.
Define what would answer the research question and what would remain ambiguous. For an economic release, specify the measure and reference period. For an election, distinguish projections, recounts, and certification. For a crypto event, name the network, product, or filing stage. This pre-commitment makes later notes comparable because each reviewer is testing the same contract interpretation rather than changing the question after a favorable source appears.
Use the workspace's configured RSS and HTTPS web sources, plus provider-configured Telegram or X readers when available. For each stored item, inspect publisher or provider identity, canonical reference where present, supplied publication time, observation time, and operational state. RSS and web canonicalization can reduce repeated copies, but a duplicate should never count as additional confirmation. A failed source remains a visible gap that may require direct manual verification.
Keep unlinked records in the evidence collection when they are useful but lack an explicit supported market relationship. Do not force a connection based on similar words. Open primary documents behind summaries, distinguish corrections from new publications, and write down why a secondary report deserves attention. The resulting evidence set is bounded and attributable: another reviewer can see what was collected, what was excluded, and which limitations affected the cycle.
Open the deterministic brief and inspect its reliability, relevance, direction, freshness, source-tier, and tagged-market inputs. Then read the original evidence and venue rules. Compare the stored trigger price with the latest available public price, quote time, bid, ask, spread, and visible depth. Use the same order of operations so a dramatic price move does not cause the reviewer to skip provenance or reinterpret weak evidence as certainty.
Record the conclusion in the authenticated workspace with the evidence consulted, unresolved question, and reason for approve, reject, or no decision. Alert delivery belongs to a separate ledger and cannot stand in for this action. If a user enters a paper request, implemented checks cover the per-order limit, rolling twenty-four-hour paper total, per-market exposure, and portfolio concentration. The simulated record remains distinct from research and from real venue activity.
At the end of a review period, choose a sample that includes approved, rejected, unresolved, and unlinked records. Reconstruct each case from the saved evidence, timestamps, market identity, price observations, policy result, actor, and note. Compare whether required fields were present and whether uncertainty was documented before the outcome became known. This audit tests repeatability; it should not turn a small retrospective sample into a performance promise.
Record one correction prospectively. Repair a broken source, clarify a watchlist label, update an account owner, simplify a note field, or change who can access exports. Preserve earlier records as they were observed. A repeatable system earns trust when another authorized reviewer can reproduce the path and identify the same data gaps, even if that reviewer reaches a different interpretation of the evidence or chooses a different real-world action. Name the owner, effective date, and next review sample before closing the change. That small control makes improvement measurable without changing the historical evidence. Record the result for audit.
Use the same bounded checklist for source provenance, explicit market identity, timestamps, trigger and current prices, policy state, disposition, and attributable notes across a mixed sample of cases.
Leave the field visibly unavailable and retain the source record in its proper context. The reviewer can document the coverage gap without filling it from assumptions or changing the original observation.
Include favorable, unfavorable, blocked, and no-action records. Trace each displayed value back to stored evidence and keep collection state, in-app decisions, alert attempts, paper records, and notes separately labelled.
