A practical guide to deterministic evidence provenance in TEMIRIN using configured records, explicit market identity, current controls, and human review.
August 3, 2026 · 6 min read
Evidence provenance answers where a record came from, when it was observed, how it entered the workspace, and which original material a reviewer can inspect. In prediction-market research, those details matter because a short headline can outlive a correction while a price changes quickly. A useful record therefore keeps the configured source, publisher or provider identity, canonical URL or reference, observed time, supplied publication or event time when available, workspace scope, and safe processing state alongside the content used in review.
Provenance does not guarantee truth. A perfectly captured article may quote an anonymous claim, repeat another publisher, or describe a different jurisdiction than the market. TEMIRIN organizes deterministic source records with links and timestamps. The human reviewer follows citations, compares the precise claim with the contract, and records uncertainty. The product’s value is a traceable starting point that can be audited later, with every conclusion attributable to the reviewer and source material.
Configured RSS and HTTPS web readers can preserve canonical records and apply content deduplication. Where a provider connection is configured, Telegram or X readers can add source messages with their own provider references. These inputs do not share identical metadata, so a normalized display should retain the source type and original identifier. A web page, feed item, channel post, and social post may all discuss the same event while offering different access, revision, authorship, and timing characteristics.
Document the configuration that produced the record without exposing credentials. Store a safe source label and workspace relationship, never a private token, password, seed phrase, or recovery phrase in evidence, notes, alerts, or exports. If a reader fails, preserve the failure in operational history rather than presenting the last successful item as fresh. Source availability is a condition of collection; it is not evidence for or against a market outcome. This distinction makes failures diagnosable without manufacturing research conclusions.
Canonicalization helps identify the stable page behind tracking parameters, redirects, or feed variants. Content deduplication helps prevent repeated RSS and web material from filling a review with copies. Neither function proves independent confirmation. Five domains may repeat one wire report, while one official page may receive a material correction at the same address. Reviewers should inspect the attribution chain and distinguish the number of stored records from the number of independent primary sources.
When a publisher revises a claim, keep the later state attributable and record how it differs from the earlier material. Do not silently edit an old decision note to match the correction. A reviewer can add a new journal entry that explains the changed source, its time, and its effect on the thesis. This append-only pattern supports honest post-event analysis. It also prevents a corrected canonical page from making the team appear to have known information that was not present during the original review.
TEMIRIN derives opportunity records from markets already configured in the workspace watchlist. A stored evidence item appears beside a market only when a supported explicit market identifier is supplied. Evidence without an identifier remains useful in its original source view. Adding a relationship requires the reviewer to verify the exact market question, outcome, date, threshold, and resolution terms.
After an explicit link exists, the reviewer still verifies the exact outcome, date, threshold, closing condition, and resolution source. A report about a national measure may not answer a regional contract; a preliminary total may not satisfy a certified-results clause. Preserve the identifier and reviewer reasoning in the record so another team member can reconstruct why the evidence was considered. Explicit identity creates an auditable join, not proof that the source resolves the market question.
The watchlist-derived opportunity preserves its first-recorded trigger price, and later synchronization can show a current price with a separate timestamp. Public bid, ask, spread, and visible depth may add market-structure context. None of those values validates the source or proves causation. A price can reflect information outside the configured coverage, and visible liquidity can change before any outside order. Keep evidence provenance and market observations side by side without merging them into a predicted edge.
A configured public wallet can provide read-only snapshots, positions, PnL context, and concentration context. Display the snapshot time and address reference safely, while avoiding claims about ownership intent, complete cost basis, or why a position changed. Current policy states explain eligibility, source or category blocks, maximum-stake overrides, free-capital percentage, the price-move guard, and kill switch around review and paper records.
An alert rule may send selected review context through configured email, Telegram, Slack, or outbound webhook delivery. The ledger records attempts and supports manual retry, preserving the alert’s relationship to the evidence record. Delivery success means the destination accepted transport; it does not acknowledge the item or create an approval. Approve and reject are authenticated in-app actions, where the configured price-move guard can block a decision when current conditions differ from the saved context.
A user may separately enter a paper order, which receives per-order, rolling twenty-four-hour, per-market exposure, and portfolio-concentration checks. Store its simulation status and later mark separately from source evidence and public-wallet activity. Journal notes, analytics, and export can then connect the reviewer’s thesis, decision, paper result, and later lesson while keeping the result explicitly simulated. Every transaction at Polymarket stays with the user, while the provenance chain explains what the workspace actually knew.
Test the chain with a sampled record from each configured source type. Begin at the journal or decision, follow the opportunity identifier to the exact watchlist market, open every explicitly linked evidence item, and verify canonical references and times. Then trace any alert attempts and paper record without treating them as evidence. Record missing or inconsistent fields as product-quality issues. A provenance system is credible when another authorized reviewer can reconstruct the stored path and clearly see where human interpretation began.
Exports should preserve enough identity for independent review while respecting workspace access and data minimization. Include safe record IDs, source labels, canonical references, market IDs, labeled timestamps, and record types; exclude secrets and avoid flattening evidence, decisions, simulations, and wallet observations into one ambiguous row. When retention or deletion changes availability, show that state honestly. Provenance is not unlimited storage: it is deliberate custody of the fields necessary to understand a research decision for the period the workspace is authorized to keep them.
Assign a reviewer to periodic provenance sampling and retain the sample definition with the findings. Check one corrected page, one repeated feed item, one provider record, and one item with an explicit market identifier. The goal is to confirm that each display can be traced to its stored origin and that later review notes remain visibly human-authored. Any mismatch should become a specific data-quality issue with an owner, affected record, and prospective correction.
Useful fields include source configuration, publisher or provider, canonical URL or reference, observed time, supplied publication time when available, workspace identity, and safe processing status.
No. It shows origin and history. A human reviewer must inspect the primary material, attribution, contract wording, and uncertainty.
The market must be in the configured watchlist and the evidence must carry a supported explicit market identifier that the reviewer verifies against the contract.
No. Provider-configured readers collect source records where available. Email, Telegram, Slack, and webhook destinations are separate outbound alert-delivery paths.
No. TEMIRIN is a read-only research and paper-decision workspace. All real transactions and wallet control remain outside the product.
