Methodology
How Parallax Sports investigates a story
We build cases, not takes. Every story is structured against a published method — the same five protocols and nine rigor rules our editors apply before anything ships. This page exists so readers can interrogate the work, and so the standard is the same whether the subject is a star player or a private citizen.
The five research protocols
1 · Separation
Output is split into a Research Plan (questions, databases, tasks) and Narrative Seeds (story beats, conditional on findings). The plan comes first; the narrative is conditional. We never write story beats as if research has been completed.
2 · Falsification-First
Every thesis names three specific kill-shot findings that would destroy it. Every research task is tagged CONFIRM or FALSIFY. At least 40 % of tasks must be FALSIFY. If we can't name three falsifying findings, the thesis is commentary — not investigation.
3 · Base-Rate
Every claim of "unusual," "rare," "first," "only," or "unprecedented" requires the base rate. No anomaly claim ships without the general rate it's measured against.
4 · Elephant in the Room
Every story names — by name — the most obvious objection a knowledgeable fan would raise, and the counter-case. Specific person, specific event, specific precedent.
5 · Verification Gate
A four-tier evidence ladder. Defaults are honest:
- 🔲 ASSERTED — Not checked. The default for everything until proven otherwise.
- 🟡 LIKELY — Plausible from existing knowledge, awaiting source.
- 🟠 SOURCED — Citation exists but hasn't been personally confirmed.
- ✅ VERIFIED — Only for facts confirmed against authoritative data (e.g., league statistics APIs). Nothing from training data alone earns ✅.
The Rigor Patch — nine pre-flight rules
The Rigor Patch was developed iteratively from observed failures — each rule prevents a specific mistake. Every story is checked against all nine before it ships.
- 1. Complete Data View. Anti-cherry-picking. We must show all relevant metrics, not just the thesis-supporting ones. If the thesis only works when metrics are excluded, we rewrite or narrow it.
- 2. Hidden Actor Specificity. Anti-conspiracy. The hidden actor must be a named person or organization. Banned: "the media," "the establishment," any "the [adjective] complex / industry / ecosystem." If we found a specific analyst, the actor is that analyst — not "the establishment."
- 3. Struggle Beat Integrity. Anti-ignored-evidence. If a finding disproves the thesis, we either rewrite the thesis, document why it survives the contradiction, or reclassify the piece. We don't ignore counter-evidence.
- 4. Evidence-Before-Thesis. We don't finalize a thesis on missing evidence. Gather first, theorize second.
- 5. Category Integrity. No apples-to-oranges comparisons — injury severity, guarantee types, playoff context, era-adjusted efficiency. Compare like to like, or surface the difference.
- 6. Double-Check Protocol. A four-question pre-flight before publish: stat check, actor check, struggle check, evidence check.
- 7. One Hidden Actor, Named. If multiple beneficiaries exist, we pick the one with the strongest evidence. Diffuse "front-office" framings are not allowed. UNRESOLVED beats a collective.
- 8. Private Individual Protection. Private citizens require all predicate facts verified before naming. Public figures (GMs, coaches, owners, agents) are fair game once the rules above are met. Priority order: public exec → agent → public figure → verified private individual → UNRESOLVED.
- 9. Match Hidden Actor to Angle. Different story angles have different hidden-actor shapes — Rise stories ask who believed first, Fall stories ask who triggered the decline, Investigative stories ask who benefits from the consensus. If no individual fits, we publish UNRESOLVED rather than substitute a collective.
The seventeen failure modes
The protocols above were derived from cataloguing seventeen specific failure modes — pre-declared verifications on training-data claims, invented statistics, no falsification, cherry-picked reference classes, missing base rates, collective hidden actors, script-disguised-as-research, commentary dressed as investigation, chronology-not-causation, straw-man counterarguments, missed elephants, unrealistic research tasks, single-stat theses, no Plan B, era-blind comparisons, conspiracy-level causal claims, and confirm-only research plans. If any of those show up in a piece, the protocols weren't followed.
What this means for the reader
When we name a person, we've named them after the rules above. When we cite a statistic, it's tagged on the verification ladder. When the evidence doesn't resolve, we say UNRESOLVED rather than reach for a tidier story. The method is the brand. Reading our editorial standards page tells you what we publish; reading this page tells you how.