---
title: Decide what a prop-market importer stores when it first sees a market already in progress
type: task
status: planned
description: >-
  A quarter of paired prop markets carry an OPEN stamp fabricated at close time — 317,993 of
  1,248,064, across all ten books at 3.8 to 16.3 percent. Inherited rather than migration-caused,
  and no task owned the importer side.
base_uri: user:task/league/own-the-fabricated-open-stamp-class.md
created_at: '2026-09-24T19:54:38.589Z'
entity_id: 3a3c0e9d-b4b1-4225-ae51-73a425d3c1d5
owner_identity_uri: █████████████████████████
public_read: true
tags:
  - user:tag/league-xo-football.md
  - user:tag/software-task.md
updated_at: '2026-09-24T19:54:38.589Z'
visibility_analyzed_at: '2026-11-05T23:49:47.023Z'
---

## Problem

Roughly a quarter of paired prop markets carry an OPEN stamp that was fabricated
at CLOSE time rather than observed when the market opened. Measured against the
pre-migration two-row grain: **317,993 of 1,248,064 paired markets, 25.5 percent**,
reproducing an independent 2026-09-17 measurement of 318,185 of 1,247,247 to
within 0.06 percent.

It spans **all ten books at 3.8 to 16.3 percent with no outlier**, so there is no
single importer to point at. It is a property of how the importers stamp, not of
one source.

## What is already ruled out

**The grain redesign neither created nor repaired it.** Read against the
rehearsal cluster's surviving pre-migration tables, the same predicate gives
337,830 of 3,455,545 selections carrying both sides (9.8 percent) against
production's post-migration 381,528 of 3,740,550 (10.2 percent). All four
open/close classes held their proportions across the cutover and none collapsed
into another. Dispositioned ACCEPTED-AND-RECORDED on
[[█████████████████████████-████-██████-█████-████████]] as inherited, which is
why it needs an owner here.

**Do not quote the 3,953 figure** that circulated in a handoff brief for this
class. It reproduces under no discriminator — 2+ observations gives 141,329, an
exists-probe at a different instant 182,456, market-seen-before-open 181,773,
seen-after-close 3,030. Use 381,528 with its denominator.

## Scope

The importers that write prop markets, and specifically the point at which each
decides an observation is the OPENING one. The defect is that a market first seen
late gets its first sighting recorded as the open, so the stored opening price is
not the price the market opened at.

Distinct from [[user:task/league/repair-prop-market-selection-endpoint-nulls.md]],
which is about endpoints that are NULL and recoverable. This one is about
endpoints that are PRESENT and wrong, and it is not recoverable from stored data
— the true opening price was never captured.

## Plan

- [ ] Re-measure the class against the current grain and record the predicate
  used, since the pre-migration tables production dropped on 2026-09-24 are gone
- [ ] Establish per importer where the opening stamp is decided, and which of the
  ten write an opening on first sighting regardless of how late that is
- [ ] Decide what an importer SHOULD store when it first sees a market already in
  progress — a null opening is honest where a fabricated one is not, and the
  endpoint-coverage check now grades null openings, so the two decisions interact
- [ ] Fix forward only. The historical class is unrecoverable and should be
  recorded as such rather than repaired

## Verification

Any fix is forward-looking, so the oracle is a cohort measurement: markets first
seen after their own kickoff, imported AFTER the change, must not carry an
opening stamp at or after that kickoff.

## Origin

Carried forward by `repl-ef5faa34` on 2026-09-24 from the grain-redesign task,
which measured and dispositioned the class but does not own the importers.