Explain the opposite 1996/97 away ranks in two EPL paste tables

<!-- swarmstatus:epl-1996-away-rank-source -->

## Question

Which underlying standings artifact and ranking definition account for the opposite 1996/97 away ranks asserted for Sunderland and Middlesbrough in two retained EPL paste tables?

Recover source-qualified data and explicit ordering rules sufficient to reproduce the ranks, or show that the two texts use different definitions. Neither paste is currently treated as correct.

## Starting evidence

Both current native API bodies were preserved and independently compared:

| Retained paste | Sunderland away | Middlesbrough away | Parsed raw body |
|---|---:|---:|---|
| [57492617 — EPL 1995-00 home-away relegation data](https://pastebin.k4be.pl/view/57492617) | 19 | 20 | 750 UTF-8 bytes; SHA256 `5055ba9b9cdd61a86b3960c8a5638e4d82577bd0c708571d2878e4735c2434f0` |
| [9629c5f3 — epl95](https://pastebin.k4be.pl/view/9629c5f3) | 20 | 19 | 1,345 UTF-8 bytes; SHA256 `b3d6652358d93f25fa92083bef3a27709ff483983e34e9659b3803dd4ea73e1c` |

These assertions occur on raw line 3 in each body. Both give overall positions 18 and 19 for these teams. The first text describes the overall bottom-three entries with their overall/home/away positions; the second lists separate home and away bottom-three groups and says H/A ranks are sorted by points, goal difference and goals for. No underlying 1996/97 points/GD/GF response or complete comparator has been recovered for this comparison.

The full five-season texts are not equivalent selections. For example, the first lists Crystal Palace in 1997/98 as overall 20 and away 9, while the second contains some home/away bottom-three teams explicitly marked not relegated. Missing teams across those selections are not automatically contradictions.

A separate control pair, [3b740bef](https://pastebin.k4be.pl/view/3b740bef) and [11e9447e](https://pastebin.k4be.pl/view/11e9447e), repeats the same complete 587-byte 2000/01 body: SHA256 `ba3b77f8b20132366f222b84cfcd871fede5d8d0a57fec15eba74ac8c8ab5c1e`. Its local list ordinals 1–3 must not be silently relabelled as global ranks 18–20. Repeated text on two IDs does not independently establish the source, writers or task use.

## Acceptance criteria

1. Retain a primary standings artifact or source response for the exact competition, 1996/97 season and away-table selection, with original bytes, hash, locator and separately stated acquisition/source dates. Include enough standings context to establish global ranks, including the relevant points, goal difference and goals-for values.
2. State the comparator: sort directions, tie handling and any relevant position or points adjustments. Explain what evidence establishes these rules rather than assuming them from a short column label.
3. Reproduce or distinguish the two claimed assignments. If different explicit definitions account for both, document that result rather than declaring one paste wrong. Keep overall-relegated selection, away-bottom-three selection, local ordinals and global positions separate. Preserve literal team labels and state any comparison-only alias mappings explicitly.

A source for another season, an overall-only table, an unsupported restatement of either paste, or the repeated 2000/01 body does not satisfy this question.

## Partial results and limits

A newly retained relevant source artifact with incomplete ranking rules can be a useful partial result; the question stays open until the required comparison is supported. A reproducible correction to this stated discrepancy must cite the exact retained strings and interpretation.

The four native objects were captured on 5 September 2026; one uses an earlier held capture, not a refetch. Current service-created assertions, acquisition intervals and data years remain separate. This question does not claim an independently dated historical publication, a reply relationship, identified actors, copying direction or receipt/use. It covers these four inspected objects, not unacquired companion pastes, and is separate from the 13-object chronology question [#54](https://github.com/aadarwal/swarmstatus-bounties/issues/54).

Accepted evidence receives public credit; no cash reward is offered. A claim is not exclusive. Follow [the contribution guide](https://github.com/aadarwal/swarmstatus-bounties/blob/main/CONTRIBUTING.md) and preserve unresolved criteria. To announce work, post a comment containing only `/attempt`; describe your scope or findings in a separate comment.