FA-99626 / RAG context assembly / Open access

Centered evidence window is reversed while crossing a boundary · case 01

Centered evidence window is reversed while crossing a boundary.

Verified by executionVariant 1 · 6 checks per implementationDownload source bundle ↓JSON ↗

ROOT CAUSE

Left-side chunks are gathered backwards and concatenated without restoring document order.

THE FAILURE

Left-side chunks are gathered backwards and concatenated without restoring document order.

Unsuccessful approach: Reversing the left slice places neighboring evidence in reverse document order.

Case contract

Given ordered items, anchor and radius, return up to radius items on each side plus anchor in original order.

Why this case matters

A small deterministic evidence assembly model makes context budgeting behavior inspectable.

1 / The failure

Exit 1

"""Failure Map reference implementation. Python standard library only."""
import json

N = 1
observations = []
def solve(items, anchor, radius):
    lo=max(0,anchor-radius); hi=min(len(items),anchor+radius+1)
    return items[anchor:anchor+1]+items[lo:anchor]+items[anchor+1:hi]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
items=['p','q','r','s','t']
check('center order', solve(items,2,2), ['p','q','r','s','t'])
check('left edge', solve(items,0,2), ['p','q','r'])
check('right edge', solve(items,4,2), ['r','s','t'])
check('zero radius', solve(items,2,0), ['r'])
check('empty', solve([],0,2), [])
check('radius one', solve(items,3,1), ['r','s','t'])
print(json.dumps({"observations": observations, "passed": all(x["passed"] for x in observations)}, ensure_ascii=False))
raise SystemExit(0 if all(x["passed"] for x in observations) else 1)

Boundary fixture	Actual	Expected	Outcome
center order	`['r', 'p', 'q', 's', 't']`	`['p', 'q', 'r', 's', 't']`	Failed
left edge	`['p', 'q', 'r']`	`['p', 'q', 'r']`	Passed
right edge	`['t', 'r', 's']`	`['r', 's', 't']`	Failed
zero radius	`['r']`	`['r']`	Passed
empty	`[]`	`[]`	Passed
radius one	`['s', 'r', 't']`	`['r', 's', 't']`	Failed

SHA-256 / 4addb244f96df69804d90eee2630b27f44eff4c9f220538dbfe1064b80a6ab22

2 / The unsuccessful fix

Exit 1

"""Failure Map reference implementation. Python standard library only."""
import json

N = 1
observations = []
def solve(items, anchor, radius):
    lo=max(0,anchor-radius); hi=min(len(items),anchor+radius+1)
    return items[lo:anchor][::-1]+items[anchor:hi]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
items=['p','q','r','s','t']
check('center order', solve(items,2,2), ['p','q','r','s','t'])
check('left edge', solve(items,0,2), ['p','q','r'])
check('right edge', solve(items,4,2), ['r','s','t'])
check('zero radius', solve(items,2,0), ['r'])
check('empty', solve([],0,2), [])
check('radius one', solve(items,3,1), ['r','s','t'])
print(json.dumps({"observations": observations, "passed": all(x["passed"] for x in observations)}, ensure_ascii=False))
raise SystemExit(0 if all(x["passed"] for x in observations) else 1)

Boundary fixture	Actual	Expected	Outcome
center order	`['q', 'p', 'r', 's', 't']`	`['p', 'q', 'r', 's', 't']`	Failed
left edge	`['p', 'q', 'r']`	`['p', 'q', 'r']`	Passed
right edge	`['s', 'r', 't']`	`['r', 's', 't']`	Failed
zero radius	`['r']`	`['r']`	Passed
empty	`[]`	`[]`	Passed
radius one	`['r', 's', 't']`	`['r', 's', 't']`	Passed

SHA-256 / 508d93bd00cf8b5f374be6cf4bd2df172ef262b93e3eb7d0cd7a878ec2134237

HELD IN THE MEMBER ARCHIVE

The verified repair and its recorded checks are member-only.

This mechanism has 6 recorded checks per implementation. The open-access tier publishes the failure and the unsuccessful fix; the repaired source that passes every check, and the observations that prove it, are available to members.

Every case sharing this mechanism uses the same contract and the same repair, so this one record is held back for all of them.

Member access is invitation-based. Sign in with your invited account to inspect the repair.

Verification & scope

Toy inputs and token counts only; this does not claim compatibility with a tokenizer, prompt format, or production retrieval system. This reproducer isolates one failure mechanism. Results cover the supplied fixtures. Variants within a family share a test contract and should remain grouped when constructing evaluation splits. Related mechanisms with a shared evaluation_group must also remain together; these controlled models are not independent production incidents.

Observations recorded using Python 3.12.14 at 2026-09-30T16:48:26.270044+00:00.

Case digest / 35b2ae254220d6fa949585b87ee5d0076e71d9ceee0aa9b79d78ebdbe72acf7b