FA-99596 / RAG context assembly / Open access

Neighboring context expands on the wrong side of an anchor · case 01

Neighboring context expands on the wrong side of an anchor.

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

ROOT CAUSE

A local evidence window uses a forward offset where the contract defines a centered window.

THE FAILURE

A local evidence window uses a forward offset where the contract defines a centered window.

Unsuccessful approach: The attempted exclusive upper bound is computed without including the anchor position, dropping the anchor or its last right neighbor.

Case contract

Given ordered texts, anchor index, and left/right counts, return the inclusive anchor window clipped to list bounds.

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, left, right):
    return items[anchor:anchor+left+right+1]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
items=['a','b','c','d','e']
check('centered', solve(items,2,1,2), ['b','c','d','e'])
check('left edge', solve(items,0,2,1), ['a','b'])
check('right edge', solve(items,4,1,2), ['d','e'])
check('single anchor', solve(items,2,0,0), ['c'])
check('empty input', solve([],0,1,1), [])
check('whole window', solve(items,2,9,9), items)
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
centered	`['c', 'd', 'e']`	`['b', 'c', 'd', 'e']`	Failed
left edge	`['a', 'b', 'c', 'd']`	`['a', 'b']`	Failed
right edge	`['e']`	`['d', 'e']`	Failed
single anchor	`['c']`	`['c']`	Passed
empty input	`[]`	`[]`	Passed
whole window	`['c', 'd', 'e']`	`['a', 'b', 'c', 'd', 'e']`	Failed

SHA-256 / 8fa14652e8bc5c9f636e414adeaa67c6f04c97c9233294390b5e1b019fbca046

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(items, anchor, left, right):
    lo=max(0,anchor-left); hi=min(len(items),anchor+right)
    return items[lo:hi]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
items=['a','b','c','d','e']
check('centered', solve(items,2,1,2), ['b','c','d','e'])
check('left edge', solve(items,0,2,1), ['a','b'])
check('right edge', solve(items,4,1,2), ['d','e'])
check('single anchor', solve(items,2,0,0), ['c'])
check('empty input', solve([],0,1,1), [])
check('whole window', solve(items,2,9,9), items)
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
centered	`['b', 'c', 'd']`	`['b', 'c', 'd', 'e']`	Failed
left edge	`['a']`	`['a', 'b']`	Failed
right edge	`['d', 'e']`	`['d', 'e']`	Passed
single anchor	`[]`	`['c']`	Failed
empty input	`[]`	`[]`	Passed
whole window	`['a', 'b', 'c', 'd', 'e']`	`['a', 'b', 'c', 'd', 'e']`	Passed

SHA-256 / 7366e767cbf5afa0a6c58390cda0d50238a7a5c9810c72bd97caf1f6d9e03b22

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 2026-09-30T16:48:25.834982+00:00.

Case digest / 00f6ad798f53c317eadd66c4e0b66eb5fc27be00f5065cfd7fa289a6e362b28e