FA-99581 / RAG context assembly / Open access

Context budget double-counts shared token overlap · case 01

Context budget double-counts shared token overlap.

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

ROOT CAUSE

The assembler charges each chunk’s full token length even when adjacent chunks share a prefix boundary.

THE FAILURE

The assembler charges each chunk’s full token length even when adjacent chunks share a prefix boundary.

Unsuccessful approach: Subtracting a single shared token misses multi-token repeated boundaries.

Case contract

Token-list chunks append under budget; each later chunk incurs only tokens after maximal exact suffix-prefix overlap. Return assembled tokens.

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(chunks, budget):
    out=[]
    for c in chunks: out.extend(c)
    return out[:budget]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
a='a'+str(N); b='b'+str(N); c='c'+str(N)
check('overlap permits fit', solve([[a,b,c],[b,c,'d']],4), [a,b,c,'d'])
check('cannot partially append', solve([[a,b],[b,c]],2), [a,b])
check('empty', solve([],N), [])
check('zero budget', solve([[a]],0), [])
check('no overlap exact', solve([[a],[b]],2), [a,b])
check('full overlap', solve([[a,b],[a,b]],2), [a,b])
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
overlap permits fit	`['a1', 'b1', 'c1', 'b1']`	`['a1', 'b1', 'c1', 'd']`	Failed
cannot partially append	`['a1', 'b1']`	`['a1', 'b1']`	Passed
empty	`[]`	`[]`	Passed
zero budget	`[]`	`[]`	Passed
no overlap exact	`['a1', 'b1']`	`['a1', 'b1']`	Passed
full overlap	`['a1', 'b1']`	`['a1', 'b1']`	Passed

SHA-256 / 2bcf5f7d2b94ed9f0569a943ab4b37ac7f861377715b28ecbf7fd0501dfb951c

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(chunks, budget):
    out=[]
    for c in chunks:
        k=1 if out and c and out[-1]==c[0] else 0
        out.extend(c[k:])
    return out[:budget]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
a='a'+str(N); b='b'+str(N); c='c'+str(N)
check('overlap permits fit', solve([[a,b,c],[b,c,'d']],4), [a,b,c,'d'])
check('cannot partially append', solve([[a,b],[b,c]],2), [a,b])
check('empty', solve([],N), [])
check('zero budget', solve([[a]],0), [])
check('no overlap exact', solve([[a],[b]],2), [a,b])
check('full overlap', solve([[a,b],[a,b]],2), [a,b])
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
overlap permits fit	`['a1', 'b1', 'c1', 'b1']`	`['a1', 'b1', 'c1', 'd']`	Failed
cannot partially append	`['a1', 'b1']`	`['a1', 'b1']`	Passed
empty	`[]`	`[]`	Passed
zero budget	`[]`	`[]`	Passed
no overlap exact	`['a1', 'b1']`	`['a1', 'b1']`	Passed
full overlap	`['a1', 'b1']`	`['a1', 'b1']`	Passed

SHA-256 / 35376a7c71f9f98245758a9bc3472c8e2c9ab6749e10c5aacef384a3c8b4c5fd

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 2026-09-30T16:48:25.802618+00:00.

Case digest / ba5634a63f4b36c420fa0051003a1b4e52cd1cd5cf0ea656545866606f567150