FA-99621 / RAG context assembly / Open access

Repeated chunk identifiers consume the evidence budget twice · case 01

Repeated chunk identifiers consume the evidence budget twice.

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

ROOT CAUSE

The assembler treats repeated references to the same chunk as independent evidence.

THE FAILURE

The assembler treats repeated references to the same chunk as independent evidence.

Unsuccessful approach: Deduplicating by text merges distinct chunks that happen to have equal wording.

Case contract

Rows are [chunk_id,text]. Preserve first occurrence by ID and return its text.

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(rows):
    return [text for ident,text in rows]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
i='id'+str(N)
check('same id distinct text', solve([[i,'first'],[i,'updated']]), ['first'])
check('same text distinct IDs', solve([[i,'same'],['other','same']]), ['same','same'])
check('different ids', solve([[i,'a'],['b','b']]), ['a','b'])
check('empty', solve([]), [])
check('singleton', solve([[i,'x']]), ['x'])
check('repeated nonadjacent', solve([[i,'x'],['j','y'],[i,'z']]), ['x','y'])
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
same id distinct text	`['first', 'updated']`	`['first']`	Failed
same text distinct IDs	`['same', 'same']`	`['same', 'same']`	Passed
different ids	`['a', 'b']`	`['a', 'b']`	Passed
empty	`[]`	`[]`	Passed
singleton	`['x']`	`['x']`	Passed
repeated nonadjacent	`['x', 'y', 'z']`	`['x', 'y']`	Failed

SHA-256 / dea221848068aaab577eee26bf4814efc6ffe4fa2056d6aebfa774b98243dab4

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(rows):
    seen=set(); out=[]
    for ident,text in rows:
        if text not in seen: seen.add(text); out.append(text)
    return out
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
i='id'+str(N)
check('same id distinct text', solve([[i,'first'],[i,'updated']]), ['first'])
check('same text distinct IDs', solve([[i,'same'],['other','same']]), ['same','same'])
check('different ids', solve([[i,'a'],['b','b']]), ['a','b'])
check('empty', solve([]), [])
check('singleton', solve([[i,'x']]), ['x'])
check('repeated nonadjacent', solve([[i,'x'],['j','y'],[i,'z']]), ['x','y'])
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
same id distinct text	`['first', 'updated']`	`['first']`	Failed
same text distinct IDs	`['same']`	`['same', 'same']`	Failed
different ids	`['a', 'b']`	`['a', 'b']`	Passed
empty	`[]`	`[]`	Passed
singleton	`['x']`	`['x']`	Passed
repeated nonadjacent	`['x', 'y', 'z']`	`['x', 'y']`	Failed

SHA-256 / a66d5f6cdfc03a9bad00ed224bd578b7bf4fe9cb4f3ed7f6bceb53938527aa2c

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 2026-09-30T16:48:26.270258+00:00.

Case digest / 487784c0f08c0e0e06d3f5e9eed3a4d7a0597f1901ae8416b88a73fd0ff71c08