FA-100611 / Mixture of experts dispatch / Open access
Dropped token IDs vanish from the dispatch result · case 01
Dropped token IDs vanish from the dispatch result.
ROOT CAUSE
The output reports only successful assignments without explicit dropped tokens.
THE FAILURE
The output reports only successful assignments without explicit dropped tokens.
Unsuccessful approach: The attempted admission fix accounts for capacity overflow but never tries a token’s alternate expert.
Case contract
Given ranked token choices and per-expert capacity, return assignments plus dropped token IDs; try each choice until a slot is available.
Why this case matters
A deterministic toy router makes this dispatch decision independently inspectable.
1 / The failure
Exit 1"""Failure Map reference implementation. Python standard library only."""
import json
N = 1
observations = []
def solve(choices,capacity):
out=[]
for i,row in enumerate(choices):
if row: out.append([i,sorted(row,key=lambda x:(-x[1],x[0]))[0][0]])
return [out,[]]
def check(label, actual, expected):
observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
e='e'+str(N); f='f'+str(N); rows=[[(e,2)],[(e,1)],[],[(f,3)]]
check('overflow listed',solve(rows,1),[[[0,e],[3,f]],[1,2]])
check('attempt needs fallback',solve([[(e,4),(f,3)],[(e,2),(f,1)]],1),[[[0,e],[1,f]],[]])
check('all capacity',solve(rows,4),[[[0,e],[1,e],[3,f]],[2]])
check('zero capacity',solve(rows,0),[[],[0,1,2,3]])
check('empty batch',solve([],N),[[],[]])
check('all empty',solve([[],[]],1),[[],[0,1]])
check('fallback route',solve([[(e,5),(f,4)],[(e,3),(f,2)]],1),[[[0,e],[1,f]],[]])
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 |
|---|---|---|---|
| overflow listed | [[[0, 'e1'], [1, 'e1'], [3, 'f1']], []] | [[[0, 'e1'], [3, 'f1']], [1, 2]] | Failed |
| attempt needs fallback | [[[0, 'e1'], [1, 'e1']], []] | [[[0, 'e1'], [1, 'f1']], []] | Failed |
| all capacity | [[[0, 'e1'], [1, 'e1'], [3, 'f1']], []] | [[[0, 'e1'], [1, 'e1'], [3, 'f1']], [2]] | Failed |
| zero capacity | [[[0, 'e1'], [1, 'e1'], [3, 'f1']], []] | [[], [0, 1, 2, 3]] | Failed |
| empty batch | [[], []] | [[], []] | Passed |
| all empty | [[], []] | [[], [0, 1]] | Failed |
| fallback route | [[[0, 'e1'], [1, 'e1']], []] | [[[0, 'e1'], [1, 'f1']], []] | Failed |
SHA-256 / ff550a91015f057784b87c7d7cac53b1e0e433f0b833430d4d3f3bdcae79147c
2 / The unsuccessful fix
Exit 1"""Failure Map reference implementation. Python standard library only."""
import json
N = 1
observations = []
def solve(choices,capacity):
out=[]; dropped=[]; used={}
for i,row in enumerate(choices):
if not row: dropped.append(i); continue
e=sorted(row,key=lambda x:(-x[1],x[0]))[0][0]
if used.get(e,0)<capacity: out.append([i,e]); used[e]=used.get(e,0)+1
else: dropped.append(i)
return [out,dropped]
def check(label, actual, expected):
observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
e='e'+str(N); f='f'+str(N); rows=[[(e,2)],[(e,1)],[],[(f,3)]]
check('overflow listed',solve(rows,1),[[[0,e],[3,f]],[1,2]])
check('attempt needs fallback',solve([[(e,4),(f,3)],[(e,2),(f,1)]],1),[[[0,e],[1,f]],[]])
check('all capacity',solve(rows,4),[[[0,e],[1,e],[3,f]],[2]])
check('zero capacity',solve(rows,0),[[],[0,1,2,3]])
check('empty batch',solve([],N),[[],[]])
check('all empty',solve([[],[]],1),[[],[0,1]])
check('fallback route',solve([[(e,5),(f,4)],[(e,3),(f,2)]],1),[[[0,e],[1,f]],[]])
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 |
|---|---|---|---|
| overflow listed | [[[0, 'e1'], [3, 'f1']], [1, 2]] | [[[0, 'e1'], [3, 'f1']], [1, 2]] | Passed |
| attempt needs fallback | [[[0, 'e1']], [1]] | [[[0, 'e1'], [1, 'f1']], []] | Failed |
| all capacity | [[[0, 'e1'], [1, 'e1'], [3, 'f1']], [2]] | [[[0, 'e1'], [1, 'e1'], [3, 'f1']], [2]] | Passed |
| zero capacity | [[], [0, 1, 2, 3]] | [[], [0, 1, 2, 3]] | Passed |
| empty batch | [[], []] | [[], []] | Passed |
| all empty | [[], [0, 1]] | [[], [0, 1]] | Passed |
| fallback route | [[[0, 'e1']], [1]] | [[[0, 'e1'], [1, 'f1']], []] | Failed |
SHA-256 / 7c2adae4a05bdfb2a7cc3368d5b22a199f4e5c8910dca9df5b5ef3bf7d3497cb
HELD IN THE MEMBER ARCHIVE
The verified repair and its recorded checks are member-only.
This mechanism has 7 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.
Sign in to the archive ↗Verification & scope
Inputs are finite toy scores and routes; the contract does not model a production model, accelerator, or learned routing quality. 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 2026-09-30T16:48:37.646328+00:00.
Case digest / 1ba070b773d4ff2a9f2d7174147510d118de1fc8374c512ce95a47219b9e4692