Skip to main content
T·M TOMFORMER live research instrument Patent pending

Answer / Reason / live composed reads

Follow every entity and relation in a two-hop and a three-hop answer.

These are connected corpus questions, not one opaque prompt. The answer from each question names the subject of the next. Every arm receives the same constituent questions with the abstention gate off; TOMX additionally returns an opaque source receipt for each supported read. The final answer is accepted only if every intermediate top-1 identity is right.

Two-hop 2 independently scored constituent questions

Where was Shafrira Goldwasser’s doctoral advisor born?

Gold final identity: Caracas Q1533

  1. 1 Shafrira Goldwasser Q11609 → doctoral advisor P184 → Manuel Blum Q92626

    Constituent question dev:0095: Shafrira Goldwasser's doctoral advisor was ___.

    TOMX

    TOP-1 ×
    South Tyrol not in top 5
    hop decision receipt rcp_e5cd85f8867bb678dfdf
    structured source read rcp_869b8c6e235af29454bc

    matched TX

    TOP-1 ×
    South Tyrol not in top 5
    hop decision receipt rcp_ab946c8c33b6f2be6d40
    no source receipt: knowledge is stored in weights

    deployed RAG

    TOP-1 ×
    Latsch not in top 5
    hop decision receipt rcp_017b6026868a7419cf94
    retrieved context rcp_79ce0702547c8a50cb7b
    retrieved context rcp_7805652818af07c0ca1f
    retrieved context rcp_eac4d828629f1701d79b
    retrieved context rcp_b58d0a334bf342d6015b
    retrieved context rcp_437908e2aed683ebeddd

    5 retrieved context receipt(s); retrieval is evidence of context, not a relation-chain proof.

  2. 2 Manuel Blum Q92626 → place of birth P19 → Caracas Q1533

    Constituent question dev:2301: Manuel Blum was born in ___.

    TOMX

    TOP-1 ×
    state of Venezuela rank 4
    hop decision receipt rcp_671bfd065fc3f2021340
    structured source read rcp_1d83a0b9226ba928163a

    matched TX

    TOP-1 ×
    Paris not in top 5
    hop decision receipt rcp_328f929bc0f1487edf16
    no source receipt: knowledge is stored in weights

    deployed RAG

    TOP-1 ×
    Manuel Blum not in top 5
    hop decision receipt rcp_b48756a9efd6929369ed
    retrieved context rcp_b92a7615731657a23bb7
    retrieved context rcp_265abb3f38da22210d97
    retrieved context rcp_e57453ee54d1aa17e979
    retrieved context rcp_fc9d1a09d77c9a55439e
    retrieved context rcp_1bd5802dc722ca02439f

    5 retrieved context receipt(s); retrieval is evidence of context, not a relation-chain proof.

Live chain score

TOMX chain FAIL
hop @1
0/2 · 0.0%
hop @5
1/2 · 50.0%

Chain pass requires every constituent hop at rank 1. Label fallbacks: 0.

matched TX chain FAIL
hop @1
0/2 · 0.0%
hop @5
0/2 · 0.0%

Chain pass requires every constituent hop at rank 1. Label fallbacks: 0.

deployed RAG chain FAIL
hop @1
0/2 · 0.0%
hop @5
0/2 · 0.0%

Chain pass requires every constituent hop at rank 1. Label fallbacks: 0.

Three-hop 3 independently scored constituent questions

Where was the doctoral advisor of Shafrira Goldwasser’s doctoral advisor born?

Gold final identity: New York City Q60

  1. 1 Shafrira Goldwasser Q11609 → doctoral advisor P184 → Manuel Blum Q92626

    Constituent question dev:0095: Shafrira Goldwasser's doctoral advisor was ___.

    TOMX

    TOP-1 ×
    South Tyrol not in top 5
    hop decision receipt rcp_e5cd85f8867bb678dfdf
    structured source read rcp_869b8c6e235af29454bc

    matched TX

    TOP-1 ×
    South Tyrol not in top 5
    hop decision receipt rcp_ab946c8c33b6f2be6d40
    no source receipt: knowledge is stored in weights

    deployed RAG

    TOP-1 ×
    Latsch not in top 5
    hop decision receipt rcp_017b6026868a7419cf94
    retrieved context rcp_79ce0702547c8a50cb7b
    retrieved context rcp_7805652818af07c0ca1f
    retrieved context rcp_eac4d828629f1701d79b
    retrieved context rcp_b58d0a334bf342d6015b
    retrieved context rcp_437908e2aed683ebeddd

    5 retrieved context receipt(s); retrieval is evidence of context, not a relation-chain proof.

  2. 2 Manuel Blum Q92626 → doctoral advisor P184 → Marvin Minsky Q204815

    Constituent question dev:2300: Manuel Blum's doctoral advisor was ___.

    TOMX

    TOP-1 ×
    state of Venezuela not in top 5
    hop decision receipt rcp_fccc264153e08ab6bcae
    structured source read rcp_5ec4eba8ccc8b843b132

    matched TX

    TOP-1 ×
    Governor-General of the Philippines not in top 5
    hop decision receipt rcp_5f924e89351ec5221391
    no source receipt: knowledge is stored in weights

    deployed RAG

    TOP-1 ×
    Manuel Blum not in top 5
    hop decision receipt rcp_d566982cea47c529c4c3
    retrieved context rcp_b92a7615731657a23bb7
    retrieved context rcp_39991ebe33f45ff6c192
    retrieved context rcp_2f87f73c5937167eef40
    retrieved context rcp_c0c4bcf3693ede7103b9
    retrieved context rcp_171b7f6645553f4de44b

    5 retrieved context receipt(s); retrieval is evidence of context, not a relation-chain proof.

  3. 3 Marvin Minsky Q204815 → place of birth P19 → New York City Q60

    Constituent question dev:0561: Marvin Minsky was born in ___.

    TOMX

    TOP-1 ×
    genitive case not in top 5
    hop decision receipt rcp_a7bd05cc9b2d53624b9a
    structured source read rcp_cdeb33580dc135e91c3f

    matched TX

    TOP-1 ×
    Saint-Brice-sous-Forêt not in top 5
    hop decision receipt rcp_924c38997b300bfd8549
    no source receipt: knowledge is stored in weights

    deployed RAG

    TOP-1 ×
    Marvin Minsky not in top 5
    hop decision receipt rcp_064686f3d3b9dd8fcf6f
    retrieved context rcp_58d15bf9c6de57826a92
    retrieved context rcp_4b4592a589699c0c8dd7
    retrieved context rcp_8b6e56f47b0c10e24ba1
    retrieved context rcp_fc73279d5e9fe6b36d6f
    retrieved context rcp_e0d98ac1a57550d8fbac

    5 retrieved context receipt(s); retrieval is evidence of context, not a relation-chain proof.

Live chain score

TOMX chain FAIL
hop @1
0/3 · 0.0%
hop @5
0/3 · 0.0%

Chain pass requires every constituent hop at rank 1. Label fallbacks: 0.

matched TX chain FAIL
hop @1
0/3 · 0.0%
hop @5
0/3 · 0.0%

Chain pass requires every constituent hop at rank 1. Label fallbacks: 0.

deployed RAG chain FAIL
hop @1
0/3 · 0.0%
hop @5
0/3 · 0.0%

Chain pass requires every constituent hop at rank 1. Label fallbacks: 0.

Adjacent benchmark / banked

Relational-hop comparison beyond these two anecdotes

2026-07-30

probe500 / family 0 / bench w5-8e6739 · n=167 · gate off. Scoring unit: exact gold entity/QID at top k, one independently scored relational-hop row per unit. This benchmark scores individual relational-hop examples; it is not a two-hop or three-hop chain-success measurement.

TOMX

verified
accuracy @1
0.4192
accuracy @5
0.5329

Deployment-pinned, banked result. Full probe and result-file hashes are not public, so it is not independently file-hash reproducible here.

matched TX

verified
accuracy @1
0.2156
accuracy @5
0.2994

Matched within 1.75% on resident parameters. This does not establish matched architecture, training bytes, compute, or overall quality.

deployed RAG

blocked
accuracy @1
0.0000
accuracy @5
0.0958

N11 qualification: shown for completeness, not used comparatively. The strongest deployed reader-free RAG’s corrected index/hash basis is not exposed by /meta, and a known LM-1 tokenizer/decoder text-boundary validity issue prevents tying these scores to the corrected artifact.

Limits adjacent to the result / N11 Two curated chains do not measure general compositional reasoning. The constituent rows are run separately and composed by this proof harness; no claim is made that one free-text forward internally executes all hops. RAG is the strongest deployed baseline available here, but its numerical deficits remain N11-qualified and blocked rather than treated as a valid fair-baseline win/loss comparison.
ANSWER · REASON

Shows: one two-hop and one three-hop corpus chain run as explicit constituent reads, an opaque source receipt for each supported TOMX read, chain success scored so that every intermediate top-1 identity must be right, and a dated banked relational-hop row displayed beside its own limits. Does not show: general compositional reasoning, a chain-success benchmark, a single free-text forward internally executing every hop, or valid comparative RAG deficits until the N11 corrected artifact/build basis is public. The hops are composed by this proof harness, and two curated chains are anecdotes.