Update EVAL_RESULTS-passk.md: difficulty wording and caveats
Browse files- EVAL_RESULTS-passk.md +3 -5
EVAL_RESULTS-passk.md
CHANGED
|
@@ -1,22 +1,20 @@
|
|
| 1 |
# EVAL_RESULTS-passk.md — full pass@5 agentic benchmark on all 29 envs (2026-10-29)
|
| 2 |
|
| 3 |
-
Model: Claude Opus 5.5
|
| 4 |
Attempts: 5 per env x 29 envs = **145 runs**. Tools: Bash+Read only (disassembly + run probes; WSL/qemu for non-x86).
|
| 5 |
Grading: OUR verifier oracle — run the challenge binary with the model's ANSWER string as argv[1]; pass = stdout "OK"
|
| 6 |
+ exit 0. Independent of any string compare. Raw logs are retained; summary table below.
|
| 7 |
|
| 8 |
## Headline (honest)
|
| 9 |
- **29/29 envs solved by Opus 5.5 with tools; 145/145 attempts correct** (after 3 infra-retries that returned empty,
|
| 10 |
-
re-run clean). Every model answer was byte-exact equal to our gold (checked separately) —
|
| 11 |
-
confirmation of unique-solvability AND of our oracles.
|
| 12 |
- Solve cost: median 44 s, mean 47 s (range 23-121 s), mean 8.4 tool turns. Cross-arch envs (MIPS/AArch64/RISC-V/
|
| 13 |
ARM32/PPC64LE) were NOT harder for Opus than x86-64 once qemu+objdump were available.
|
| 14 |
- Implication we state plainly to buyers: the public flag-hunt tier is **warm-up level for a frontier agent**
|
| 15 |
(all-pass groups give ~zero GRPO gradient at that capability level). Our answer: (a) use them as the free demo/
|
| 16 |
calibration tier, (b) sell a **private hard tier** targeting mid-band pass rates for the buyer's own model,
|
| 17 |
(c) offer per-model pass-rate calibration (we run it on our harness before you buy).
|
| 18 |
-
- Not claimed: SOTA comparisons, open-weight numbers yet (next step), pass@N
|
| 19 |
-
unseen by any model).
|
| 20 |
|
| 21 |
## Method notes / gotchas found while building this
|
| 22 |
- WSL re-parses argv from Windows (shell metacharacters in answers like `<`, `;`, `&`, backtick mangled) -> grading
|
|
|
|
| 1 |
# EVAL_RESULTS-passk.md — full pass@5 agentic benchmark on all 29 envs (2026-10-29)
|
| 2 |
|
| 3 |
+
Model: Claude Opus 5.5 through a hosted frontier agent, fresh session per attempt.
|
| 4 |
Attempts: 5 per env x 29 envs = **145 runs**. Tools: Bash+Read only (disassembly + run probes; WSL/qemu for non-x86).
|
| 5 |
Grading: OUR verifier oracle — run the challenge binary with the model's ANSWER string as argv[1]; pass = stdout "OK"
|
| 6 |
+ exit 0. Independent of any string compare. Raw logs are retained; summary table below.
|
| 7 |
|
| 8 |
## Headline (honest)
|
| 9 |
- **29/29 envs solved by Opus 5.5 with tools; 145/145 attempts correct** (after 3 infra-retries that returned empty,
|
| 10 |
+
re-run clean). Every model answer was byte-exact equal to our gold (checked separately) — this confirms the shipped binary accepts those answers and verifies our grading chain; it is not proof that each answer is unique. The run was non-isolated, so treat it as an oracle check, not a sealed-valuation result.
|
|
|
|
| 11 |
- Solve cost: median 44 s, mean 47 s (range 23-121 s), mean 8.4 tool turns. Cross-arch envs (MIPS/AArch64/RISC-V/
|
| 12 |
ARM32/PPC64LE) were NOT harder for Opus than x86-64 once qemu+objdump were available.
|
| 13 |
- Implication we state plainly to buyers: the public flag-hunt tier is **warm-up level for a frontier agent**
|
| 14 |
(all-pass groups give ~zero GRPO gradient at that capability level). Our answer: (a) use them as the free demo/
|
| 15 |
calibration tier, (b) sell a **private hard tier** targeting mid-band pass rates for the buyer's own model,
|
| 16 |
(c) offer per-model pass-rate calibration (we run it on our harness before you buy).
|
| 17 |
+
- Not claimed: SOTA comparisons, open-weight numbers yet (next step), or a full pass@N calibration of the private tier. Hard-tier evidence so far is limited; see `EVAL_RESULTS-frontier.md` §C.
|
|
|
|
| 18 |
|
| 19 |
## Method notes / gotchas found while building this
|
| 20 |
- WSL re-parses argv from Windows (shell metacharacters in answers like `<`, `;`, `&`, backtick mangled) -> grading
|