Skip to content

Commit 4a433b4

Browse files
echobtechobt
committed
docs(deepagent): m29d HF full re-upload N=9 report + card
Full folder re-upload of prod_hard_deepswe_med N=9 to BaseIntelligence/deepagent@test; card pretty_name DeepAgent with M28 diversity, M27 floors, M16 N=10 supersession.
1 parent 3b260af commit 4a433b4

4 files changed

Lines changed: 183 additions & 2 deletions

File tree

‎deepagent/datasets/prod_hard_deepswe_med/README.md‎

Lines changed: 6 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,5 @@
11
---
2-
pretty_name: DeepAgent Hardness — DeepSWE-median diversified (M28)
2+
pretty_name: DeepAgent
33
tags:
44
- code
55
- software-engineering
@@ -22,6 +22,8 @@ size_categories:
2222

2323
This revision is the **current product hardness corpus**. Structural difficulty stays on the **DeepSWE median** floors (M27). **M28** adds coverage volume under a **repo diversity cap** so one upstream no longer dominates the set.
2424

25+
> **Not M16 N=10.** Historical `datasets/test_n10` / early M16 `test` trees are **not** the live product. Current certified hardness is **N=9** under M27 floors + M28 diversity.
26+
2527
## M28 coverage + diversity policy
2628

2729
| Policy | Value |
@@ -41,8 +43,10 @@ Preferred targets N≥12/15 remain aspirational; shortfall is honest (many floor
4143

4244
| Item | Status |
4345
|------|--------|
44-
| **Current product (this revision)** | `datasets/prod_hard_deepswe_med` · M27 floors + **M28 diversity (max 2/repo)** · N=9 |
46+
| **Current product (this revision)** | `datasets/prod_hard_deepswe_med` · M27 floors + **M28 diversity (max 2/repo)** · **N=9** |
47+
| **M16 N=10 / `test_n10` claims** | **Superseded** — early M16 live wave and any card/docs that called N=10 the current product are historical only |
4548
| Prior M27 HF `test` (N=5, denser werkzeug) | Superseded by this diversified tree (drops weak 3rd-werkzeug keep; adds non-werkzeug dual-truth keeps) |
49+
| Prior M28c HF `test` (same N=9 ids) | Re-asserted by M29d full-folder re-upload with refreshed card (`pretty_name: DeepAgent`) |
4650
| Historical softer band | `datasets/prod_hard_keep` (M25/M26) retained **locally for audit only** — not HF `test` |
4751

4852
Do **not** treat model dual-solve alone as a product drop (M25 intrinsic policy). Hardness refusals remain dual-truth fail, prompt–verifier misalignment, structural floors, and high-confidence intrinsic `EASY_REQUEST`.

‎deepagent/datasets/prod_hard_deepswe_med/hf_upload_m29d_report.json‎

Lines changed: 93 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,93 @@
1+
{
2+
"feature": "m29d-hf-full-reupload-n9-card",
3+
"fulfills": [
4+
"VAL-PUB-005"
5+
],
6+
"generated_at_utc": "2026-07-20T12:09:14Z",
7+
"ok": true,
8+
"pushed": true,
9+
"schema_ok": true,
10+
"repo_id": "BaseIntelligence/deepagent",
11+
"revision": "test",
12+
"did_not_push_main": true,
13+
"full_folder_upload": true,
14+
"not_card_only": true,
15+
"local_src": "datasets/prod_hard_deepswe_med",
16+
"upload_stage_src": "datasets/_m29d_hf_stage",
17+
"stage_note": "Mirrored product packs + corpus meta; excluded generate_*.log, _m28* work dirs, .env; restored committed tests/Dockerfile tags",
18+
"band": "deepswe_median_m27",
19+
"pack_count": 9,
20+
"task_ids": [
21+
"realpr-click-3442",
22+
"realpr-itemadapter-101",
23+
"realpr-oauthlib-889",
24+
"realpr-packaging-1120",
25+
"realpr-packaging-1267",
26+
"realpr-rich-3930",
27+
"realpr-werkzeug-2637",
28+
"realpr-werkzeug-3116",
29+
"realpr-wtforms-923"
30+
],
31+
"diversity": {
32+
"N": 9,
33+
"unique_repos": 7,
34+
"max_packs_per_repo": 2,
35+
"packs_per_repo": {
36+
"scrapy/itemadapter": 1,
37+
"pypa/packaging": 2,
38+
"Textualize/rich": 1,
39+
"pallets-eco/wtforms": 1,
40+
"pallets/werkzeug": 2,
41+
"oauthlib/oauthlib": 1,
42+
"pallets/click": 1
43+
},
44+
"ok_n_ge_8": true,
45+
"ok_unique_repos_ge_5": true,
46+
"ok_max_packs_le_2": true
47+
},
48+
"upload_commit_sha": "c17bc27918d2e4ac1196ac3af049f1c7b58a161d",
49+
"upload_commit_url": "https://huggingface.co/datasets/BaseIntelligence/deepagent/commit/c17bc27918d2e4ac1196ac3af049f1c7b58a161d",
50+
"upload_message": "upload_folder complete",
51+
"card": {
52+
"readme": "datasets/prod_hard_deepswe_med/README.md",
53+
"pretty_name": "DeepAgent",
54+
"product_readme": "datasets/prod_hard_deepswe_med/PRODUCT_README.md",
55+
"notes": [
56+
"pretty_name: DeepAgent",
57+
"Certified N=9",
58+
"M28 diversity cap max 2 packs/repo",
59+
"M27 DeepSWE-median floors table",
60+
"Explicit supersession of M16 N=10 / test_n10 claims"
61+
]
62+
},
63+
"pull_back": {
64+
"out": "/tmp/da_m29d_hf_pull",
65+
"ok": true,
66+
"pack_count": 9,
67+
"task_ids_match_local": true,
68+
"exact_set_match": true,
69+
"sample_pack": "realpr-itemadapter-101",
70+
"sample_files_present": [
71+
"task.toml",
72+
"instruction.md",
73+
"environment/Dockerfile",
74+
"tests",
75+
"solution"
76+
],
77+
"coverage_stats_N": 9,
78+
"coverage_stats_unique_repos": 7,
79+
"coverage_stats_max_packs_per_repo": 2,
80+
"card_pretty_name_DeepAgent": true,
81+
"card_N9": true,
82+
"card_m28_max2": true,
83+
"card_m27_floors": true,
84+
"card_supersedes_m16_n10": true
85+
},
86+
"secrets": {
87+
"local_pre_upload_scan_ok": true,
88+
"pull_tree_scan_ok": true,
89+
"cli_json_token_free": true,
90+
"hit_count": 0
91+
},
92+
"remote_list_match": true
93+
}

‎deepagent/datasets/prod_hard_deepswe_med/hf_upload_m29d_report.md‎

Lines changed: 72 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,72 @@
1+
# m29d — HF full re-upload N=9 + refreshed card
2+
3+
**Feature:** `m29d-hf-full-reupload-n9-card`
4+
**Fulfills:** VAL-PUB-005
5+
**UTC:** 2026-07-20
6+
7+
## What shipped
8+
9+
| Field | Value |
10+
|---|---|
11+
| Local src | `datasets/prod_hard_deepswe_med` (staged via `datasets/_m29d_hf_stage`) |
12+
| Stage note | Full folder pack tree + corpus meta; excluded `generate*.log`, `_m28*` work dirs, `.env`; restored committed `tests/Dockerfile` image tags |
13+
| Repo | `BaseIntelligence/deepagent` |
14+
| Revision | **`test` only** (not `main`; not card-only) |
15+
| Pack count | **9** |
16+
| Band | `deepswe_median_m27` + M28 diversity (max 2/repo) |
17+
| unique_repos | **7** |
18+
| max packs/repo | **2** |
19+
| Upload SHA | `c17bc27918d2e4ac1196ac3af049f1c7b58a161d` |
20+
| Commit URL | https://huggingface.co/datasets/BaseIntelligence/deepagent/commit/c17bc27918d2e4ac1196ac3af049f1c7b58a161d |
21+
| Message | `upload_folder complete` |
22+
23+
## Task ids (local == remote list == pull)
24+
25+
1. `realpr-click-3442`
26+
2. `realpr-itemadapter-101`
27+
3. `realpr-oauthlib-889`
28+
4. `realpr-packaging-1120`
29+
5. `realpr-packaging-1267`
30+
6. `realpr-rich-3930`
31+
7. `realpr-werkzeug-2637`
32+
8. `realpr-werkzeug-3116`
33+
9. `realpr-wtforms-923`
34+
35+
- Local ↔ pull `tasks/*` set match: **yes** (exact)
36+
- Remote/pull `pack_manifest`: count=9, task_ids exact match
37+
- Pull out: `/tmp/da_m29d_hf_pull` pack_count=9
38+
- Sample pack layout (itemadapter-101): task.toml, instruction.md, environment/, tests/, solution/ present
39+
- Meta present on pull: coverage_stats.json, median_stats.json, PRODUCT_README.md, PROVENANCE.md, README.md
40+
41+
## Card / supersession
42+
43+
- Dataset card `README.md` YAML front-matter: `pretty_name: DeepAgent`
44+
- Documents **N=9**, M28 diversity **max 2 packs/repo**, unique_repos=7
45+
- M27 floors table (files≥4 OR hybrid; hunks≥14; added≥400; F2P≥5; dual-truth)
46+
- Explicit supersession of **M16 N=10 / `test_n10`** claims (historical only; not current product)
47+
- Re-asserts M28c N=9 diversified product after public-surface card refresh
48+
49+
## Secrets
50+
51+
- Pre-upload local text scan on stage: **0 hits**
52+
- Pull tree scan: **0 hits**
53+
- Upload/pull CLI JSON: no token fields (auth via gitignored `.env` only)
54+
- Stage excludes: generate logs, `_m28*` work dirs, `.env`
55+
56+
## Commands
57+
58+
```bash
59+
cd /projects/Agent-SWE/deepagent
60+
set -a && source .env && set +a
61+
62+
.venv/bin/deepagent upload --src datasets/_m29d_hf_stage \
63+
--repo-id BaseIntelligence/deepagent --revision test --dry-run --json
64+
65+
.venv/bin/deepagent upload --src datasets/_m29d_hf_stage \
66+
--repo-id BaseIntelligence/deepagent --revision test --json
67+
68+
.venv/bin/deepagent pull --repo-id BaseIntelligence/deepagent \
69+
--revision test --out /tmp/da_m29d_hf_pull --json
70+
```
71+
72+
Services.yaml: `upload_prod_hard_deepswe_med_m29`, `pull_test_m29`.

‎deepagent/datasets/prod_hard_deepswe_med/secrets_scan_m29d.json‎

Lines changed: 12 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,12 @@
1+
{
2+
"generated_at_utc": "2026-07-20T12:07:17Z",
3+
"feature": "m29d-hf-full-reupload-n9-card",
4+
"stage": "datasets/_m29d_hf_stage",
5+
"local_pre_upload_hit_count": 0,
6+
"leak_scan_secretish_count": 1,
7+
"hits": [],
8+
"leak_scan_sample": [
9+
"leak_scan_unavailable:TypeError:'LeakScanResult' object is not iterable"
10+
],
11+
"ok": true
12+
}

0 commit comments

Comments
 (0)