Burnmydays Claude Opus 4.8 (1M context) commited on
Commit
45ca42d
·
1 Parent(s): 9eeaeb4

docs: refresh Codex handoff + add STATUS snapshot

Browse files

- CODEX.md: rewritten as a current, self-contained handoff (correct desktop
path, current state, integrity rules). Headline Codex task restored: the
turn-delta Codex-token parser (per-day cache_create estimation) — left for
Codex; the ratio-based Alpha/Beta fix + marker were done this session.
- STATUS.md: one-page owner snapshot (built / left / badges / verify / docs map).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Files changed (2) hide show
  1. CODEX.md +54 -45
  2. STATUS.md +44 -0
CODEX.md CHANGED
@@ -1,50 +1,59 @@
1
- # Using Codex on this repo
2
-
3
- You (Luthen) may have OpenAI Codex access. This file is both your instructions
4
- for driving Codex against this repo AND the thing that earns the **Codex $10k
5
- sponsor prize** (which requires Codex-attributed commits in the repo/Space).
6
-
7
- ## Why Codex matters for the prize
8
- The sponsor track rewards repos where Codex did real work, shown through
9
- Codex-attributed commits. So: let Codex make actual edits and commit them with
10
- its own attribution. Don't hand-copy its output let it write to disk and commit.
11
-
12
- ## Setup
13
- 1. Install Codex CLI (if not already): follow OpenAI's current install docs.
14
- 2. From the repo root: `cd ~/moses-sigrank`
15
- 3. Point Codex at this folder so it has full file context.
16
-
17
- ## Good first tasks to hand Codex (each = one attributed commit)
18
- These are real, useful, and low-risk — perfect for generating genuine commits:
19
-
20
- 1. **Refine the 2:1 anchor** (`ingest.py` `parse_codex`)
21
- Current: `est_fresh_input = 2 * output`. Ask Codex to implement the
22
- turn-delta method as an optional, more-accurate path (estimate cache_create
23
- from per-turn input deltas when daily/session granularity is present), keeping
24
- the 2:1 floor as fallback. Keep the caveat flags.
25
-
26
- 2. **Add unit tests** (`test_metrics.py`)
27
- Ask Codex to write pytest cases that lock the canonical numbers:
28
- MO§ES Υ=18,437, leverage=2042, 10x DEV=3.31, telescoping identity
29
- (o/i)(cw/o)(cr/cw) == cr/i for every SEED operator.
30
-
31
- 3. **Self-cost for Codex rows**
32
- ccusage supplies cost for Claude. For Codex JSON, wire OpenAI per-1M prices
33
- into `parse_codex` meta so Codex rows show real (not list-estimate) $/1M.
34
-
35
- 4. **Board polish** a small visual marker (asterisk / dimmed row) on estimated
36
- rows so the leaderboard itself shows measured vs estimated at a glance.
37
-
38
- ## Verifying Codex's work before commit
39
- After Codex edits, ALWAYS run the stress test (see TODO.md "verify" block):
 
 
 
40
  ```
41
- cd ~/moses-sigrank && python3 -c "import py_compile,glob; [py_compile.compile(f,doraise=True) for f in glob.glob('*.py')]"
42
- python3 metrics.py # canonical numbers must still print
 
43
  ```
44
- Then commit with Codex attribution.
45
 
46
  ## DO NOT let Codex touch
47
- - The canonical SEED numbers in `metrics.py` (frozen, verified this session).
48
  - The Υ formula `(C·O)/I²` or the telescoping identity — these are the thesis.
49
- - Anything that turns the 2:1 anchor into a silent strict assumption (it must
50
- stay flagged/provisional).
 
 
 
 
 
 
1
+ # Codex handoff MO§ES SigRank
2
+
3
+ Read this first. It's the instruction set for driving **OpenAI Codex** against this
4
+ repo and the thing that earns the **Codex $10k sponsor prize**.
5
+
6
+ Local path (desktop storage): `/Users/dericmchenry/Desktop/moses-sigrank`
7
+ GitHub (current): `github.com/Burnmydays/hf-` → upload target: `github.com/SunrisesIllNeverSee`
8
+
9
+ ## How the prize works (important)
10
+ The sponsor track rewards repos where **Codex did real work**, shown through
11
+ **Codex-attributed commits**. It does NOT require that 100% of the repo is Codex's.
12
+ Two rules:
13
+ 1. **Genuine work only.** Let Codex actually write the code and commit it with its own
14
+ attribution. Do not hand-copy output or fake attribution — that risks disqualification.
15
+ 2. **Verify the official rule** on the hackathon page before relying on this.
16
+
17
+ ## Current state (as of this handoff)
18
+ Done and pushed to Burnmydays/hf- (main):
19
+ - Codex parser fixed — `_codex_input_estimate` in `ingest.py` (Beta = output × real Claude
20
+ io_ratio; Alpha = output × 2.0 AA baseline). Two pathways, both flagged `*` estimated.
21
+ - `./sigrank --all` (run every provider in turn).
22
+ - Instructions sharpened (app.py "Clock Your Signal" tab + README).
23
+ - Wild corpus = 10 tokscale.ai operators; board = 11 rows; Supabase migrated + synced.
24
+
25
+ ## Tasks for Codex (each = one attributed commit)
26
+ 1. **Turn-delta Codex parser** — the headline Codex-token-parsing work. The current
27
+ estimate (`_codex_input_estimate` in `ingest.py`) is a single ratio applied to total
28
+ output. Make it more accurate: when the Codex JSON has **daily/session granularity**,
29
+ estimate `cache_create` from **per-day context growth** (each day's input above the
30
+ prior day's floor ≈ new cache writes) instead of one flat ratio. Keep the ratio path
31
+ (Alpha 2:1 / Beta Claude-ratio) as the fallback when only totals are present, and keep
32
+ the `*` estimated flag + caveat naming the method used. This is real parsing work on
33
+ Codex's own token data the natural Codex-attributed commit.
34
+ 2. **`test_metrics.py`** — pytest that locks the canonical numbers and identities:
35
+ - MO§ES Υ = 18,436.98, leverage = 2042.2, 10x DEV = 3.31.
36
+ - Telescoping identity `(o/i)·(cw/o)·(cr/cw) == cr/i` for every SEED operator with cache.
37
+ - Both Codex pathways: Alpha `output×2`, Beta `output×io_ratio`.
38
+ 3. **Real Codex `$/1M`** wire OpenAI per-1M prices into `parse_codex_submission` meta so
39
+ Codex rows can show real (not list-estimate) cost. Keep the `*` estimated flag on input.
40
+ 4. **Upload the repo to `github.com/SunrisesIllNeverSee`** (the prize-submission location).
41
+
42
+ ## Verify BEFORE every Codex commit
43
  ```
44
+ cd /Users/dericmchenry/Desktop/moses-sigrank
45
+ .venv/bin/python -c "import py_compile,glob; [py_compile.compile(f,doraise=True) for f in glob.glob('*.py')]"
46
+ .venv/bin/python metrics.py # MO§ES must print Y 18436.98, lev 2042.2, 10xDEV 3.31, $/1M 0.527
47
  ```
48
+ (use `python3` if there's no `.venv`.)
49
 
50
  ## DO NOT let Codex touch
51
+ - The MO§ES (ccusage) SEED row in `metrics.py` (canonical, Υ 18436.98).
52
  - The Υ formula `(C·O)/I²` or the telescoping identity — these are the thesis.
53
+ - The Codex estimation must stay **flagged (`*`)** — never a silent strict assumption.
54
+
55
+ ## Already done — NOT available as Codex commits
56
+ - 2:1 → real-ratio anchor refinement (Alpha/Beta unified). Done by Claude this session.
57
+ - Board `~`/`*` estimated-row marker. Done by Claude this session.
58
+
59
+ See `SCRATCHPAD.md` for live cross-agent state and `TODO.md` for the full task board.
STATUS.md ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # STATUS — MO§ES SigRank (where things stand)
2
+
3
+ Snapshot for the owner. Deadline: **2026-06-15 23:59 UTC**.
4
+ Repo: `github.com/Burnmydays/hf-` (main `9eeaeb4`). · Upload target: `SunrisesIllNeverSee`.
5
+
6
+ ---
7
+
8
+ ## ✅ Built, verified, and pushed
9
+ - **Core engine** — `metrics.py`: 4 integers → full ledger. Canonical MO§ES Υ **18,436.98**.
10
+ - **Leaderboard** — 11 rows live (MO§ES + 10 tokscale.ai operators), log-scaled Υ, $/1M column.
11
+ - **Codex parser (fixed)** — `_codex_input_estimate`: Beta = output × your real Claude ratio;
12
+ Alpha = output × 2.0 (AA baseline). Both flagged `*`. No more hardcoded `/9.0`.
13
+ - **Local importer** — `./sigrank` (Claude), `./sigrank --codex` (Codex), `./sigrank --all`.
14
+ - **Instructions** — "Clock Your Signal" tab + README: measure each provider separately.
15
+ - **Persistence** — Supabase migrated (`submitted_at`, `hf_user`, `sigrank_sessions` + RLS);
16
+ board synced to 11 rows; Greatest Hits read path verified end-to-end.
17
+ - **MiniCPM narration** — non-blocking, template fallback, `@GPU` for ZeroGPU.
18
+
19
+ ## ⏳ Left to do
20
+ 1. **Deploy to the HF Space** (parked on your call)
21
+ - Confirm how code reaches the Space (HF git remote vs GitHub auto-sync vs manual).
22
+ - Set Space secrets from `SECRETS.local.md`: `SUPABASE_URL` + `SUPABASE_ANON_KEY`
23
+ (add `SUPABASE_SERVICE_KEY` only if you want signed-in visitor rows to persist).
24
+ 2. **Codex handoff** → upload to `SunrisesIllNeverSee` + the remaining Codex-attributed
25
+ commits (`test_metrics.py`, real Codex `$/1M`). See `CODEX.md`.
26
+ 3. **Submission** — move Space into `build-small-hackathon` org · 60s video · social post ·
27
+ GitHub link in README.
28
+
29
+ ## 🏅 Badges
30
+ ✅ Off Brand · ✅ Tiny Titan · ✅ Best MiniCPM | ⏳ Best Demo (needs video) · ⏳ Codex $10k (needs Codex commits)
31
+
32
+ ## 🔎 Verify anytime
33
+ ```
34
+ cd /Users/dericmchenry/Desktop/moses-sigrank
35
+ .venv/bin/python metrics.py # canonical numbers
36
+ ./sigrank --all --no-color # your live Claude + Codex read
37
+ ```
38
+
39
+ ## 🗂 Where things are documented
40
+ - `CODEX.md` — Codex handoff instructions (grab from desktop).
41
+ - `TODO.md` — full task board (done at bottom).
42
+ - `SCRATCHPAD.md` — live cross-agent state.
43
+ - `SUPABASE_MIGRATION.md` — the DB migration (already applied).
44
+ - `SECRETS.local.md` — Supabase keys (gitignored, never uploaded).