jonny commited on
Commit
b6b9678
Β·
verified Β·
1 Parent(s): 1b2a2fb

card: five blocks, not four; label the ladder

Browse files
Files changed (1) hide show
  1. README.md +5 -2
README.md CHANGED
@@ -49,6 +49,9 @@ set can change; `main` is not a stable contract. Topos pins the SHA it was gated
49
  snapshot_download("Dialogues/horos", revision="1873a1b667dc2d1bd630835f1f4f47025f3864d9")
50
  ```
51
 
 
 
 
52
  ```
53
  labels β‰₯ tau_high, none low, nothing in band β†’ ACT on the scope set (1..N scopes)
54
  high none, nothing else β†’ abstain: no personal data
@@ -58,7 +61,7 @@ nothing anywhere, none included β†’ escalate to the pack's LLM (ign
58
 
59
  ## Report card
60
 
61
- Every release publishes all four blocks below, in full, **including the numbers that got
62
  worse** β€” this is a standing contract, not a summary of a good run. Every row is versioned
63
  and every cross-arm claim carries an interval. Each block earns its slot by having caught
64
  something the others hid:
@@ -79,7 +82,7 @@ something the others hid:
79
  `swallowed` and `escalated` stay separate throughout: they have opposite fixes, and
80
  collapsing them into "not answered" hides which one you have.
81
 
82
- Blocks 1–3 regenerate from one command in `topos-eval`
83
  (`scripts/scope_head_report_card.py --head A --label v1 --head B --label v2`). That
84
  repo is private, so the numbers here are auditable by us and take-it-or-leave-it for
85
  you β€” the benchmark itself is what we'd have to open to change that.
 
49
  snapshot_download("Dialogues/horos", revision="1873a1b667dc2d1bd630835f1f4f47025f3864d9")
50
  ```
51
 
52
+ **The ladder.** Four branches, because "no scopes" is two different states β€” the model
53
+ deciding there is no personal data here, and the model having no idea:
54
+
55
  ```
56
  labels β‰₯ tau_high, none low, nothing in band β†’ ACT on the scope set (1..N scopes)
57
  high none, nothing else β†’ abstain: no personal data
 
61
 
62
  ## Report card
63
 
64
+ Every release publishes all five blocks below, in full, **including the numbers that got
65
  worse** β€” this is a standing contract, not a summary of a good run. Every row is versioned
66
  and every cross-arm claim carries an interval. Each block earns its slot by having caught
67
  something the others hid:
 
82
  `swallowed` and `escalated` stay separate throughout: they have opposite fixes, and
83
  collapsing them into "not answered" hides which one you have.
84
 
85
+ Blocks 1–3 and 5 regenerate from one command in `topos-eval`
86
  (`scripts/scope_head_report_card.py --head A --label v1 --head B --label v2`). That
87
  repo is private, so the numbers here are auditable by us and take-it-or-leave-it for
88
  you β€” the benchmark itself is what we'd have to open to change that.