Pinky RRRR: honest steer attribution — the evidence chain names the source (v1.101.0, #60) #136

Merged

ClaudiuSchuster merged 1 commit from oss-oo/pinky-rrrr-attrib into main

2026-10-17 03:42:11 +00:00

ClaudiuSchuster commented

2026-10-17 03:41:36 +00:00

Owner

R1 — honest sender attribution (owner finding #60)

A steer-delivered transcript message rode the real_user user_prompt shape, so the sender rendered as generic user — the owner watched B³’s words land in gen-4 as “user ⧉”.

The cure is an evidence chain, never a guess:

the message’s steer anchor pins the delivery (its body is the VERBATIM posted bytes);
the anchor’s send stamp + normalized byte equality prove WHICH steer prompt file carried the words (the obs- child seam the lane rows know);
the SOURCE is the session whose own tool part ran the posting call — it names the TARGET sid AND the verbatim text, inside the send-time window (the console POST always fires at send). Latest hit is the poster; a photo-finish between two sessions is UNPROVEN → stays user.

Machine steers (context guardian, quota throttle, starvation watch, turn-retry) carry their identity BY CONSTRUCTION (start_steer records at spawn).
user stays exclusively genuine owner input: legacy rows, cleaned prompt files, dead source sids all settle honestly as user.
Symmetry: agent replies keep their agent class untouched — who asked (source session) and who answered (agent) both render honestly.
The drawer renders the source title in the sender slot (mint kept — still words INTO the session); tooltip carries the provenance + sid. Findings persist in a capped album under STEER_DIR; fresh unproven messages rescan (parts flush lag), older ones settle.

R2 — the QQQQ-F3 docs fold

The specimen recorded 13 legs while its docs said 11 (Q2/Q7 double-recorded). The two double-recorded proofs got their own leg ids — docstring, in-code labels and Makefile now say 13 honestly (chosen over docs-only: every record() is its own proof point).

Evidence

Suite: 1733 ran (+20 new: tests/test_steer_attribution.py — chain/album/honesty/symmetry; tests/test_round_rrrr.py — wiring tripwires + specimen pin), 5 skipped, only the expected pre-tag version tie.
Vault: tests/rigs/rrrr_attrib_cdp.py — 12/12 legs (API truth + rendered drawer: source named, unproven stays user, machine identity, symmetry, tooltip evidence, computed style, clean chain).
Super-Linter slim-v8.7.0 station recipe: BASH, BASH_EXEC, MARKDOWN, PYTHON_PYLINT, PYTHON_RUFF, YAML — all green.

## R1 — honest sender attribution (owner finding #60) A steer-delivered transcript message rode the `real_user` user_prompt shape, so the sender rendered as generic `user` — the owner watched B³’s words land in gen-4 as “user ⧉”. **The cure is an evidence chain, never a guess:** 1. the message’s steer anchor pins the delivery (its body is the VERBATIM posted bytes); 2. the anchor’s send stamp + normalized byte equality prove WHICH steer prompt file carried the words (the `obs-` child seam the lane rows know); 3. the SOURCE is the session whose own tool part ran the posting call — it names the TARGET sid AND the verbatim text, inside the send-time window (the console POST always fires at send). Latest hit is the poster; a photo-finish between two sessions is UNPROVEN → stays `user`. - Machine steers (context guardian, quota throttle, starvation watch, turn-retry) carry their identity BY CONSTRUCTION (`start_steer` records at spawn). - `user` stays exclusively genuine owner input: legacy rows, cleaned prompt files, dead source sids all settle honestly as `user`. - **Symmetry**: agent replies keep their `agent` class untouched — who asked (source session) and who answered (agent) both render honestly. - The drawer renders the source title in the sender slot (mint kept — still words INTO the session); tooltip carries the provenance + sid. Findings persist in a capped album under STEER_DIR; fresh unproven messages rescan (parts flush lag), older ones settle. ## R2 — the QQQQ-F3 docs fold The specimen recorded 13 legs while its docs said 11 (Q2/Q7 double-recorded). The two double-recorded proofs got their own leg ids — docstring, in-code labels and Makefile now say 13 honestly (chosen over docs-only: every `record()` is its own proof point). ## Evidence - Suite: 1733 ran (+20 new: `tests/test_steer_attribution.py` — chain/album/honesty/symmetry; `tests/test_round_rrrr.py` — wiring tripwires + specimen pin), 5 skipped, only the expected pre-tag version tie. - Vault: `tests/rigs/rrrr_attrib_cdp.py` — 12/12 legs (API truth + rendered drawer: source named, unproven stays `user`, machine identity, symmetry, tooltip evidence, computed style, clean chain). - Super-Linter slim-v8.7.0 station recipe: BASH, BASH_EXEC, MARKDOWN, PYTHON_PYLINT, PYTHON_RUFF, YAML — all green.

ClaudiuSchuster added 1 commit

2026-10-17 03:41:36 +00:00

Pinky RRRR: honest steer attribution — the evidence chain names the source (v1.101.0, #60 )

Checks / Functional checks (pull_request) Successful in 19s

Details Explain this check

8f57d0b151

R1 (finding #60): a steer-delivered transcript message rode the
real_user user_prompt shape, so the sender rendered as generic
'user' (B³'s words landed in gen-4 as 'user ⧉'). The cure is an
evidence chain, never a guess: the message's steer anchor pins the
delivery; the anchor's send stamp + normalized byte equality prove
WHICH steer prompt file carried the words (the obs- child seam the
lane rows know); the SOURCE is the session whose own tool part ran
the posting call (target sid + verbatim text in the send-time
window; a photo-finish between two sessions is unproven and stays
'user'). Machine steers (guardian/throttle/starvation/turn-retry)
carry their identity by construction, recorded at spawn.
'user' stays exclusively genuine owner input; unproven rows
(legacy, cleaned files, dead sids) settle honestly. The drawer
renders the source title in the sender slot (mint kept — words INTO
the session), tooltip carries the provenance; agent replies keep
their 'agent' class (both directions honest). Album persists
findings under STEER_DIR; fresh unproven messages rescan (parts
flush lag), older ones settle.

R2 (QQQQ-F3): the specimen recorded 13 legs while its docs said 11
— the double-recorded Q2/Q7 proofs got their own ids, docstring +
Makefile now say 13 honestly.

Suite: +20 legs (tests/test_steer_attribution.py — the chain,
album, honesty, symmetry; tests/test_round_rrrr.py — wiring
tripwires + the vault specimen pin); vault: rrrr_attrib_cdp.py
(12 legs, API + rendered drawer), registered + make target.

ClaudiuSchuster merged commit c74a0b55de into main

2026-10-17 03:42:11 +00:00

ClaudiuSchuster deleted branch oss-oo/pinky-rrrr-attrib

2026-10-17 03:42:11 +00:00

ClaudiuSchuster referenced this pull request from a commit

2026-10-17 03:42:12 +00:00

Pinky RRRR: honest steer attribution — the evidence chain names the source (v1.101.0, #60) (#136)