Edition README
SRIC: external semantic admission
README.md
1 # SRIC Block 24: External Latent-Reasoner Causal-Semantic Admission
2
3 **Release:** v0.3.2
4 **Frozen scientific protocol:** v0.3.0
5 **Canonical status:** `INFRASTRUCTURE_VALIDATED_EXTERNAL_EXECUTION_PENDING`
6 **Author:** Komorebi P.
7
8 This package is the first source-locked SRIC implementation intended to audit a third-party latent-reasoning checkpoint. It targets the public multimode GPT-2 Coconut checkpoint trained on PrOntoQA and turns the synthetic Block 16-23 methodology into a fail-closed external-model protocol.
9
10 The package does **not** contain the external repository, checkpoint, GPT-2 weights, or public dataset. It does **not** report a result for the real checkpoint. The actual checkpoint has not been executed in this build environment.
11
12 ## What the package asks
13
14 The primary causal audit asks whether the model's recurrent latent trace is:
15
16 1. replayable through a source-faithful adapter;
17 2. useful for the protected answer;
18 3. temporally ordered;
19 4. structurally meaningful beyond vector norm or covariance;
20 5. stable under semantic-preserving surface changes;
21 6. specific to the current proof instance beyond final-answer identity.
22
23 Only a model passing the causal admission gates may enter semantic escalation. The semantic compiler then searches for a low-rank quotient that predicts later latent states, answer-residual future state, anonymous intervention effects, and the final answer while remaining stable across semantic twins. Gold proof steps and explicit-CoT mode are withheld until after the quotient is sealed.
24
25 ## Primary claim tier
26
27 The PrOntoQA primary tier is explicitly labeled **content-specific**, not algorithmically procedure-specific. The public test family is dominated by fixed-form unary chains. A same-answer, different-proof transplant can show that traces are tied to task or proof-instance content, but it cannot by itself establish that the model represents different reasoning algorithms.
28
29 A stronger structure-sensitive follow-up is specified for ProsQA in `PROSQA_FOLLOWUP_PROTOCOL.md`.
30
31 ## Sealed chronology
32
33 ```text
34 verify source commit and Git blobs
35 -> hash checkpoint and local GPT-2 snapshot
36 -> choose deterministic task population
37 -> reserve group-disjoint covariance calibration support
38 -> freeze causal audit support and donor plans
39 -> freeze semantic discovery/calibration/final partitions
40 -> write pre-audit seal
41 -> estimate covariance on calibration support only
42 -> run causal audit on disjoint audit support
43 -> if CONTENT_SPECIFIC, collect semantic discovery support
44 -> collect semantic calibration support
45 -> select and seal quotient
46 -> only then collect semantic final twins/counterfactuals
47 -> reveal proof-instance and explicit-CoT witnesses
48 -> run causal subspace patching
49 -> write result receipt
50 ```
51
52 ## External-runtime and terminal-summary hardening in v0.3.2
53
54 Version 0.3.2 is an infrastructure-only patch. It does not alter the v0.3.0 scientific gates, task profiles, donor rules, semantic chronology, or claim boundary. It adds:
55
56 - automatic discovery of an installed Python 3.10, 3.11, or 3.12 runtime;
57 - a revision-pinned asset downloader for the exact `checkpoint_37` file;
58 - a minimal allowlist for the GPT-2 base-model snapshot instead of downloading every framework variant;
59 - automatic retry through the standard HTTP/LFS path if the local Hugging Face Xet transfer fails;
60 - `-DisableXet` and `-LocalFilesOnly` setup modes;
61 - an asset receipt recording complete checkpoint SHA-256, size, minimal base-model tree SHA-256, revisions, and transfer mode;
62 - tests that reject unpinned or overbroad asset plans.
63
64 The external checkpoint remains unexecuted in the release environment. Asset-transfer failure is classified as infrastructure failure and cannot be interpreted as evidence about the model.
65
66 ## Scientific safeguards frozen in v0.3.0
67
68 - Exact external repository commit and required Git blobs are verified.
69 - Hugging Face checkpoint and base-model revisions are pinned by setup commands.
70 - The complete checkpoint file and local base-model tree are hashed into the seal.
71 - Covariance-matched noise is calibrated on a group-disjoint support, not the causal audit rows.
72 - The semantic final support is not captured before quotient selection and sealing.
73 - Gold explicit-CoT is reserved for post-seal comparison and cannot aid quotient construction.
74 - PrOntoQA's maximum specificity claim is narrowed to content/proof-instance specificity.
75 - A separate source-faithful recurrence parity test compares the adapter engine against the reviewed upstream recurrence.
76 - The engineered fixture is staged before execution and is always marked as a non-real positive control.
77
78 ## Status taxonomy
79
80 | Status | Meaning |
81 |---|---|
82 | `ADAPTER_INVALID` | Exact replay failed; later causal claims are inadmissible. |
83 | `LATENT_UNUSED` | Removing latent recurrence did not produce the frozen minimum effect. |
84 | `LATENT_UNORDERED` | Latent computation may matter, but the declared step order was not supported. |
85 | `LATENT_MAGNITUDE_ONLY` | Matched corruptions did not show enough dependence on latent structure. |
86 | `SURFACE_UNSTABLE` | A semantic-preserving twin changed the trace's functional effect too strongly. |
87 | `LATENT_GENERIC` | The trace was useful and ordered but did not show enough same-answer task specificity. |
88 | `CONTENT_SPECIFIC` | The trace was tied to proof-instance content beyond final-answer identity. |
89 | `PROCEDURE_SPECIFIC` | Reserved for datasets with genuinely varied solution structures and a structure-near versus structure-far contrast. |
90 | `ONTOLOGY_UNDERIDENTIFIED` | No low-rank semantic candidate passed the construction gates. |
91 | `SEMANTIC_CANDIDATE` | A quotient passed construction but not the post-seal proof-instance witness. |
92 | `SEMANTICALLY_IDENTIFIED` | Post-seal witness passed, but causal patching did not. |
93 | `CAUSALLY_VALIDATED` | The sealed quotient passed both post-seal witness and causal patching gates. |
94
95 ## Engineered validation result
96
97 The included fixture is deliberately constructed to contain a causal semantic packet. It is a mechanical positive control, not evidence about Coconut.
98
99 - Frozen fixture seal: `68b04ea973f5030a7085d2b1891e87bedbf058db007d004fefee047beaee70b9`
100 - Fixture result receipt: `924157159d14b8d3715f2f886c2ad294b55f5cac15a3b63f1552df1b3731e620`
101 - Causal status: `CONTENT_SPECIFIC`
102 - Semantic status: `CAUSALLY_VALIDATED`
103 - Selected quotient: `joint`, rank 3
104 - Audit rows: 40
105 - Covariance calibration rows: 8
106 - Explicit-CoT used during construction: `false`
107 - Final semantic support captured before quotient seal: `false`
108
109 ## Windows execution
110
111 ### 1. Validate the package without external weights
112
113 ```powershell
114 $ErrorActionPreference='Stop'; Expand-Archive .\SRIC_Block24_External_LRM_Causal_Semantic_Admission_v0.3.2.zip .\SRIC_Block24 -Force; Set-Location .\SRIC_Block24\SRIC_Block24_External_LRM_Causal_Semantic_Admission_v0.3.2; .\Run_Fake_Validation.ps1
115 ```
116
117 ### 2. Create the pinned external runtime
118
119 ```powershell
120 .\Setup_External_PrOntoQA_Coconut.ps1
121 ```
122
123 Useful fail-closed alternatives:
124
125 ```powershell
126 # Force the standard HTTP/LFS path instead of Xet
127 .\Setup_External_PrOntoQA_Coconut.ps1 -DisableXet
128
129 # Verify and stage assets already present in the local Hugging Face cache
130 .\Setup_External_PrOntoQA_Coconut.ps1 -LocalFilesOnly
131
132 # CPU-only setup for adapter debugging; not the recommended final audit path
133 .\Setup_External_PrOntoQA_Coconut.ps1 -CpuOnly
134 ```
135
136 The setup script creates an isolated Python 3.10-3.12 environment, checks out the pinned upstream commit, downloads the exact checkpoint plus only the required GPT-2 files, verifies the source blobs, hashes the complete local assets, and writes `external_runtime\runtime_paths.json`. If Xet transfer is unavailable, it retries through the standard HTTP/LFS path.
137
138 ### 3. Run a 12-task smoke audit
139
140 ```powershell
141 .\Run_Real_LRM_Pilot.ps1 -Profile smoke -Device cuda
142 ```
143
144 ### 4. Run the primary 64-task audit
145
146 ```powershell
147 .\Run_Real_LRM_Pilot.ps1 -Profile primary -Device cuda
148 ```
149
150 ### 5. Run the 256-task audit only after reviewing the primary receipt
151
152 ```powershell
153 .\Run_Real_LRM_Pilot.ps1 -Profile full -Device cuda
154 ```
155
156 The default scientific order is smoke -> primary -> full. Do not alter thresholds after viewing a final result. A failure is a valid terminal result.
157
158 ## Important boundary
159
160 This package validates infrastructure and an engineered positive control. It does not establish that the external Coconut checkpoint uses its latent states, represents proof-instance content, exposes a discoverable semantic quotient, or supports causal semantic editing. Those questions remain open until the source-locked checkpoint is run.
161
162 See:
163
164 - `BLOCK24_PROTOCOL.md`
165 - `BRAINSTORM_AND_DESIGN_DECISIONS.md`
166 - `CLAIM_BOUNDARY.md`
167 - `REAL_EXECUTION_STATUS.md`
168 - `REAL_RUN_CHECKLIST.md`
169 - `PROSQA_FOLLOWUP_PROTOCOL.md`
170
171 ## Terminal-summary fix in v0.3.2
172
173 The 12-task smoke profile intentionally disables covariance-matched noise. Its result therefore records `causal_calibration: null`. Earlier summarization code incorrectly assumed this optional field was always an object and could raise an `AttributeError` after the result itself had already been written and verified. Version 0.3.2 handles this terminal path explicitly. The scientific result JSON and all frozen causal gates are unchanged.