Research-Stack/6-Documentation/docs/lean/PIST_RECEIPT_DENSITY_BACKFILL_V1.md
allaun 69988453bc chore(infra): stage working tree modifications
Build: 8604 jobs, 0 errors (lake build)
2026-06-22 01:23:17 -05:00

8 KiB

INFRA:DEAD rds -- AWS RDS is gone. Any file referencing rds_connect.py or this hostname is stale and must be ported.

PIST Receipt Density Backfill v1

Status: CALIBRATED_ENGINEERING_DELTA
Scope: RRC equation corpus routing evidence backfill
Script: 4-Infrastructure/shim/pist_receipt_density_injector.py


Purpose

This backfill converts existing RRC equation classification rows plus optional PIST spectral/classifier outputs into explicit receipt_density records.

It is a routing-evidence population pass, not a promotion pass.

receipt_density populated != theorem proved
receipt_density populated != claim promoted
receipt_density populated == route has structural/witness evidence for RRC use

Phase 2.1 verification sequence

Run the local regression harness first:

python3 4-Infrastructure/shim/test_pist_receipt_density_injector.py

Expected result: all tests print PASS.

Then emit the audit JSON only:

python3 4-Infrastructure/shim/pist_receipt_density_injector.py

Inspect:

shared-data/rrc_receipt_density_backfill.json

Only after inspection, run the guarded sidecar write:

python3 4-Infrastructure/shim/pist_receipt_density_injector.py --write-rds

Finally validate the sidecar table:

python3 4-Infrastructure/shim/validate_receipt_density_sidecar.py

Optional validation report:

python3 4-Infrastructure/shim/validate_receipt_density_sidecar.py \
  --out shared-data/rrc_receipt_density_sidecar_validation.json

The validator fails if:

sidecar table is empty
any row has promotion != not_promoted
any density/confidence is null or outside [0, 1]
any receipt hash/source is missing
any receipt_density_status is not CANDIDATE or HOLD

Phase 2.2 shape-alignment calibration

After classifier-backed density has been generated, run the alignment pass:

python3 4-Infrastructure/shim/rrc_pist_shape_alignment.py \
  --fail-on-raw-disagreement

Then inspect:

jq '.summary.shape_alignment_counts, .summary.warning_counts' \
  shared-data/rrc_receipt_density_backfill.json

Expected transition for the current corpus:

pist_shape_disagreement -> removed
structural_semantic_label_divergence -> present for compatible PIST/RRC label-space divergence
shape_alignment_counts.compatible_structural_projection -> most or all records

Reason:

PIST exact/proxy label = structural/spectral morphology
RRC shape label        = semantic/domain routing class

So a row like:

PIST exact label: LogogramProjection
RRC shape: CognitiveLoadField

is not necessarily an error. It can mean PIST sees a symbolic/logogram surface while RRC supplies the semantic routing class.

After alignment, rerun the guarded RDS sidecar write and readback validator:

python3 4-Infrastructure/shim/pist_receipt_density_injector.py --write-rds
python3 4-Infrastructure/shim/validate_receipt_density_sidecar.py

Commit the aligned JSON artifacts after validation.


Default inputs

6-Documentation/docs/rrc_equation_classification.md
shared-data/rrc_pist_exact_validation.json

The first file provides equation IDs, RRC shape hints, status, and declared axes.

The second file provides optional PIST spectral/classifier outputs such as:

proxy_pred
exact_pred
matrix_hash
canonical_hash
spectral_gap
rank_estimate
laplacian_zero_count

The injector explicitly filters Markdown table header/separator rows such as Equation and ---, which older validation passes could accidentally treat as equations.


Default output

shared-data/rrc_receipt_density_backfill.json

Optional JSONL output for later DB/RDS import:

python3 4-Infrastructure/shim/pist_receipt_density_injector.py \
  --jsonl-out shared-data/rrc_receipt_density_backfill.jsonl

Run command

python3 4-Infrastructure/shim/pist_receipt_density_injector.py

Strict mode, useful for CI:

python3 4-Infrastructure/shim/pist_receipt_density_injector.py --fail-on-missing-pist

Custom input/output:

python3 4-Infrastructure/shim/pist_receipt_density_injector.py \
  --rrc-file 6-Documentation/docs/rrc_equation_classification.md \
  --pist-report shared-data/rrc_pist_exact_validation.json \
  --out shared-data/rrc_receipt_density_backfill.json

Guarded RDS sidecar write:

python3 4-Infrastructure/shim/pist_receipt_density_injector.py \
  --write-rds \
  --rds-table ene.rrc_receipt_density

The writer uses:

4-Infrastructure/shim/rds_connect.py

so connection parameters resolve through the shared RDS path:

DATABASE_URL
RDS_* env vars
IAM token / boto3 / AWS CLI
password fallback

Record schema

Each record has the form:

{
  "receipt_version": "pist-receipt-density-v1",
  "equation_id": "bandwidth_adjusted_threshold",
  "rrc_shape": "CognitiveLoadField",
  "domain": "analysis",
  "source_status": "CANDIDATE",
  "receipt_density": 0.7125,
  "confidence": 0.6842,
  "density_components": {
    "status_score": 0.45,
    "axis_score": 1.0,
    "spectral_quality": 0.73,
    "shape_agreement": 0.82
  },
  "shape_prediction": {
    "ground_truth_hint": "CognitiveLoadField",
    "proxy_pred": "...",
    "exact_pred": "...",
    "matrix_hash": "...",
    "canonical_hash": "...",
    "spectral_gap": 0.42,
    "rank_estimate": 8,
    "laplacian_zero_count": 1
  },
  "shape_alignment": {
    "alignment_version": "rrc-pist-shape-alignment-v1",
    "alignment_status": "compatible_structural_projection",
    "alignment_confidence": 0.72,
    "label_space_model": "PIST=morphology; RRC=semantic routing"
  },
  "top_axes": ["projection_declared", "negative_control_strength"],
  "status": "CANDIDATE",
  "promotion": "not_promoted",
  "source": "pist_receipt_density_injector_v1",
  "receipt_hash": "...",
  "warnings": ["structural_semantic_label_divergence"]
}

RDS sidecar schema

The guarded writer creates or upserts into:

ene.rrc_receipt_density

Fields:

equation_id
rrc_shape
domain
source_status
receipt_density
receipt_density_source
receipt_density_hash
receipt_density_status
receipt_density_warnings
confidence
top_axes
shape_prediction
density_components
promotion
payload
updated_at

The table has a check constraint:

promotion = 'not_promoted'

This makes promotion drift fail at the database boundary.


Density calculation

The density score is computed from four bounded components:

status_score
axis_score
spectral_quality
shape_agreement

Current weighting:

receipt_density =
  0.26 * status_score
+ 0.24 * axis_score
+ 0.26 * spectral_quality
+ 0.24 * shape_agreement

Confidence is slightly more classifier-weighted:

confidence =
  0.20 * status_score
+ 0.20 * axis_score
+ 0.28 * spectral_quality
+ 0.32 * shape_agreement

This is intentionally conservative. A missing PIST prediction can still produce a low-density record from declared RRC axes, but the record receives a missing_pist_prediction warning.


Claim boundary

The script writes every record as:

"promotion": "not_promoted"

This is the central anti-drift boundary.

The generated density says:

This equation has routing evidence.

It does not say:

This equation is true.
This equation is proved.
This equation is promoted to REVIEWED.

Promotion still requires external receipts, Lean/kernel verification where applicable, or human/adversarial review depending on the claim class.


CI candidate

A minimal CI step should run:

python3 4-Infrastructure/shim/test_pist_receipt_density_injector.py
python3 4-Infrastructure/shim/pist_receipt_density_injector.py --fail-on-missing-pist
python3 4-Infrastructure/shim/rrc_pist_shape_alignment.py --fail-on-raw-disagreement

The sidecar validator should run only in environments with RDS credentials.


Keeper phrase

PIST stops being just a repair engine when its classifications become receipt density for the RRC corpus.
PIST is seeing morphology; RRC is naming semantics.