---
id: s3-source-inventory-template
title: Source Inventory Template
module: GROW-S3
module_slug: grow-s3-data-provenance
cluster: Systems
type: template
version: v0.2.2
status: Draft
tier: membership
contract_role: Produces C4 → Evaluation
canonical_url: "https://grow.goodcombinator.ai/library/registry/s3-source-inventory-template"
download_url: "https://grow.goodcombinator.ai/library/registry/s3-source-inventory-template.md"
license: CC-BY-4.0 (proposed — owner confirmation required)
source: GROW by Good Combinator
retrieved_at: 2026-05-29
---

# Source Inventory Template

## Purpose
The Source Inventory is the canonical register of every input a GROW-built system is allowed to read, retrieve, embed, summarize, or cite. Nothing enters lineage that is not first listed here. If an agent reaches for data not on the inventory, that is a provenance violation and a reliability incident, not a data-quality problem to solve at output time.

A complete inventory answers four questions for every byte the system touches:
1. Where did it come from and who owns it?
2. Is it authoritative for the claim we will use it to support?
3. How fresh is it, and how often can it move under us?
4. What source confidence do we attach to outputs derived from it?

The `confidence_band` field on every row is the value supplied to `s2-scoring-system` per locked contract C4.

## Required Fields
| Field | Type | Required | Notes |
|---|---|---|---|
| `source_id` | string (kebab-case) | yes | Stable across versions; never renumbered. |
| `name` | string | yes | Human-readable. |
| `type` | enum | yes | `dataset` \| `document` \| `api` \| `connector` \| `upload` \| `generated` |
| `owner` | string | yes | Role or named individual accountable for the source. |
| `authoritative` | bool | yes | True only if this is the source-of-record for the claim type. |
| `update_frequency` | string | yes | E.g., `realtime`, `hourly`, `daily`, `weekly`, `event-driven`, `static`. |
| `last_validated` | date (ISO 8601) | yes | Last time owner confirmed the source is fit-for-purpose. |
| `permissions` | string | yes | Access scope; `public` \| `internal` \| `restricted` \| `regulated`. |
| `confidence_band` | enum | yes | `high` \| `medium` \| `low` \| `unknown` |
| `band_rationale` | string | yes | One sentence explaining why this source carries its assigned `confidence_band`. Per locked contract C4, this is a structured per-row field, not narrative prose. |

Optional but recommended: `retention_class`, `pii_flag`, `jurisdiction`, `cost_per_call`, `rate_limit`, `change_log_url`.

## Worked Example — Doug's Operating Stack
The following is a real-shape inventory for a multi-surface workload spanning a Florida special district seat, a 30A STR campus, and a podcast production pipeline.

| source_id | name | type | owner | authoritative | update_frequency | last_validated | permissions | confidence_band | band_rationale |
|---|---|---|---|---|---|---|---|---|---|
| `fl-dep-lpa0381` | FL DEP Grant LPA0381 disbursement ledger | dataset | District Treasurer | true | weekly | 2026-05-12 | regulated | high | Statutory system-of-record for grant disbursement; reconciled weekly by the treasurer against bank settlement. |
| `wcpa-parcels` | Walton County Property Appraiser parcel records | api | County GIS | true | daily | 2026-05-20 | public | high | County-published parcel system-of-record under FS Ch. 193; refreshed daily with documented change log. |
| `ownerrez-bookings` | OwnerRez bookings + payouts feed (Point Preserve) | connector | STR Ops | true | hourly | 2026-05-25 | internal | high | Booking system-of-record for confirmed reservations; high for booking events themselves, downgraded for revenue claims until reconciled with Airbnb. |
| `aifg-transcripts` | AI for Good podcast Otter/Descript transcripts | document | Producer | false | event-driven | 2026-05-22 | internal | medium | Machine-generated transcript with measurable named-entity and timestamp drift; safe for topic-level reuse but not verbatim quotation without review. |
| `slack-pp-archive` | Point Preserve internal Slack workspace export | dataset | Founder | false | weekly | 2026-04-30 | restricted | medium | Conversational and informal; not authoritative for any external claim, acceptable only as supporting context behind a primary source. |
| `ecoguardian-stream` | EcoGuardian sensor telemetry (water quality, turbidity) | api | EcoGuardian.AI | true | realtime | 2026-05-27 | internal | medium | Authoritative sensor feed but calibration window is open; high only post-calibration, medium during the current verification cycle. |
| `fasd-bulletins` | FL Association of Special Districts bulletin archive | document | Commissioner office | false | weekly | 2026-05-15 | public | medium | Reputable secondary commentary on statute and practice, but not a primary legal source; cite as context, not authority. |
| `airbnb-csv-dump` | Manual Airbnb reservation CSV exports | upload | STR Ops | false | monthly | 2026-04-01 | internal | low | Manual export with monthly latency and known overlap with `ownerrez-bookings`; usable only as a cross-check, never standalone. |

### Why these confidence bands
The `band_rationale` column above is the canonical, per-row justification consumed by `s2-scoring-system` under contract C4. The notes below restate the operating intuition for builders reading the inventory the first time; they do not replace the structured field.

- `fl-dep-lpa0381` and `wcpa-parcels` are statutory systems-of-record. **High.**
- `ownerrez-bookings` is the booking system-of-record, but reconciliation gaps with Airbnb mean medium-confidence for revenue claims — promoted to high only after reconciliation. Listed as high here for booking events themselves.
- `aifg-transcripts` are machine-generated with named-entity error; **medium**.
- `slack-pp-archive` is conversational, not authoritative for any external claim; **medium** at best, often **low** for factual reuse.
- `airbnb-csv-dump` is manual, late, and overlaps `ownerrez-bookings` for the same events; **low** when used standalone.

## Operating Rules
1. New source → inventory row before first read. No backfilling after the fact.
2. `last_validated` older than the `update_frequency` window flips `confidence_band` down one notch automatically until re-validated.
3. `authoritative=true` is exclusive per claim type. Two authoritative sources for the same claim is a governance bug, not a redundancy.
4. Deprecation is a row update (status note), never a deletion — provenance records reference historical `source_id`s.
5. Quarterly review: owners re-attest, stale rows are downgraded, and the inventory diff is logged into the lineage map.