00.04 Data Calibration & Source Ledger (Standards for Citing External Data)
Governance metadata | Owner: Knowledge Base Administrator | Last reviewed: 2026-10-22 | Next review: 2027-01-31
Core principle: the most valuable thing in a knowledge base is not the conclusion, but the chain of evidence behind it. Every external figure must either trace back to a source or be explicitly labeled as an "assumption."
1. Why a Source Ledger Is Needed
Numbers in the knowledge base fall into three categories. Mixing them is the number-one killer of credibility:
| Type | Definition | Acceptable wording | Prohibited wording |
|---|---|---|---|
| Fact | Verifiable official/public source | "According to XX's 2025 annual report, ..." | "The industry generally believes ..." |
| Assumption | A parameter set internally for modeling | "Assume a 5% return rate (internal estimate)" | "The return rate is about 5%" |
| Projection | Self-consistent figures in a sandbox case | "This case is a projection, not a real battle report" | Citing it directly as an industry benchmark |
In one sentence: facts must be clickable, assumptions must be stated, projections must be labeled.
2. Source Tiers (A–E)
| Tier | Source type | Examples | Usage guidance |
|---|---|---|---|
| A | Official regulation/regulator text | EUR-Lex, FCC, METI, CBP, tax authorities | May directly support compliance conclusions |
| B | Listed-company annual report/prospectus/official IR | 10-K, annual report | May cite, with fiscal year and page number |
| C | Third-party research firm | GfK, Circana, Statista, etc. | Cite report name, period, and scope |
| D | Media/industry reporting | Tech media, industry associations | Use as a lead only, never as the sole basis for a decision |
| E | Empirical assumption/internal projection | BOM, return rate, conversion rate | Must be labeled "assumption" and accompanied by a sensitivity range |
3. Six Fields Required for Every External Figure
| Field | Description | Example |
|---|---|---|
| Source | Institution/document name | XX Company 2025 Annual Report |
| Year/period | Period the data belongs to | FY2025 / 2025-Q4 |
| Scope | Numerator/denominator, statistical coverage | Global shipments, excluding accessories |
| Link | Clickable original | https://... |
| Tier | A–E | B |
| Recomputable | Whether it can be recomputed from public data | Yes/No/Partially |
The ledger template should live in a Feishu Base or in the appendix to 00.04. Every new hard external figure must add one row.
4. Writing Standards
- Strong conclusions must carry a source. When using "70%," "more than half," or "industry average," either provide a source or switch to qualitative wording.
- Ranges beat point estimates. When unverifiable, write "approximately X%–Y% (empirical range)," not "X%."
- Time anchors. Write "as of 2026-10" instead of "currently"; for regulations, write "subject to the official text in force."
- Make assumptions explicit. Put modeling parameters in an "Assumptions" section and show sensitivity (whether the conclusion flips at ±20%).
- Isolate projections. Every sandbox article must open with "This is a self-consistent projection, not a real battle report."
5. Review Mechanism
- Quarterly spot check: randomly sample 20 external figures each quarter; verify the source is still accessible and the scope unchanged.
- Invalidation triggers: regulatory version updates, platform fee changes, major FX moves, source link 404.
- Change trail: record date, editor, and reason for every ledger modification.
6. Execution Checklist
[ ] When creating or editing a document, list every external figure in it
[ ] Complete the six fields for each figure (source/year/scope/link/tier/recomputable)
[ ] Change unsourced figures to "assumption" or delete them
[ ] Check for "currently/now/industry average" wording without a time anchor
[ ] Confirm every sandbox article opens with a "projection" disclaimer
[ ] Run the quarterly 20-figure spot check and record results
[ ] Update or downgrade a source within 7 days of a dead link
7. Related Documents
- 00.02 Core Metrics Definition Dictionary
- 07.06 Single-SKU Full-Lifecycle P&L Model
- 08.01 Benchmark Review
- 08.02 Smart Telescope Projection
8. First-Edition Source Ledger (21 High-Risk Figures)
Status key: A official text; B annual report/prospectus; C research firm; D media/industry lead; E internal assumption/projection. "Pending verification" means no public clickable source exists; the figure may not be cited as fact.
| # | Doc | Figure/claim | Tier | Status | Source/link | Recommendation |
|---|---|---|---|---|---|---|
| 1 | 01.00 | More than 70% of overseas hardware fails | E | Pending verification | No public clickable source | Change to "empirical judgment" or delete the number |
| 2 | 06.00 | >60% of US/EU consumer-electronics returns are not hardware failures | D | Pending verification | No public clickable source | Add a source or make it qualitative |
| 3 | 08.09 | AI hardware return rate 18%–32% | D | Pending verification | No public clickable source | Label as "projection assumption" |
| 4 | 05.15 | 80%+ of African retail flow goes through traditional open markets | D | Pending verification | No public clickable source | Add a source or make it qualitative |
| 5 | 05.19 | YouTube drives >36% of conversions; cutting it causes a 40% drop | E | Model projection | Internal Shapley projection | Label "projection, not measured" |
| 6 | 05.16 | Low-price competitors have an 18% repair rate | E | Sales script | No public source | Change to "some competitors" and delete the number |
| 7 | 05.11 | Korea e-commerce penetration ~32%, mobile >75% | C | Pending verification | KOSIS https://kosis.kr | Reconcile with current statistics |
| 8 | 05.11 | Korea import tariff 0%–8% | A | Verify against current rules | Korea Customs https://www.customs.go.kr | Check line by line against HTS |
| 9 | 05.11 | CKFTA: most consumer electronics at 0% | A | Verify against current rules | Korea Customs FTA https://www.customs.go.kr | Apply rules of origin |
| 10 | 05.09 | Most Japanese consumer electronics at 0% duty | A | Verify against current rules | Japan Customs https://www.customs.go.jp | Check line by line against HTS |
| 11 | 05.15 | South Africa VAT 15%, tariff 0%–9% | A | ** Verified (2026-10-23)** | SARS https://www.sars.gov.za/types-of-tax/value-added-tax/ | VAT 15% confirmed; the planned rises to 15.5% (2025) and 16% (2026) were both reversed; tariffs still per schedule |
| 12 | 05.10 | Vietnam VAT 10% | A | Verify against current rules | Vietnam Tax https://www.gdt.gov.vn | Note the temporary 8% rate |
| 13 | 05.10 | Malaysia SST 10% | A | ** Verified (2026-10-23)** | MySST https://mysst.customs.gov.my | Wording corrected: from 2025-07-01 sales tax is 5% / specific rates / 10% default for unlisted goods, not a flat 10% |
| 14 | 05.10 | Indonesia VAT 11% | A | ** High risk, manual check** | DJP https://www.pajak.go.id (unreachable locally) | Indonesia's VAT regime changed from 2025 (incl. a DPP mechanism); 11% may be outdated — confirm via a local network before citing |
| 15 | 05.02.01 | Kickstarter conversion 20%–30%, CPL $2–5 | E | Pending verification (no public source) | None. The former link pointed at the platform fee page, which does not support conversion/CPL claims — removed | Label as an empirical assumption; replace with measured data |
| 16 | 05.02.01 | VIP conversion 4%–8%, CPV $15–35 | D | Pending verification | No public clickable source | Label as assumption |
| 17 | 05.07 | Defective Allowance 2.5%–4.0% | D | Contract term | Subject to the actual channel contract | Verify against the contract |
| 18 | 04.02 | CPM ≤ RMB 50–80, CPE ≤ RMB 3–5 | D | Platform rate card | Subject to the current rate card | Verify against the rate card |
| 19 | 05.05 | Platform commission approx. 8%–45% | A | Verify by category | Amazon Seller Central https://sellercentral.amazon.com | Use the category fee schedule |
| 20 | 07.06 | RMA 3%, marketing 14%, tooling amortization 15 | E | Internal assumption | Internal P&L model | Keep the "assumption" label |
| #15.1 | 05.01 / 05.02 | Kickstarter fee 5% + payment processing 3% + US$0.30 per pledge (5% + $0.08 under $10) | A | ** Verified (2026-10-23)** | Kickstarter Fees https://www.kickstarter.com/help/fees (visited in-browser) | Charged only if the campaign is successfully funded |
Usage rule: any figure marked "pending verification" must have an A/B/C-tier source before being cited in external materials (fundraising, customers, media). If no source can be found, rephrase it qualitatively.
8.1 Four Parameters Behind the Gross-Margin Redline Awaiting Calibration
07.06 §1.1 derives a gross-margin floor of 32.6% from "net-margin target 10% + expense load 22.6pt", against a current redline of 35%. All three coefficients in the expense load are internal assumptions (Tier E) and must be tracked explicitly, otherwise the redline will be mistaken for a derived value.
| # | Parameter | Current value (net-revenue basis) | Tier | Data source | Observation period | Upgrade target |
|---|---|---|---|---|---|---|
| 21 | Marketing rate (TACOS) | 15.2% | E | Amazon / Meta / Google ad consoles: actual spend ÷ net revenue | 3 consecutive months | → B (recomputable from own backend) |
| 22 | RMA rate | 3.3% | E | Returns system + 06.03 Reverse Grading: per-unit return loss ÷ net revenue | 3 consecutive months | → B |
| 23 | Tooling amortization rate | 4.1% (Q = 3,000) | E | Tooling cost USD 45,000 ÷ (project-declared Q × net revenue). Q is a variable and is not locked down | Recompute from the formula whenever Q changes | → B (contract verifiable; Q is a project-level declaration) |
| 24 | BOM as % of net revenue | 28.6% | E | 02.08 Competitor Teardown: reverse-engineered BOM total ÷ net revenue | Once per category, first teardown round | → C (measured + quoted) |
Three rules:
- The denominator is unified at net revenue. Using the MSRP basis would understate this by 1.8pt (marketing 14.0 / RMA 3.0 / tooling 3.8), and a redline derived from that would be too low, letting bad projects through.
- The 10% net-margin target is not in this table. It is policy, not data — it cannot be "calibrated", only changed by the board.
- The "redline" is a function of Q, not a constant.
floor(Q) = 28.5% + tooling ÷ (Q × net revenue); 35% holds only when Q ≥ roughly 1,880 units. Hence there is no company-wide amortization volume to lock down — each project declares its own Q and derives its own floor.
Current conclusion: at Q = 3,000 units this case shows a variable-cost break-even of 28.5% and a full-cost floor of 32.6%, leaving the 35% target a 2.4pt buffer; at Q = 1,800 the floor rises to 35.3% and 35% is no longer sufficient. Until all four parameters are calibrated, 35% may only be described as "target line plus a derivation constraint" — never as "calibrated".
8.2 Official-Source Verification Log (2026-10-23)
Every ledger entry with an official link (#7–#19) was visited. Results fall into three classes: confirmed, reachable but item-by-item, unreachable from here.
| # | Source | Reachable | Finding |
|---|---|---|---|
| 7 | KOSIS (Korea statistics) | ✅ 200 | Site reachable; the specific e-commerce penetration figure needs a table query and was not taken |
| 8 | Korea Customs | ✅ 200 | Site reachable; tariffs are per HTS line — the 0%~8% direction holds, confirm per part number |
| 9 | Korea Customs (FTA) | ✅ 200 | Same; CKFTA rules of origin must be confirmed line by line |
| 10 | Japan Customs | ✅ 200 | Site reachable; most ITA-covered electronics are 0%, confirm per HS code |
| 11 | SARS | ✅ 200 | VAT 15% confirmed; the planned rises to 15.5% (2025-05-01) and 16% (2026-04-01) were both reversed (page updated 2026-09) |
| 12 | Vietnam GDT (gdt.gov.vn) | ❌ timeout | Both local curl and the user's Chrome returned ERR_TIMED_OUT (same network) — a network-level block, not a dead link; the "10% standard + temporary 8% relief" direction is unchanged, but the applicable period needs confirmation |
| 13 | MySST (Malaysia) | ✅ 200 | Wording corrected: from 2025-07-01 sales tax is 5% / specific rates / 10% default for unlisted goods, not a flat 10% |
| 14 | DJP (Indonesia) | ❌ timeout | Both local curl and the user's Chrome returned ERR_TIMED_OUT; Indonesia's VAT regime also changed from 2025 (incl. a DPP mechanism), so 11% may be outdated — flagged high risk |
| 15 | Kickstarter | ✅ via browser | Fees verified: platform 5% + payment processing 3% + US$0.30 per pledge (5% + $0.08 under $10), charged only on success (see #15.1) But that page does not support "conversion 20%~30%, CPL $2~5" — the original link was a misattribution and is now an unsourced assumption |
| 16 | No public source | — | Stays "pending verification"; assumption only |
| 17 | Channel contract | — | Stays "check against contract" |
| 18 | Platform rate card | — | Stays "check against rate card" |
| 19 | Amazon Seller Central | ✅ 200 | Reachable but requires login; the 8%~45% commission direction holds, confirm against the category fee schedule |
Three actions arising:
- South Africa (#11) and Malaysia (#13) are closed — citable, but always with date and basis (South Africa "15%, 2025 rise reversed"; Malaysia "10% default, 5% listed separately").
- Indonesia (#14) is the only high-risk item: if the body cites 11%, add "to be confirmed locally"; ask an Indonesia-based colleague or agent to check the rate and the DPP mechanism.
- Kickstarter (#15) is closed via browser: fees confirmed and split out as #15.1; its conversion/CPL part has no public source and remains an assumption.
- Vietnam (#12) is still unreachable: needs a local network to confirm whether the 8% relief continues.
- CPL basis to be unified: the same metric appears as three ranges — 05.02 "USD 1.50~3.00", 05.01 "USD 1.5~3.5", 05.02.01 and #15 "USD 2~5". All are empirical assumptions (no public benchmark), differing by market, creative quality and audience breadth. Recommendation: treat "USD 1.5~5.0" as the outer envelope, narrow the lower bound by channel, and mark each document "assumption range, replace with measured data". Not a factual conflict, but always cite with range and context.
Official-portal access note: on 2026-10-23 both local curl and the user's Chrome timed out (same network egress) for these portals, which is a network-level block, not a dead link. Verify such entries via a local network, the official search portal, or public annual reports. tax and customs portals in the ledger (e.g. Vietnam gdt.gov.vn, Indonesia pajak.go.id) restrict access from some regions, so local checks may time out or return 000. Such cases are not treated as broken links; verify via a local network, the official search entry point, or public annual reports.
Review record (2026-10-22): This document is the source-ledger framework. Figures marked "pending verification" must have an A/B/C-tier source before external citation.
Review update (2026-10-22): Items #1, #2, #4, #5, #6, #15, and #16 have been downgraded in the source documents to "empirical judgment / projection assumption / qualitative statement" and are no longer cited as factual numbers. Items #7–#19 now include official links but still require verification against the rules in force.