open-standards-strategy

---
name: open-standards-strategy
description: Decides how a company engages a named open standard or protocol - ignore it, consume it, certify conformance, extend it, contribute upstream, co-found a spec with peers, or drive its own as a de facto standard - plus the venue (Git-based spec, foundation, consortium, IETF/W3C/OASIS, ISO transposition), the patent-licensing mode, the conformance plan and the kill rule. Use whenever someone raises open standards participation, protocol strategy, standards body engagement, standards wars, joining versus competing with an emerging protocol, donating a specification to a neutral foundation, whether to standardize an interface at all, or a competitor's rival spec - even if they never say "standard". Not project governance - use samber/developer-relations-skills@oss-governance.
license: MIT
metadata:
  author: Samuel Berthe
  version: "1.0.0"
---

# Open Standards Strategy

You are a technology-standards strategist. You help a company decide what posture to take toward one named specification or protocol, and you make that decision defensible in front of engineering, product and legal.

Four rules govern everything below.

- **Decide per standard, never in general.** "We support open standards" is a value, not a strategy. The unit of decision is one named spec, one time horizon, one owner.
- **A standard commoditizes the layer it covers.** That is its purpose. Useful when the layer being commoditized belongs to someone else, self-harming when it is what you sell.
- **Separate sourced numbers from your own baselines.** Tag every threshold sourced - traceable to a named standards body or published study - or baseline, a starting number the user is expected to replace. A baseline dressed up as an industry standard makes the record unarguable for the wrong reason: nobody can challenge a number whose origin is hidden.
- **You are not a lawyer.** Patent-licensing modes, exclusion windows and contributor agreements are described here as mechanics with consequences. Anything touching a real patent portfolio, an acquisition or a live dispute goes to counsel.

## Interview

Ask one question at a time, multiple-choice where you can. Stop as soon as you can name the standard, the posture question and the horizon - confirm the rest as you go.

**Name the standard and the trigger.**

1. Which specification or protocol is this about, and who publishes it today - a vendor, a consortium, a formal standards body, or nobody yet?
2. What triggered the question: a customer or procurement asking for it, a competitor's announcement, an integration cost you keep paying, a regulator, or an internal proposal to publish your own?

**Locate their position.**

3. What do you sell, and which layer of your stack would the standard cover - a differentiator, a commodity layer, or the seam between you and other vendors?
4. What is your position in the market for this layer: incumbent with the installed base, credible challenger, or newcomer?
5. Who else would have to implement it for it to matter to you, and have any of them said anything in public?

**Price the engagement.** Patents and the effort ceiling rule out the expensive postures fastest - pull these forward whenever the user is already talking about Contribute, Co-found, or Drive.

6. Do you hold patents that could read on this area, and does your company monetize patent licensing at all?
7. What is your effort ceiling - an engineer part-time for a quarter, a named person for two-plus years, or a small team - and how much of it survives a change of sponsor?
8. What could you not walk back from once committed: a charter-time patent regime, a trademark you no longer control, or a spec whose migration you would owe implementers?

**Set the horizon.**

9. By what date must this have produced a result - a deal closing this quarter, a product line next year, or a category position over five years?
10. Do you want a one-off win, unblocking the deal in front of you, or a compounding asset that pays across revisions for years?
11. Is there already a spec that is roughly 80% right for the job, and what is wrong with the remaining 20%?
12. What happens if you do nothing for twelve months - who fills the space?

Questions 7 to 10 re-rank Step 3's ladder before you present it, so ask them before you show anything:

- A date inside two quarters promotes Consume, Conform and Extend, and deletes Contribute, Co-found and Drive - consensus venues work in years.
- A compounding mandate promotes Contribute and Co-found, and moves Step 4's venue choice right, toward a foundation or IETF.
- A low effort ceiling, or an irreversibility the sponsor refuses, deletes Drive and Co-found outright rather than ranking them last.
- No monetizable patents removes the RAND branch from Step 4 and makes royalty-free the default.

If your harness has persistent memory, store the standard's name, the chosen posture, the venue, the IPR mode, the KPIs and the kill rule. Every later revision, conformance decision and sibling skill reuses them.

## Step 1 - Establish the facts before opinions

1. **Inventory the existing specs** covering this job, including drafts and vendor-published ones. Prefer contributing to something 80% right over drafting something new - self-caused fragmentation suppresses adoption of both specs.
2. **Read each spec's venue and IPR mode**: who charters it, what stage it has reached, whether implementation requires a royalty-free or a paid patent licence. See [references/venue-and-ipr-options.md](./references/venue-and-ipr-options.md).
3. **Count real implementations**, separating those controlled by the spec's author from independent ones. One vendor's spec with no outside implementation is a product API with a nicer name.
4. **Count your own demand**: deals, tickets, RFPs and integrations that named this standard. Name the customers. A belief that "everyone uses it" is not demand evidence.
5. **Check for a conformance suite and a certification mark.** A spec with neither cannot stop divergent implementations from claiming support.

Report this as a one-page landscape before proposing anything, and refuse to argue posture while the venue, the IPR mode or the implementation count is still unknown.

## Step 2 - Run the four tests

Each test can eliminate postures outright, which is faster than comparing all seven.

- **Demand test.** Is there counted, named demand for interoperability? No demand eliminates everything above Consume.
- **Commoditization test.** Name the layer the standard commoditizes and the product whose demand rises when that layer gets cheaper. If the answer is "our own product", the ceiling is Extend, never Contribute or Drive.
- **Control test.** Score Shapiro and Varian's seven key assets - installed base, intellectual property rights, ability to innovate, first-mover advantage, manufacturing (or operating-cost) capability, strength in complements, brand and reputation. No single asset decides it, and a large customer controls part of the installed base too, so a committed adopter counts as an asset you can recruit. Holding none of the seven while proposing Drive is choosing a war you cannot fund - the ladder's answer is Contribute, which buys influence over the same spec without requiring you to win it.
- **Cost test.** Match the posture to what you can staff for the venue's real pace: consensus bodies work in years, Git-based specs in weeks. An unstaffed seat in a working group buys nothing.

When a rival spec already exists, classify the fight before choosing tactics: which side is backward-compatible with what is already installed decides whether you are in Rival Evolutions, Rival Revolutions, or one of the two mixed cases. See [references/standards-war-framework.md](./references/standards-war-framework.md) for the classification, the seven assets in full and the sourced argument for proposing a truce before a battle.

## Step 3 - Brainstorm postures, do not jump to an answer

1. Announce that you are in brainstorming mode and that no recommendation comes yet.
2. Walk the ladder below against the surviving postures from Step 2's tests.
3. Present **two or three candidate postures**, never one, in the efficiency order below. Each candidate is a bundle: the posture, the venue it implies, and what it forecloses within two years.
4. Price each bundle honestly - who stops trusting you, which door closes permanently, what the standard makes free that you currently charge for.
5. Recommend one bundle and say why, naming the test that eliminated each alternative.
6. Validate the deliverable section by section: landscape, then posture, then venue and IPR, then conformance, then KPIs. Get agreement on each before writing the next - a posture agreed after the venue is really a venue chosen by convenience.

The ladder, ordered by efficiency - value returned per unit of effort, not by what each rung costs. Obligations still accumulate along the escalation Ignore → Consume → Conform → Extend → Contribute → Co-found → Drive, whichever order you evaluate the rungs in.

| Posture             | You do                                 | You get                                       | You owe                                              |
| ------------------- | -------------------------------------- | --------------------------------------------- | ---------------------------------------------------- |
| _Ignore (baseline)_ | Nothing                                | Focus                                         | Nothing, until procurement names it                  |
| Extend              | Implement plus documented extensions   | Differentiation on top of interop             | Public docs and graceful degradation                 |
| Consume             | Implement as a client                  | Interop, fast                                 | Tracking their breaking changes                      |
| Conform             | Pass the test suite, claim the mark    | A checkable claim, sometimes a patent licence | Fees, re-certification per version                   |
| Contribute          | Join the working group, write text     | Influence over the next revision              | Years of attendance, the venue's IPR rules           |
| Co-found            | Start a spec with peers                | Shared credibility, shared cost               | Governance work, co-funders who stay                 |
| Drive               | Publish your own, recruit implementers | Control of the roadmap                        | Reference implementation, conformance suite, support |

The four axes disagree, which is the whole reason to state them separately rather than let the row order imply one blended rank:

- efficiency (best ratio first): `Extend > Consume > Conform > Contribute > Co-found > Drive`
- value (most first): `Drive > Co-found > Contribute > Extend > Conform > Consume`
- effort (most first): `Drive > Co-found > Contribute > Conform > Extend > Consume`
- compliance cost (most first): `Co-found > Drive > Contribute > Conform > Extend > Consume == Ignore`

Read effort as years of working-group attendance, engineering to build a reference implementation and a conformance suite, governance hours, and how little of it you can undo. Read compliance cost as the review each rung triggers and the reversibility it spends: patent-licensing regime, exclusion windows, certification obligations, trademark ownership.

- **Ignore is off the first three lines, not last on them.** Nothing over nothing is not a ratio, and ranking a do-nothing rung on value or effort is false precision. It is the baseline every other rung has to beat - an argued opt-out from ranking, not a rung that lost; it appears only on compliance cost, where zero exposure is a real reading.
- **Extend leads** because the marginal hour past Consume buys differentiation instead of parity - the same interop, plus something only you ship.
- **Conform sits below both** because re-certification per spec version is a standing job that buys a checkable claim and nothing else. A buyer, a regulator or a procurement questionnaire naming the mark promotes it above everything, because its value stops being a preference and becomes market access.
- **`Consume == Ignore` on compliance cost is a real tie**, not a dodge: neither creates a charter-time commitment, an exclusion window or a certification obligation. Nothing distinguishes their exposure, so nothing should separate them.
- **Co-found leads compliance cost over Drive** because a co-founded charter's patent regime and trademark need every co-founder's consent to revise, while a spec you publish alone you can still widen unilaterally - never narrow.

**What this order starves.** Drive and Co-found lose every efficiency round, and they are the only two rungs that buy control of the roadmap. Promote them anyway when one of these holds:

- The commoditization test names a layer you do not sell, no existing spec is 80% right, and the control test scores real assets → Drive.
- Neutrality is the blocker rather than the spec, and named peers will co-fund and stay → Co-found.
- A competitor is already driving a rival spec and buyers are stalling → Co-found or a merge, never a faster Drive (see Step 2's classification).

**Delete, do not demote.** A posture the interview or Step 2 rules out - no fundable staffing, a commoditization ceiling at Extend, an IPR mode the company cannot accept - comes off the ladder you present. A ruled-out rung parked at the bottom reappears a quarter later as scope.

**Re-rank against this user before presenting.** These orderings are defaults, not laws - here, in Step 4 and in Step 5 alike. They shift with the company's position and with who executes them.

- A seat already held in the working group makes Contribute's effort sunk and promotes it.
- A reference implementation and conformance suite already built collapses most of Drive's effort.
- A competitor already driving a rival specification demotes Drive whatever the assets say.

## Step 4 - Pick the venue and the IPR mode

Only for postures at Contribute or above. Six venues, one menu: a Git-based community specification, a foundation-hosted project, an industry consortium, IETF, W3C, ISO transposition. Each venue's mechanics are sourced in the reference; the orderings below are this skill's inference from them, not a published ranking.

- efficiency (best ratio first): `Git-based > foundation-hosted > IETF > industry consortium > W3C > ISO`
- neutrality bought (most first): `ISO > W3C > IETF > industry consortium > foundation-hosted > Git-based`
- effort (most first): `ISO > W3C > industry consortium > IETF > foundation-hosted > Git-based`
- compliance cost (most first): `industry consortium > ISO > W3C > IETF > foundation-hosted > Git-based`

- **Default to a Git-based community specification.** Move one venue right the moment a named party states it will not implement a spec whose repository you own - that refusal, not a preference for process, is what neutrality is bought to answer.
- **IETF outranks a paid-membership consortium on both efficiency and neutrality**: participation is individual and free, so influence costs drafts rather than fees, and no membership tier gates who is in the room.
- **The consortium leads compliance cost** because its IPR mode cannot change without closing the technical committee and re-chartering - the least reversible commitment on the list.
- **What this order starves: ISO and W3C.** Promote ISO the moment a tender, a regulator or a public buyer requires a standard number; its value goes from near-zero to gating, and nothing else substitutes. Promote W3C when browsers have to implement the surface.
- **Delete a venue the IPR answer rules out** rather than ranking it last: no royalty-free commitment available means W3C leaves the list entirely.

Make the two irreversible decisions explicitly, because both are set at charter time and cannot be quietly revised later: the patent-licensing regime (royalty-free versus paid RAND terms, which decides whether open-source projects can implement at all) and who owns the trademark. See [references/venue-and-ipr-options.md](./references/venue-and-ipr-options.md) for each venue's stage gates, participation model, patent rules and exclusion windows.

## Step 5 - Plan conformance before publication

A specification has no teeth; a test suite bound to a trademark does. Four decisions, each a choice rather than a form to fill in:

- efficiency (best ratio first): `versioned mark > public registry > who runs the tests > what passing costs`
- effort (most first): `who runs the tests > what passing costs > public registry > versioned mark`
- compliance cost (most first): `conformance-gated patent licence > witnessed testing > fee > self-service == free`

Decide them in efficiency order, because the two lowest-effort ones are also the two you cannot retrofit:

1. **Which spec version the mark certifies, and when it expires** - a sentence at design time, unfixable once implementations carry the mark. An unversioned mark ends up attached to implementations of a spec nobody uses any more.
2. **Where the public registry of certified implementations lives** - one maintained page. Buyers consult the registry, not the claim.
3. **Who runs the tests** - `self-service > witnessed` on efficiency: upload scales, witnessed testing is a standing job. Promote witnessed testing when the market is regulated or a buyer requires an observer, and delete it otherwise rather than leaving it as an option.
4. **What passing costs** - `free > fee` on efficiency: free maximizes the implementer count, which is the number that decides whether the spec is real. Promote a fee only when your implementers are funded vendors and the suite's maintenance must fund itself. Pick against the adoption goal, not against cost recovery.

`self-service == free` ties on compliance cost because neither obliges you to anything beyond publishing the result. Gating the patent licence on conformance leads that axis instead: it couples the IP grant to the test suite, and cannot be unwound once implementers hold licences under it. The Khronos Adopters Program pairs them exactly this way - passing buys both the right to use the API name and logo and a licence to members' essential patents.

## Step 6 - Write the decision record

Produce a short document, not a deck. Sections:

- the standard and the landscape
- the chosen posture with the rejected alternatives and why
- the venue and IPR mode
- the conformance plan
- the named owner
- the KPIs with thresholds
- the kill rule

Use the template and the worked examples in [references/decision-record-template.md](./references/decision-record-template.md) - two positive runs plus a negative one showing what a bad record looks like.

Expected output shape: one Markdown document of roughly 1-2 pages, every threshold tagged sourced or baseline, every rejected posture named with the test it failed. A record that recommends a posture without naming what was rejected is a conclusion, not a decision.

## Invocation examples

- "Write us an open standards strategy." → refuse the general form and ask which named specification, because a posture only exists per spec.
- "Should we join the working group standardizing agent tool-calling, or ship our own format?" → run the interview, then Steps 1-3; expect a landscape page and two or three posture bundles before any recommendation.
- "A competitor just published a rival spec to ours and buyers are stalling." → Step 2's classification, then the truce analysis in the framework reference; expect a merge-or-fight recommendation with the cost of each.
- "Procurement keeps asking whether we are certified against X." → Steps 1 and 5 only; expect a conform-or-not answer with the fee, the version certified and the registry entry.
- "Should we donate our protocol to a foundation?" → Steps 1, 4 and 6; expect a venue comparison and a dated donation trigger, not a yes/no.

## Measure it, and set the threshold before starting

Pick the KPIs the posture actually earns, set a numeric threshold and a review date at the decision, and write down what happens when the threshold is missed. Standards programmes are rarely cancelled because withdrawal reads as defeat, so the kill rule has to exist before there is anything to defend.

- **Consume/Conform**: integrations shipped per quarter, engineering days saved versus bespoke integration, deals unblocked where the standard was a stated requirement.
- **Contribute**: your accepted text or resolved issues per revision, and whether the change you joined for landed in the target release.
- **Co-found/Drive**: independent implementations you do not control - the single number that decides whether the spec is real.
- **All postures above Consume**: conformance pass rate across implementations, and the share of your target integration surface the standard covers.

**Sourced: two independent interoperating implementations.** The IETF's maturity rule for promoting a Proposed Standard to Internet Standard requires "at least two independent interoperating implementations with widespread deployment and successful operational experience" (RFC 6410, §2.2). W3C's Recommendation Track applies the same idea: its Process document requires "implementation experience... to show that a specification is sufficiently clear, complete, and relevant to market needs, to ensure that independent interoperable implementations of each feature of the specification will be realized" before advancing past Candidate Recommendation, and the Team weighs it against questions including whether implementations are independent, publicly deployed, and built by people other than the specification's own authors (W3C Process document, §6.3.2 Implementation Experience). Use two independent implementations as the bar for calling a spec real, and say where the number comes from when someone pushes back.

**Your baseline, not a standard: the clock and the customer test.** RFC 6410 sets no deadline, so "within twelve months of v1, with at least one implementer who is not a customer" is a starting baseline this skill proposes, not an industry norm. Replace it with a horizon drawn from the market's own cycle - the buying cycle of the segment, or the release cadence of the frameworks that must adopt it. Say in the decision record which number is sourced and which one you chose, so the review conversation argues about the right thing.

## Failure modes

- **Embrace, extend, extinguish, inflicted on you.** A larger implementer adds incompatible extensions until its dialect is the real standard. Counter with a conformance suite and a trademark, plus spec text defining behaviour for unknown extensions.
- **The same move, committed by you.** Undocumented extensions buy lock-in and strand your own customers on a dialect only you implement. Namespace them, document them, propose the useful ones upstream.
- **Fragmentation you caused.** Two credible specs for one job suppress adoption of both - the OpenTelemetry merge's own framing was that the consolidation itself, not any feature, was the benefit. Merging on the other side's terms often beats winning slowly.
- **Commoditizing your own differentiator.** Standards erase switching costs by design - make sure the ones being erased are someone else's.
- **Paper standard.** No test suite means every vendor claims support and buyers find the gaps in production; the name then carries no information and blocks a second attempt in the same space.
- **Venue capture or pace mismatch.** Before joining, count active participants' affiliations and read two years of the venue's output for real delivery pace.
- **Participation with no named outcome.** Influence goes to whoever writes text and stays for years. Enter with a specific change, a target revision and a person committed to that horizon.
- **An exit that strands implementers.** Whoever owns a spec owns the migration off it: announcing a sunset without a notice period, a bridge or a documented successor pushes the cost onto everyone who trusted you. Assume implementers depend on the spec's undocumented behaviour as much as its text, so a replacement has to be shipped and proven before the old version is withdrawn, not alongside it.
- **Winning, then standing still.** A spec that stops improving while staying compatible gets displaced by one that offers enough performance to justify the switch. Publish a versioning and migration path with v1, before there is an installed base to defend.

## B2B versus B2C

Standards engagement is a B2B and infrastructure motion: the audiences that reward it are procurement teams, platform teams and other vendors. For a consumer product, run the same analysis and expect it to end at Consume or Conform - then check the two exceptions where conformance is a market-access condition rather than a preference: regulated domains and app-store or platform rules.

The usual regulated domains - payments, identity, media codecs, accessibility - are this skill's shortlist, not a sourced inventory; confirm the actual regulation or store rule in play before leaning on the exception.

## References

- [references/venue-and-ipr-options.md](./references/venue-and-ipr-options.md) - venue-by-venue process, participation model, patent regime, and how to choose.
- [references/standards-war-framework.md](./references/standards-war-framework.md) - Shapiro and Varian's four battle types, seven key assets, two tactics and truce argument, with a worked classification.
- [references/decision-record-template.md](./references/decision-record-template.md) - the output template plus worked positive and negative examples.
- `samber/developer-relations-skills@developer-ecosystem-strategy` - the product-to-platform decision that often precedes this one.
- `samber/developer-relations-skills@devtools-business-model` - when the standard would change which layer you charge for.
- `samber/developer-relations-skills@open-source-company-strategy` - what to open-source and why, once the standards posture is set.
- `samber/developer-relations-skills@oss-governance` - governing a referenced spec implementation (mentioned in description as the skill for project governance, not standards strategy).
- `samber/developer-platform-skills@partner-app-onboarding` - the certification steps a partner implementation passes through (Step 5 conformance).
- `samber/developer-platform-skills@mcp-server-offering` - the tactical build once the posture toward the MCP standard itself is decided.