Skip to content
finalstate.aiWhen a Company Wiki Becomes a Graveyard, Fix the Clock
008Atlas

008THE GRAVEYARD

When a Company Wiki Becomes a Graveyard, Fix the Clock

A record of a machine rattle is only a starting point. Knowledge escrow proposes capture during work, preparation when needed and supervised rehearsal; access, sources and current context still govern what can be used.

A question about a machine rattle reaches a wiki page with a broken source connection.The illustrative machine and wiki page remain separated by a missing source trail. Asking a live question does not turn the old page into supported operating advice.WikiRattleSource??
Illustrative maintenance case, not a reported incident. The old page cannot establish which machine condition its advice concerns.

Preparation can come before anyone knows the question

A technician asked to document for an unknown successor must choose what context to preserve before the need is clear. That timing can waste effort. It is one reuse problem, not an explanation for every neglected wiki. Authors may reuse their own records, help peers or have immediate reasons to document. Permissions, maintenance and support also matter.

WHAT WON'T COME OUT

We can know more than we can tell.

Michael Polanyi, The Structure of Tacit-Knowing, Duke lecture, 1964, p. 12; excerpt

In the lecture, Polanyi describes recognizing faces and learning through advice, example and trial and error (pp. 11–15). Tacit knowing involves integrating particulars; listing them does not reproduce that skill. The rattle record may help someone learn without containing everything the technician knows.

CAPTURE IN THE WORK

Keep the incident attached to its evidence

A rattle incident record attaches to its job evidence within an access boundary.The incident tag gains a continuous source tether to the job record. A scope boundary encloses both, representing the context and permissions retained with the evidence. No complete skill extraction is shown.JobRattleScope
Proposed capture practice. Keep the original evidence and the limits of the account, including missing operating context.

In this example, a technician notices a rattle change under load. Preserve the account beside the job record, decision, correction and observed outcome. Record who supplied it, when, which operating conditions it covers and who may see it. This is not free capture, and it will not extract every tacit distinction. It gives a later reader something to inspect and question.

MOVE THE COST

Prepare for the question. Withhold what it cannot support.

A real question triggers a sourced extract while unsupported material remains behind the release boundary.The question arrives before any extract is prepared. Under the illustrated permission and currency conditions, a scoped extract crosses the boundary while its source tether remains intact. A separate unsupported claim from the same record stays on hold. A drawn boundary alone is not a successful check.RattleQuestionScopedHold
Conditional branches in the proposed workflow. The released extract assumes established provenance, access and current applicability; the held claim lacks that basis. No automatic clearance or measured cost saving.

A real question triggers preparation from sources the asker is allowed to use. Release only claims whose provenance, current applicability and access scope are established. Keep citations and visible redactions with the extract. Missing access, sources or freshness blocks the unsupported answer: the archive cannot support this answer yet. Knowledge escrow is this proposed timing arrangement, not a demonstrated reduction in total effort.

THE REUSE WORK

Someone still has to make the record usable

A schematic timeline places capture, preparation and use along one reuse process, with a return route.The rattle record persists along a timeline without a cost axis. A return route allows the producer to become a later reuser. The same person can occupy different roles; no separate staffing requirement, measured saving or automatic model competence is implied.CapturePrepareUseRoles can overlap
Schematic application to the maintenance record. Roles can overlap, and reuse work remains. This is not an empirical cost curve.M. Lynne Markus, JMIS 18(1), Summer 2001, pp. 57–93; DOI 10.1080/07421222.2001.11045671

M. Lynne Markus's 2001 theory of knowledge reuse distinguishes producers, intermediaries and consumers; one person may occupy all three roles (pp. 60–62). Reuse needs differ with the recipient and purpose. Retrieval, interpretation and restoring context require effort; incentives alone may not solve access and support problems (pp. 82–83, 87–88). A model may assist preparation, but this paper does not establish that an AI archive performs these jobs reliably.

THREE ASKERS

The same incident supports different permitted extracts

One rattle record connects to three differently scoped extracts, including a redacted adviser copy.Each asker's question triggers its extract. Source tethers stay attached as the copies take different forms. The adviser copy has an opaque redaction; none of the three is evidence of unrestricted access or a successful handoff.RattleStand-inAdviserNew hire
Illustrative permissions, not a claim that these roles are entitled to the whole archive. The adviser's permitted extract retains only a source reference that its scope allows.
  • Stand-in: operating evidence for the rattle question
  • Buyer's adviser: an incident trail within diligence access
  • New hire: material for supervised practice

Keep the underlying record intact. Each question gets a scoped extract with its source connection, omissions and currency limits. Redaction must not expose restricted material merely to prove provenance. If removing context leaves a claim unsupported, withhold it. An available record does not imply that every asker may receive it.

THE REHEARSAL

Record the missed cue before the correction

A rehearsal leaves a recorded gap before an expert amendment is attached.The rattle record is used in a controlled trial. The missed condition appears as a separate gap record; only afterward does an amendment join it. Both remain attached to the evidence, and a retest question stays open.Record?GapAmendmentRetest?
Illustrative rehearsal result. The observed miss is retained; the amendment does not turn it into a pass. No real machine procedure or performance measurement is supplied.

Use a controlled training rig and agree the expected stop or escalation condition before the trial. In this example, the stand-in misses the change in rattle under load. Record the miss and the missing context before the expert amends the account. Try again without expert prompting, with the archive available, and record what the stand-in does before offering help. A correction is not a demonstrated improvement, and one successful trial would cover only the tested condition.

THE STAND-IN TEST

Can the next person use it and recognize its limits?

The incident record enters a rehearsal and receives a documented correction before another trial.A forward route carries the sourced record to the rehearsal. The gap is retained, and a return route attaches the amendment to the record. The next trial remains open; capture never jumps directly to a competence verdict.RecordRehearseGap?Amend / retry
  • Choose one exception and preserve its evidence
  • Ask a permitted question and check the proposed answer
  • Rehearse, record the gap, amend and try again

Keep the rule, owner and scenario attached to the tested condition. Record what happened, what remains unknown and when the evidence needs review. A passed trial informs that handoff; it does not certify general competence or make later learning unnecessary.

Read the transcript

01 · THE GRAVEYARD

A technician hears a rattle and opens the company wiki. In this illustrative case, the page describes an earlier machine setup, with an old date and no trail back to the job record. The rattle question reaches the page, but its source connection is missing. The operator cannot tell whether the advice applies. Timing may be part of the problem: somebody prepared a page before knowing this question. A timing change still needs evidence, access and practical learning.

02 · THE CLOCK

Documenting for an unknown successor asks a technician to predict what that person will need and preserve the right context now. Some of that preparation may never be used. But the author can benefit too, by consulting the record later or exchanging useful accounts with peers. Neglected wikis have more than one cause. Moving preparation closer to a real question is a proposal to test, alongside the cost of capture, maintenance, retrieval and interpretation.

03 · WHAT WON'T COME OUT

Michael Polanyi stated the limit in his nineteen sixty-four Duke lecture: We can know more than we can tell. He discussed recognizing faces and learning through advice, example and trial and error. Knowing which particulars to integrate is different from possessing a list of them. The technician may hear a meaningful change without being able to explain every contributing cue. That does not make the skill unteachable, or make a recording of the rattle a substitute for learning it.

04 · CAPTURE IN THE WORK

Keep evidence close to the work. In this example, the rattle changed under load. Attach the incident account to the job record, decision, correction and observed outcome. Preserve who supplied it, when, its operating context and its access boundary. The source connection stays with the record. This is not free capture, and it will not extract every tacit distinction. It preserves material that a later reader can inspect, question and use in practice.

05 · MOVE THE COST

A real question arrives. Only then prepare an extract from permitted sources. For the supported part, establish provenance, access and current applicability. Release that scoped extract with its source connection and any redactions visible. Hold the unsupported claim behind the boundary. Missing access, sources or freshness means the archive cannot support this answer yet. Calling the arrangement knowledge escrow does not make preparation costless or establish that an AI system can perform it reliably.

06 · THE REUSE WORK

Lynne Markus's two thousand and one theory distinguishes a knowledge producer, an intermediary who prepares it and a consumer who uses it. These roles can overlap: the technician may later use the same record. Different recipients need different context and support. Retrieval and interpretation require effort alongside documentation. Incentives alone may not solve the access and support problems. Moving some preparation to demand changes when work happens; it does not erase that work.

07 · Advertisement · Twain GTM Agents

A good outbound message should know why it is being sent. Twain GTM agents research accounts, qualify contacts, and turn live signals into relevant outreach. They work with your CRM, personas, and GTM stack, so the message is not just personalized. It is backed by a specific reason to start the conversation.

08 · THREE ASKERS

The stand-in asks about the rattle and receives the permitted operating extract. The buyer's adviser asks about the incident history and receives a redacted extract within the diligence scope. The new hire asks what to practise and receives material for supervised learning. Each extract keeps an allowed source reference; the original record remains intact. A citation cannot justify revealing restricted material. If redaction removes the basis for a claim, withhold that claim too.

09 · THE REHEARSAL

Use a controlled training rig, with the stop or escalation condition agreed before the trial. In this example, the stand-in misses the change in rattle under load. Record the missed cue and the missing operating context. Then the expert amends the account while the original miss stays on record. Try again without expert prompting, with the archive still available. Record what the stand-in does before offering help. The correction has not yet demonstrated improvement. Even a successful trial would show performance on this condition, not general competence.

10 · THE STAND-IN TEST

Choose one exception, keep its evidence and ask a question within the reader's access. Check what the proposed answer can support before rehearsing it. Record the rehearsal gap before returning a correction to the incident record. Then try another trial and retain its outcome, including uncertainty or a stop. Keep the rule, owner and scenario with the tested condition. Page count and a fluent answer cannot tell you whether the next person is ready.

01 / 10 · THE GRAVEYARD0:00 / 5:23